dect
/
asterisk
Archived
13
0
Fork 0
This repository has been archived on 2022-02-17. You can view files and clone it, but cannot push or open issues or pull requests.
asterisk/main/event.c

1772 lines
48 KiB
C
Raw Permalink Normal View History

/*
* Asterisk -- An open source telephony toolkit.
*
* Copyright (C) 2007 - 2008, Digium, Inc.
*
* Russell Bryant <russell@digium.com>
*
* See http://www.asterisk.org for more information about
* the Asterisk project. Please do not directly contact
* any of the maintainers of this project for assistance;
* the project provides a web site, mailing lists and IRC
* channels for your use.
*
* This program is free software, distributed under the terms of
* the GNU General Public License Version 2. See the LICENSE file
* at the top of the source tree.
*/
/*! \file
*
* \brief Internal generic event system
*
* \author Russell Bryant <russell@digium.com>
*/
#include "asterisk.h"
ASTERISK_FILE_VERSION(__FILE__, "$Revision$")
#include "asterisk/_private.h"
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
#include "asterisk/event.h"
#include "asterisk/linkedlists.h"
#include "asterisk/dlinkedlists.h"
#include "asterisk/lock.h"
#include "asterisk/utils.h"
#include "asterisk/unaligned.h"
#include "asterisk/utils.h"
#include "asterisk/taskprocessor.h"
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
#include "asterisk/astobj2.h"
Merged revisions 295711 via svnmerge from https://origsvn.digium.com/svn/asterisk/branches/1.8 ................ r295711 | russell | 2010-11-19 18:50:00 -0600 (Fri, 19 Nov 2010) | 36 lines Merged revisions 295710 via svnmerge from https://origsvn.digium.com/svn/asterisk/branches/1.6.2 ........ r295710 | russell | 2010-11-19 18:45:51 -0600 (Fri, 19 Nov 2010) | 29 lines Fix cache of device state changes for multiple servers. This patch addresses a regression where device states across multiple servers were not being processing completely correctly. The code works to determine the overall state by looking at the last known state of a device on each server. However, there was a regression due to some invasive rewrites of how the cache works that led to the cache only storing the last device state change for a device, regardless of which server it was on. The code is set up to cache device state change events by ensuring that each event in the cache has a unique device name + entity ID (server ID). The code that was responsible for comparing raw information elements (which EID is) always returned a match due to a memcmp() with a length of 0. There isn't much code to fix the actual bug. This patch also introduces a new CLI command that was very useful for debugging this problem. The command allows you to dump the contents of the event cache. (closes issue #18284) Reported by: klaus3000 Patches: issue18284.rev1.txt uploaded by russell (license 2) Tested by: russell, klaus3000 (closes issue #18280) Reported by: klaus3000 Review: https://reviewboard.asterisk.org/r/1012/ ........ ................ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@295712 f38db490-d61c-443f-a65b-d21fe96a405b
2010-11-20 00:52:47 +00:00
#include "asterisk/cli.h"
static struct ast_taskprocessor *event_dispatcher;
/*!
* \brief An event information element
*
* \note The format of this structure is important. Since these events may
* be sent directly over a network, changing this structure will break
* compatibility with older versions. However, at this point, this code
* has not made it into a release, so it is still fair game for change.
*/
struct ast_event_ie {
enum ast_event_ie_type ie_type:16;
/*! Total length of the IE payload */
uint16_t ie_payload_len;
unsigned char ie_payload[0];
} __attribute__((packed));
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
/*!
* \brief The payload for a string information element
*/
struct ast_event_ie_str_payload {
/*! \brief A hash calculated with ast_str_hash(), to speed up comparisons */
uint32_t hash;
/*! \brief The actual string, null terminated */
char str[1];
} __attribute__((packed));
/*!
* \brief An event
*
* An ast_event consists of an event header (this structure), and zero or
* more information elements defined by ast_event_ie.
*
* \note The format of this structure is important. Since these events may
* be sent directly over a network, changing this structure will break
* compatibility with older versions. However, at this point, this code
* has not made it into a release, so it is still fair game for change.
*/
struct ast_event {
/*! Event type */
enum ast_event_type type:16;
/*! Total length of the event */
uint16_t event_len:16;
/*! The data payload of the event, made up of information elements */
unsigned char payload[0];
} __attribute__((packed));
/*!
* \brief A holder for an event
*
* \details This struct used to have more of a purpose than it does now.
* It is used to hold events in the event cache. It can be completely removed
* if one of these two things is done:
* - ast_event gets changed such that it never has to be realloc()d
* - astobj2 is updated so that you can realloc() an astobj2 object
*/
struct ast_event_ref {
struct ast_event *event;
};
struct ast_event_ie_val {
AST_LIST_ENTRY(ast_event_ie_val) entry;
enum ast_event_ie_type ie_type;
enum ast_event_ie_pltype ie_pltype;
union {
uint32_t uint;
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
struct {
uint32_t hash;
const char *str;
};
void *raw;
} payload;
size_t raw_datalen;
};
/*! \brief Event subscription */
struct ast_event_sub {
enum ast_event_type type;
ast_event_cb_t cb;
char description[64];
void *userdata;
uint32_t uniqueid;
AST_LIST_HEAD_NOLOCK(, ast_event_ie_val) ie_vals;
AST_RWDLLIST_ENTRY(ast_event_sub) entry;
};
static uint32_t sub_uniqueid;
/*! \brief Event subscriptions
* The event subscribers are indexed by which event they are subscribed to */
static AST_RWDLLIST_HEAD(ast_event_sub_list, ast_event_sub) ast_event_subs[AST_EVENT_TOTAL];
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
static int ast_event_cmp(void *obj, void *arg, int flags);
static int ast_event_hash_mwi(const void *obj, const int flags);
static int ast_event_hash_devstate(const void *obj, const int flags);
static int ast_event_hash_devstate_change(const void *obj, const int flags);
#ifdef LOW_MEMORY
#define NUM_CACHE_BUCKETS 17
#else
#define NUM_CACHE_BUCKETS 563
#endif
#define MAX_CACHE_ARGS 8
/*!
* \brief Event types that are kept in the cache.
*/
static struct {
/*!
* \brief Container of cached events
*
* \details This gets allocated in ast_event_init() when Asterisk starts
* for the event types declared as using the cache.
*/
struct ao2_container *container;
/*! \brief Event type specific hash function */
ao2_hash_fn *hash_fn;
/*!
* \brief Information Elements used for caching
*
* \details This array is the set of information elements that will be unique
* among all events in the cache for this event type. When a new event gets
* cached, a previous event with the same values for these information elements
* will be replaced.
*/
enum ast_event_ie_type cache_args[MAX_CACHE_ARGS];
} ast_event_cache[AST_EVENT_TOTAL] = {
[AST_EVENT_MWI] = {
.hash_fn = ast_event_hash_mwi,
.cache_args = { AST_EVENT_IE_MAILBOX, AST_EVENT_IE_CONTEXT },
},
[AST_EVENT_DEVICE_STATE] = {
.hash_fn = ast_event_hash_devstate,
.cache_args = { AST_EVENT_IE_DEVICE, },
},
[AST_EVENT_DEVICE_STATE_CHANGE] = {
.hash_fn = ast_event_hash_devstate_change,
.cache_args = { AST_EVENT_IE_DEVICE, AST_EVENT_IE_EID, },
},
};
Merged revisions 295711 via svnmerge from https://origsvn.digium.com/svn/asterisk/branches/1.8 ................ r295711 | russell | 2010-11-19 18:50:00 -0600 (Fri, 19 Nov 2010) | 36 lines Merged revisions 295710 via svnmerge from https://origsvn.digium.com/svn/asterisk/branches/1.6.2 ........ r295710 | russell | 2010-11-19 18:45:51 -0600 (Fri, 19 Nov 2010) | 29 lines Fix cache of device state changes for multiple servers. This patch addresses a regression where device states across multiple servers were not being processing completely correctly. The code works to determine the overall state by looking at the last known state of a device on each server. However, there was a regression due to some invasive rewrites of how the cache works that led to the cache only storing the last device state change for a device, regardless of which server it was on. The code is set up to cache device state change events by ensuring that each event in the cache has a unique device name + entity ID (server ID). The code that was responsible for comparing raw information elements (which EID is) always returned a match due to a memcmp() with a length of 0. There isn't much code to fix the actual bug. This patch also introduces a new CLI command that was very useful for debugging this problem. The command allows you to dump the contents of the event cache. (closes issue #18284) Reported by: klaus3000 Patches: issue18284.rev1.txt uploaded by russell (license 2) Tested by: russell, klaus3000 (closes issue #18280) Reported by: klaus3000 Review: https://reviewboard.asterisk.org/r/1012/ ........ ................ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@295712 f38db490-d61c-443f-a65b-d21fe96a405b
2010-11-20 00:52:47 +00:00
/*!
* \brief Names of cached event types, for CLI tab completion
*
* \note These names must match what is in the event_names array.
*/
static const char * const cached_event_types[] = { "MWI", "DeviceState", "DeviceStateChange", NULL };
/*!
* \brief Event Names
*/
static const char * const event_names[AST_EVENT_TOTAL] = {
[AST_EVENT_ALL] = "All",
[AST_EVENT_CUSTOM] = "Custom",
[AST_EVENT_MWI] = "MWI",
[AST_EVENT_SUB] = "Subscription",
[AST_EVENT_UNSUB] = "Unsubscription",
[AST_EVENT_DEVICE_STATE] = "DeviceState",
[AST_EVENT_DEVICE_STATE_CHANGE] = "DeviceStateChange",
[AST_EVENT_CEL] = "CEL",
[AST_EVENT_SECURITY] = "Security",
Merged revisions 295711 via svnmerge from https://origsvn.digium.com/svn/asterisk/branches/1.8 ................ r295711 | russell | 2010-11-19 18:50:00 -0600 (Fri, 19 Nov 2010) | 36 lines Merged revisions 295710 via svnmerge from https://origsvn.digium.com/svn/asterisk/branches/1.6.2 ........ r295710 | russell | 2010-11-19 18:45:51 -0600 (Fri, 19 Nov 2010) | 29 lines Fix cache of device state changes for multiple servers. This patch addresses a regression where device states across multiple servers were not being processing completely correctly. The code works to determine the overall state by looking at the last known state of a device on each server. However, there was a regression due to some invasive rewrites of how the cache works that led to the cache only storing the last device state change for a device, regardless of which server it was on. The code is set up to cache device state change events by ensuring that each event in the cache has a unique device name + entity ID (server ID). The code that was responsible for comparing raw information elements (which EID is) always returned a match due to a memcmp() with a length of 0. There isn't much code to fix the actual bug. This patch also introduces a new CLI command that was very useful for debugging this problem. The command allows you to dump the contents of the event cache. (closes issue #18284) Reported by: klaus3000 Patches: issue18284.rev1.txt uploaded by russell (license 2) Tested by: russell, klaus3000 (closes issue #18280) Reported by: klaus3000 Review: https://reviewboard.asterisk.org/r/1012/ ........ ................ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@295712 f38db490-d61c-443f-a65b-d21fe96a405b
2010-11-20 00:52:47 +00:00
[AST_EVENT_NETWORK_CHANGE] = "NetworkChange",
};
/*!
* \brief IE payload types and names
*/
static const struct ie_map {
enum ast_event_ie_pltype ie_pltype;
const char *name;
} ie_maps[AST_EVENT_IE_TOTAL] = {
[AST_EVENT_IE_NEWMSGS] = { AST_EVENT_IE_PLTYPE_UINT, "NewMessages" },
[AST_EVENT_IE_OLDMSGS] = { AST_EVENT_IE_PLTYPE_UINT, "OldMessages" },
[AST_EVENT_IE_MAILBOX] = { AST_EVENT_IE_PLTYPE_STR, "Mailbox" },
[AST_EVENT_IE_UNIQUEID] = { AST_EVENT_IE_PLTYPE_UINT, "UniqueID" },
[AST_EVENT_IE_EVENTTYPE] = { AST_EVENT_IE_PLTYPE_UINT, "EventType" },
[AST_EVENT_IE_EXISTS] = { AST_EVENT_IE_PLTYPE_UINT, "Exists" },
[AST_EVENT_IE_DEVICE] = { AST_EVENT_IE_PLTYPE_STR, "Device" },
[AST_EVENT_IE_STATE] = { AST_EVENT_IE_PLTYPE_UINT, "State" },
[AST_EVENT_IE_CONTEXT] = { AST_EVENT_IE_PLTYPE_STR, "Context" },
[AST_EVENT_IE_EID] = { AST_EVENT_IE_PLTYPE_RAW, "EntityID" },
[AST_EVENT_IE_CEL_EVENT_TYPE] = { AST_EVENT_IE_PLTYPE_UINT, "CELEventType" },
[AST_EVENT_IE_CEL_EVENT_TIME] = { AST_EVENT_IE_PLTYPE_UINT, "CELEventTime" },
[AST_EVENT_IE_CEL_EVENT_TIME_USEC] = { AST_EVENT_IE_PLTYPE_UINT, "CELEventTimeUSec" },
[AST_EVENT_IE_CEL_USEREVENT_NAME] = { AST_EVENT_IE_PLTYPE_UINT, "CELUserEventName" },
[AST_EVENT_IE_CEL_CIDNAME] = { AST_EVENT_IE_PLTYPE_STR, "CELCIDName" },
[AST_EVENT_IE_CEL_CIDNUM] = { AST_EVENT_IE_PLTYPE_STR, "CELCIDNum" },
[AST_EVENT_IE_CEL_EXTEN] = { AST_EVENT_IE_PLTYPE_STR, "CELExten" },
[AST_EVENT_IE_CEL_CONTEXT] = { AST_EVENT_IE_PLTYPE_STR, "CELContext" },
[AST_EVENT_IE_CEL_CHANNAME] = { AST_EVENT_IE_PLTYPE_STR, "CELChanName" },
[AST_EVENT_IE_CEL_APPNAME] = { AST_EVENT_IE_PLTYPE_STR, "CELAppName" },
[AST_EVENT_IE_CEL_APPDATA] = { AST_EVENT_IE_PLTYPE_STR, "CELAppData" },
[AST_EVENT_IE_CEL_AMAFLAGS] = { AST_EVENT_IE_PLTYPE_STR, "CELAMAFlags" },
[AST_EVENT_IE_CEL_ACCTCODE] = { AST_EVENT_IE_PLTYPE_UINT, "CELAcctCode" },
[AST_EVENT_IE_CEL_UNIQUEID] = { AST_EVENT_IE_PLTYPE_STR, "CELUniqueID" },
[AST_EVENT_IE_CEL_USERFIELD] = { AST_EVENT_IE_PLTYPE_STR, "CELUserField" },
[AST_EVENT_IE_CEL_CIDANI] = { AST_EVENT_IE_PLTYPE_STR, "CELCIDani" },
[AST_EVENT_IE_CEL_CIDRDNIS] = { AST_EVENT_IE_PLTYPE_STR, "CELCIDrdnis" },
[AST_EVENT_IE_CEL_CIDDNID] = { AST_EVENT_IE_PLTYPE_STR, "CELCIDdnid" },
[AST_EVENT_IE_CEL_PEER] = { AST_EVENT_IE_PLTYPE_STR, "CELPeer" },
[AST_EVENT_IE_CEL_LINKEDID] = { AST_EVENT_IE_PLTYPE_STR, "CELLinkedID" },
[AST_EVENT_IE_CEL_PEERACCT] = { AST_EVENT_IE_PLTYPE_STR, "CELPeerAcct" },
[AST_EVENT_IE_CEL_EXTRA] = { AST_EVENT_IE_PLTYPE_STR, "CELExtra" },
[AST_EVENT_IE_SECURITY_EVENT] = { AST_EVENT_IE_PLTYPE_STR, "SecurityEvent" },
[AST_EVENT_IE_EVENT_VERSION] = { AST_EVENT_IE_PLTYPE_UINT, "EventVersion" },
[AST_EVENT_IE_SERVICE] = { AST_EVENT_IE_PLTYPE_STR, "Service" },
[AST_EVENT_IE_MODULE] = { AST_EVENT_IE_PLTYPE_STR, "Module" },
[AST_EVENT_IE_ACCOUNT_ID] = { AST_EVENT_IE_PLTYPE_STR, "AccountID" },
[AST_EVENT_IE_SESSION_ID] = { AST_EVENT_IE_PLTYPE_STR, "SessionID" },
[AST_EVENT_IE_SESSION_TV] = { AST_EVENT_IE_PLTYPE_STR, "SessionTV" },
[AST_EVENT_IE_ACL_NAME] = { AST_EVENT_IE_PLTYPE_STR, "ACLName" },
[AST_EVENT_IE_LOCAL_ADDR] = { AST_EVENT_IE_PLTYPE_STR, "LocalAddress" },
[AST_EVENT_IE_REMOTE_ADDR] = { AST_EVENT_IE_PLTYPE_STR, "RemoteAddress" },
[AST_EVENT_IE_EVENT_TV] = { AST_EVENT_IE_PLTYPE_STR, "EventTV" },
[AST_EVENT_IE_REQUEST_TYPE] = { AST_EVENT_IE_PLTYPE_STR, "RequestType" },
[AST_EVENT_IE_REQUEST_PARAMS] = { AST_EVENT_IE_PLTYPE_STR, "RequestParams" },
[AST_EVENT_IE_AUTH_METHOD] = { AST_EVENT_IE_PLTYPE_STR, "AuthMethod" },
[AST_EVENT_IE_SEVERITY] = { AST_EVENT_IE_PLTYPE_STR, "Severity" },
[AST_EVENT_IE_EXPECTED_ADDR] = { AST_EVENT_IE_PLTYPE_STR, "ExpectedAddress" },
[AST_EVENT_IE_CHALLENGE] = { AST_EVENT_IE_PLTYPE_STR, "Challenge" },
[AST_EVENT_IE_RESPONSE] = { AST_EVENT_IE_PLTYPE_STR, "Response" },
[AST_EVENT_IE_EXPECTED_RESPONSE] = { AST_EVENT_IE_PLTYPE_STR, "ExpectedResponse" },
};
const char *ast_event_get_type_name(const struct ast_event *event)
{
enum ast_event_type type;
type = ast_event_get_type(event);
if (type < 0 || type >= ARRAY_LEN(event_names)) {
ast_log(LOG_ERROR, "Invalid event type - '%d'\n", type);
return "";
}
return event_names[type];
}
int ast_event_str_to_event_type(const char *str, enum ast_event_type *event_type)
{
int i;
for (i = 0; i < ARRAY_LEN(event_names); i++) {
Merged revisions 295711 via svnmerge from https://origsvn.digium.com/svn/asterisk/branches/1.8 ................ r295711 | russell | 2010-11-19 18:50:00 -0600 (Fri, 19 Nov 2010) | 36 lines Merged revisions 295710 via svnmerge from https://origsvn.digium.com/svn/asterisk/branches/1.6.2 ........ r295710 | russell | 2010-11-19 18:45:51 -0600 (Fri, 19 Nov 2010) | 29 lines Fix cache of device state changes for multiple servers. This patch addresses a regression where device states across multiple servers were not being processing completely correctly. The code works to determine the overall state by looking at the last known state of a device on each server. However, there was a regression due to some invasive rewrites of how the cache works that led to the cache only storing the last device state change for a device, regardless of which server it was on. The code is set up to cache device state change events by ensuring that each event in the cache has a unique device name + entity ID (server ID). The code that was responsible for comparing raw information elements (which EID is) always returned a match due to a memcmp() with a length of 0. There isn't much code to fix the actual bug. This patch also introduces a new CLI command that was very useful for debugging this problem. The command allows you to dump the contents of the event cache. (closes issue #18284) Reported by: klaus3000 Patches: issue18284.rev1.txt uploaded by russell (license 2) Tested by: russell, klaus3000 (closes issue #18280) Reported by: klaus3000 Review: https://reviewboard.asterisk.org/r/1012/ ........ ................ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@295712 f38db490-d61c-443f-a65b-d21fe96a405b
2010-11-20 00:52:47 +00:00
if (ast_strlen_zero(event_names[i]) || strcasecmp(event_names[i], str)) {
continue;
}
*event_type = i;
return 0;
}
return -1;
}
const char *ast_event_get_ie_type_name(enum ast_event_ie_type ie_type)
{
if (ie_type <= 0 || ie_type >= ARRAY_LEN(ie_maps)) {
ast_log(LOG_ERROR, "Invalid IE type - '%d'\n", ie_type);
return "";
}
return ie_maps[ie_type].name;
}
enum ast_event_ie_pltype ast_event_get_ie_pltype(enum ast_event_ie_type ie_type)
{
if (ie_type <= 0 || ie_type >= ARRAY_LEN(ie_maps)) {
ast_log(LOG_ERROR, "Invalid IE type - '%d'\n", ie_type);
return AST_EVENT_IE_PLTYPE_UNKNOWN;
}
return ie_maps[ie_type].ie_pltype;
}
int ast_event_str_to_ie_type(const char *str, enum ast_event_ie_type *ie_type)
{
int i;
for (i = 0; i < ARRAY_LEN(ie_maps); i++) {
if (strcasecmp(ie_maps[i].name, str)) {
continue;
}
*ie_type = i;
return 0;
}
return -1;
}
size_t ast_event_get_size(const struct ast_event *event)
{
size_t res;
res = ntohs(event->event_len);
return res;
}
static void ast_event_ie_val_destroy(struct ast_event_ie_val *ie_val)
{
switch (ie_val->ie_pltype) {
case AST_EVENT_IE_PLTYPE_STR:
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
ast_free((char *) ie_val->payload.str);
break;
case AST_EVENT_IE_PLTYPE_RAW:
ast_free(ie_val->payload.raw);
break;
case AST_EVENT_IE_PLTYPE_UINT:
case AST_EVENT_IE_PLTYPE_BITFLAGS:
case AST_EVENT_IE_PLTYPE_EXISTS:
case AST_EVENT_IE_PLTYPE_UNKNOWN:
break;
}
ast_free(ie_val);
}
/*! \brief Subscription event check list. */
struct ast_ev_check_list {
AST_LIST_HEAD_NOLOCK(, ast_event_ie_val) ie_vals;
};
/*!
* \internal
* \brief Check if a subscription ie_val matches an event.
*
* \param sub_ie_val Subscripton IE value to check
* \param check_ie_vals event list to check against
*
* \retval 0 not matched
* \retval non-zero matched
*/
static int match_sub_ie_val_to_event(const struct ast_event_ie_val *sub_ie_val, const struct ast_ev_check_list *check_ie_vals)
{
const struct ast_event_ie_val *event_ie_val;
int res = 0;
AST_LIST_TRAVERSE(&check_ie_vals->ie_vals, event_ie_val, entry) {
if (sub_ie_val->ie_type == event_ie_val->ie_type) {
break;
}
}
if (!event_ie_val) {
/* We did not find the event ie the subscriber cares about. */
return 0;
}
if (sub_ie_val->ie_pltype != event_ie_val->ie_pltype) {
if (sub_ie_val->ie_pltype == AST_EVENT_IE_PLTYPE_EXISTS) {
/* The subscription only cares that this ie exists. */
return 1;
}
/* Payload types do not match. */
return 0;
}
switch (sub_ie_val->ie_pltype) {
case AST_EVENT_IE_PLTYPE_UINT:
res = (sub_ie_val->payload.uint == event_ie_val->payload.uint);
break;
case AST_EVENT_IE_PLTYPE_BITFLAGS:
/*
* If the subscriber has requested *any* of the bitflags we are providing,
* then it's a match.
*/
res = (sub_ie_val->payload.uint & event_ie_val->payload.uint);
break;
case AST_EVENT_IE_PLTYPE_STR:
res = !strcmp(sub_ie_val->payload.str, event_ie_val->payload.str);
break;
case AST_EVENT_IE_PLTYPE_RAW:
res = (sub_ie_val->raw_datalen == event_ie_val->raw_datalen
&& !memcmp(sub_ie_val->payload.raw, event_ie_val->payload.raw,
sub_ie_val->raw_datalen));
break;
case AST_EVENT_IE_PLTYPE_EXISTS:
/* Should never get here since check_ie_vals cannot have this type. */
break;
case AST_EVENT_IE_PLTYPE_UNKNOWN:
/*
* Should never be in a subscription event ie val list and
* check_ie_vals cannot have this type either.
*/
break;
}
return res;
}
enum ast_event_subscriber_res ast_event_check_subscriber(enum ast_event_type type, ...)
{
va_list ap;
enum ast_event_ie_type ie_type;
enum ast_event_subscriber_res res = AST_EVENT_SUB_NONE;
struct ast_event_ie_val *ie_val;
struct ast_event_sub *sub;
struct ast_ev_check_list check_ie_vals = {
.ie_vals = AST_LIST_HEAD_NOLOCK_INIT_VALUE
};
const enum ast_event_type event_types[] = { type, AST_EVENT_ALL };
int i;
int want_specific_event;/* TRUE if looking for subscribers wanting specific parameters. */
if (type >= AST_EVENT_TOTAL) {
ast_log(LOG_ERROR, "%u is an invalid type!\n", type);
return res;
}
want_specific_event = 0;
va_start(ap, type);
for (ie_type = va_arg(ap, enum ast_event_ie_type);
ie_type != AST_EVENT_IE_END;
ie_type = va_arg(ap, enum ast_event_ie_type))
{
struct ast_event_ie_val *ie_value = alloca(sizeof(*ie_value));
int insert = 0;
memset(ie_value, 0, sizeof(*ie_value));
ie_value->ie_type = ie_type;
ie_value->ie_pltype = va_arg(ap, enum ast_event_ie_pltype);
switch (ie_value->ie_pltype) {
case AST_EVENT_IE_PLTYPE_UINT:
ie_value->payload.uint = va_arg(ap, uint32_t);
insert = 1;
break;
case AST_EVENT_IE_PLTYPE_BITFLAGS:
ie_value->payload.uint = va_arg(ap, uint32_t);
insert = 1;
break;
case AST_EVENT_IE_PLTYPE_STR:
ie_value->payload.str = va_arg(ap, const char *);
insert = 1;
break;
case AST_EVENT_IE_PLTYPE_RAW:
{
void *data = va_arg(ap, void *);
size_t datalen = va_arg(ap, size_t);
ie_value->payload.raw = alloca(datalen);
memcpy(ie_value->payload.raw, data, datalen);
ie_value->raw_datalen = datalen;
insert = 1;
break;
}
case AST_EVENT_IE_PLTYPE_UNKNOWN:
case AST_EVENT_IE_PLTYPE_EXISTS:
/* Unsupported payload type. */
break;
}
if (insert) {
want_specific_event = 1;
AST_LIST_INSERT_TAIL(&check_ie_vals.ie_vals, ie_value, entry);
} else {
ast_log(LOG_WARNING, "Unsupported PLTYPE(%d)\n", ie_value->ie_pltype);
}
}
va_end(ap);
for (i = 0; i < ARRAY_LEN(event_types); i++) {
AST_RWDLLIST_RDLOCK(&ast_event_subs[event_types[i]]);
if (want_specific_event) {
AST_RWDLLIST_TRAVERSE(&ast_event_subs[event_types[i]], sub, entry) {
AST_LIST_TRAVERSE(&sub->ie_vals, ie_val, entry) {
if (!match_sub_ie_val_to_event(ie_val, &check_ie_vals)) {
/* The current subscription ie did not match an event ie. */
break;
}
}
if (!ie_val) {
/* Everything matched. A subscriber is looking for this event. */
break;
}
}
} else {
/* Just looking to see if there are ANY subscribers to the event type. */
sub = AST_RWLIST_FIRST(&ast_event_subs[event_types[i]]);
}
AST_RWDLLIST_UNLOCK(&ast_event_subs[event_types[i]]);
if (sub) {
break;
}
}
return sub ? AST_EVENT_SUB_EXISTS : AST_EVENT_SUB_NONE;
}
/*!
* \internal
* \brief Check if an ie_val matches an event
*
* \param event event to check against
* \param ie_val IE value to check
* \param event2 optional event, if specified, the value to compare against will be pulled
* from this event instead of from the ie_val structure. In this case, only the IE
* type and payload type will be pulled from ie_val.
*
* \retval 0 not matched
* \retval non-zero matched
*/
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
static int match_ie_val(const struct ast_event *event,
const struct ast_event_ie_val *ie_val, const struct ast_event *event2)
{
switch (ie_val->ie_pltype) {
case AST_EVENT_IE_PLTYPE_UINT:
{
uint32_t val = event2 ? ast_event_get_ie_uint(event2, ie_val->ie_type) : ie_val->payload.uint;
return (val == ast_event_get_ie_uint(event, ie_val->ie_type)) ? 1 : 0;
}
case AST_EVENT_IE_PLTYPE_BITFLAGS:
{
uint32_t flags = event2 ? ast_event_get_ie_uint(event2, ie_val->ie_type) : ie_val->payload.uint;
/*
* If the subscriber has requested *any* of the bitflags that this event provides,
* then it's a match.
*/
return (flags & ast_event_get_ie_bitflags(event, ie_val->ie_type)) ? 1 : 0;
}
case AST_EVENT_IE_PLTYPE_STR:
{
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
const char *str;
uint32_t hash;
hash = event2 ? ast_event_get_ie_str_hash(event2, ie_val->ie_type) : ie_val->payload.hash;
if (hash != ast_event_get_ie_str_hash(event, ie_val->ie_type)) {
return 0;
}
str = event2 ? ast_event_get_ie_str(event2, ie_val->ie_type) : ie_val->payload.str;
if (str && !strcmp(str, ast_event_get_ie_str(event, ie_val->ie_type))) {
return 1;
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
}
return 0;
}
case AST_EVENT_IE_PLTYPE_RAW:
{
const void *buf = event2 ? ast_event_get_ie_raw(event2, ie_val->ie_type) : ie_val->payload.raw;
Merged revisions 295711 via svnmerge from https://origsvn.digium.com/svn/asterisk/branches/1.8 ................ r295711 | russell | 2010-11-19 18:50:00 -0600 (Fri, 19 Nov 2010) | 36 lines Merged revisions 295710 via svnmerge from https://origsvn.digium.com/svn/asterisk/branches/1.6.2 ........ r295710 | russell | 2010-11-19 18:45:51 -0600 (Fri, 19 Nov 2010) | 29 lines Fix cache of device state changes for multiple servers. This patch addresses a regression where device states across multiple servers were not being processing completely correctly. The code works to determine the overall state by looking at the last known state of a device on each server. However, there was a regression due to some invasive rewrites of how the cache works that led to the cache only storing the last device state change for a device, regardless of which server it was on. The code is set up to cache device state change events by ensuring that each event in the cache has a unique device name + entity ID (server ID). The code that was responsible for comparing raw information elements (which EID is) always returned a match due to a memcmp() with a length of 0. There isn't much code to fix the actual bug. This patch also introduces a new CLI command that was very useful for debugging this problem. The command allows you to dump the contents of the event cache. (closes issue #18284) Reported by: klaus3000 Patches: issue18284.rev1.txt uploaded by russell (license 2) Tested by: russell, klaus3000 (closes issue #18280) Reported by: klaus3000 Review: https://reviewboard.asterisk.org/r/1012/ ........ ................ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@295712 f38db490-d61c-443f-a65b-d21fe96a405b
2010-11-20 00:52:47 +00:00
uint16_t ie_payload_len = event2 ? ast_event_get_ie_raw_payload_len(event2, ie_val->ie_type) : ie_val->raw_datalen;
return (buf
&& ie_payload_len == ast_event_get_ie_raw_payload_len(event, ie_val->ie_type)
&& !memcmp(buf, ast_event_get_ie_raw(event, ie_val->ie_type), ie_payload_len)) ? 1 : 0;
}
case AST_EVENT_IE_PLTYPE_EXISTS:
{
return ast_event_get_ie_raw(event, ie_val->ie_type) ? 1 : 0;
}
case AST_EVENT_IE_PLTYPE_UNKNOWN:
return 0;
}
return 0;
}
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
static int dump_cache_cb(void *obj, void *arg, int flags)
{
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
const struct ast_event_ref *event_ref = obj;
const struct ast_event *event = event_ref->event;
const struct ast_event_sub *event_sub = arg;
struct ast_event_ie_val *ie_val = NULL;
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
AST_LIST_TRAVERSE(&event_sub->ie_vals, ie_val, entry) {
if (!match_ie_val(event, ie_val, NULL)) {
break;
}
}
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
if (!ie_val) {
/* All parameters were matched on this cache entry, so dump it */
event_sub->cb(event, event_sub->userdata);
}
return 0;
}
/*! \brief Dump the event cache for the subscribed event type */
void ast_event_dump_cache(const struct ast_event_sub *event_sub)
{
ao2_callback(ast_event_cache[event_sub->type].container, OBJ_NODATA,
dump_cache_cb, (void *) event_sub);
}
static struct ast_event *gen_sub_event(struct ast_event_sub *sub)
{
struct ast_event_ie_val *ie_val;
struct ast_event *event;
event = ast_event_new(AST_EVENT_SUB,
AST_EVENT_IE_UNIQUEID, AST_EVENT_IE_PLTYPE_UINT, sub->uniqueid,
AST_EVENT_IE_EVENTTYPE, AST_EVENT_IE_PLTYPE_UINT, sub->type,
AST_EVENT_IE_DESCRIPTION, AST_EVENT_IE_PLTYPE_STR, sub->description,
AST_EVENT_IE_END);
if (!event)
return NULL;
AST_LIST_TRAVERSE(&sub->ie_vals, ie_val, entry) {
switch (ie_val->ie_pltype) {
case AST_EVENT_IE_PLTYPE_UNKNOWN:
break;
case AST_EVENT_IE_PLTYPE_EXISTS:
ast_event_append_ie_uint(&event, AST_EVENT_IE_EXISTS, ie_val->ie_type);
break;
case AST_EVENT_IE_PLTYPE_UINT:
ast_event_append_ie_uint(&event, ie_val->ie_type, ie_val->payload.uint);
break;
case AST_EVENT_IE_PLTYPE_BITFLAGS:
ast_event_append_ie_bitflags(&event, ie_val->ie_type, ie_val->payload.uint);
break;
case AST_EVENT_IE_PLTYPE_STR:
ast_event_append_ie_str(&event, ie_val->ie_type, ie_val->payload.str);
break;
case AST_EVENT_IE_PLTYPE_RAW:
ast_event_append_ie_raw(&event, ie_val->ie_type, ie_val->payload.raw, ie_val->raw_datalen);
break;
}
if (!event)
break;
}
return event;
}
/*! \brief Send AST_EVENT_SUB events to this subscriber of ... subscriber events */
void ast_event_report_subs(const struct ast_event_sub *event_sub)
{
struct ast_event *event;
struct ast_event_sub *sub;
enum ast_event_type event_type = -1;
struct ast_event_ie_val *ie_val;
if (event_sub->type != AST_EVENT_SUB)
return;
AST_LIST_TRAVERSE(&event_sub->ie_vals, ie_val, entry) {
if (ie_val->ie_type == AST_EVENT_IE_EVENTTYPE) {
event_type = ie_val->payload.uint;
break;
}
}
if (event_type == -1)
return;
AST_RWDLLIST_RDLOCK(&ast_event_subs[event_type]);
AST_RWDLLIST_TRAVERSE(&ast_event_subs[event_type], sub, entry) {
if (event_sub == sub) {
continue;
}
event = gen_sub_event(sub);
if (!event) {
continue;
}
event_sub->cb(event, event_sub->userdata);
ast_event_destroy(event);
}
AST_RWDLLIST_UNLOCK(&ast_event_subs[event_type]);
}
struct ast_event_sub *ast_event_subscribe_new(enum ast_event_type type,
ast_event_cb_t cb, void *userdata)
{
struct ast_event_sub *sub;
if (type < 0 || type >= AST_EVENT_TOTAL) {
ast_log(LOG_ERROR, "%u is an invalid type!\n", type);
return NULL;
}
if (!(sub = ast_calloc(1, sizeof(*sub)))) {
return NULL;
}
sub->type = type;
sub->cb = cb;
sub->userdata = userdata;
sub->uniqueid = ast_atomic_fetchadd_int((int *) &sub_uniqueid, 1);
return sub;
}
int ast_event_sub_append_ie_uint(struct ast_event_sub *sub,
enum ast_event_ie_type ie_type, uint32_t unsigned_int)
{
struct ast_event_ie_val *ie_val;
if (ie_type <= 0 || ie_type >= AST_EVENT_IE_TOTAL) {
return -1;
}
if (!(ie_val = ast_calloc(1, sizeof(*ie_val)))) {
return -1;
}
ie_val->ie_type = ie_type;
ie_val->payload.uint = unsigned_int;
ie_val->ie_pltype = AST_EVENT_IE_PLTYPE_UINT;
AST_LIST_INSERT_TAIL(&sub->ie_vals, ie_val, entry);
return 0;
}
int ast_event_sub_append_ie_bitflags(struct ast_event_sub *sub,
enum ast_event_ie_type ie_type, uint32_t flags)
{
struct ast_event_ie_val *ie_val;
if (ie_type <= 0 || ie_type >= AST_EVENT_IE_TOTAL) {
return -1;
}
if (!(ie_val = ast_calloc(1, sizeof(*ie_val)))) {
return -1;
}
ie_val->ie_type = ie_type;
ie_val->payload.uint = flags;
ie_val->ie_pltype = AST_EVENT_IE_PLTYPE_BITFLAGS;
AST_LIST_INSERT_TAIL(&sub->ie_vals, ie_val, entry);
return 0;
}
int ast_event_sub_append_ie_exists(struct ast_event_sub *sub,
enum ast_event_ie_type ie_type)
{
struct ast_event_ie_val *ie_val;
if (ie_type <= 0 || ie_type >= AST_EVENT_IE_TOTAL) {
return -1;
}
if (!(ie_val = ast_calloc(1, sizeof(*ie_val)))) {
return -1;
}
ie_val->ie_type = ie_type;
ie_val->ie_pltype = AST_EVENT_IE_PLTYPE_EXISTS;
AST_LIST_INSERT_TAIL(&sub->ie_vals, ie_val, entry);
return 0;
}
int ast_event_sub_append_ie_str(struct ast_event_sub *sub,
enum ast_event_ie_type ie_type, const char *str)
{
struct ast_event_ie_val *ie_val;
if (ie_type <= 0 || ie_type >= AST_EVENT_IE_TOTAL) {
return -1;
}
if (!(ie_val = ast_calloc(1, sizeof(*ie_val)))) {
return -1;
}
ie_val->ie_type = ie_type;
ie_val->ie_pltype = AST_EVENT_IE_PLTYPE_STR;
if (!(ie_val->payload.str = ast_strdup(str))) {
ast_free(ie_val);
return -1;
}
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
ie_val->payload.hash = ast_str_hash(str);
AST_LIST_INSERT_TAIL(&sub->ie_vals, ie_val, entry);
return 0;
}
int ast_event_sub_append_ie_raw(struct ast_event_sub *sub,
enum ast_event_ie_type ie_type, void *data, size_t raw_datalen)
{
struct ast_event_ie_val *ie_val;
if (ie_type <= 0 || ie_type >= AST_EVENT_IE_TOTAL) {
return -1;
}
if (!(ie_val = ast_calloc(1, sizeof(*ie_val)))) {
return -1;
}
ie_val->ie_type = ie_type;
ie_val->ie_pltype = AST_EVENT_IE_PLTYPE_RAW;
ie_val->raw_datalen = raw_datalen;
if (!(ie_val->payload.raw = ast_malloc(raw_datalen))) {
ast_free(ie_val);
return -1;
}
memcpy(ie_val->payload.raw, data, raw_datalen);
AST_LIST_INSERT_TAIL(&sub->ie_vals, ie_val, entry);
return 0;
}
int ast_event_sub_activate(struct ast_event_sub *sub)
{
if (ast_event_check_subscriber(AST_EVENT_SUB,
AST_EVENT_IE_EVENTTYPE, AST_EVENT_IE_PLTYPE_UINT, sub->type,
AST_EVENT_IE_END) != AST_EVENT_SUB_NONE) {
struct ast_event *event;
event = gen_sub_event(sub);
if (event && ast_event_queue(event)) {
ast_event_destroy(event);
}
}
AST_RWDLLIST_WRLOCK(&ast_event_subs[sub->type]);
AST_RWDLLIST_INSERT_TAIL(&ast_event_subs[sub->type], sub, entry);
AST_RWDLLIST_UNLOCK(&ast_event_subs[sub->type]);
return 0;
}
struct ast_event_sub *ast_event_subscribe(enum ast_event_type type, ast_event_cb_t cb,
const char *description, void *userdata, ...)
{
va_list ap;
enum ast_event_ie_type ie_type;
struct ast_event_sub *sub;
if (!(sub = ast_event_subscribe_new(type, cb, userdata))) {
return NULL;
}
ast_copy_string(sub->description, description, sizeof(sub->description));
va_start(ap, userdata);
for (ie_type = va_arg(ap, enum ast_event_ie_type);
ie_type != AST_EVENT_IE_END;
ie_type = va_arg(ap, enum ast_event_ie_type))
{
enum ast_event_ie_pltype ie_pltype;
ie_pltype = va_arg(ap, enum ast_event_ie_pltype);
switch (ie_pltype) {
case AST_EVENT_IE_PLTYPE_UNKNOWN:
break;
case AST_EVENT_IE_PLTYPE_UINT:
{
uint32_t unsigned_int = va_arg(ap, uint32_t);
ast_event_sub_append_ie_uint(sub, ie_type, unsigned_int);
break;
}
case AST_EVENT_IE_PLTYPE_BITFLAGS:
{
uint32_t unsigned_int = va_arg(ap, uint32_t);
ast_event_sub_append_ie_bitflags(sub, ie_type, unsigned_int);
break;
}
case AST_EVENT_IE_PLTYPE_STR:
{
const char *str = va_arg(ap, const char *);
ast_event_sub_append_ie_str(sub, ie_type, str);
break;
}
case AST_EVENT_IE_PLTYPE_RAW:
{
void *data = va_arg(ap, void *);
size_t data_len = va_arg(ap, size_t);
ast_event_sub_append_ie_raw(sub, ie_type, data, data_len);
break;
}
case AST_EVENT_IE_PLTYPE_EXISTS:
ast_event_sub_append_ie_exists(sub, ie_type);
break;
}
}
va_end(ap);
ast_event_sub_activate(sub);
return sub;
}
void ast_event_sub_destroy(struct ast_event_sub *sub)
{
struct ast_event_ie_val *ie_val;
while ((ie_val = AST_LIST_REMOVE_HEAD(&sub->ie_vals, entry))) {
ast_event_ie_val_destroy(ie_val);
}
ast_free(sub);
}
const char *ast_event_subscriber_get_description(struct ast_event_sub *sub)
{
return sub ? sub->description : NULL;
}
struct ast_event_sub *ast_event_unsubscribe(struct ast_event_sub *sub)
{
struct ast_event *event;
AST_RWDLLIST_WRLOCK(&ast_event_subs[sub->type]);
AST_DLLIST_REMOVE(&ast_event_subs[sub->type], sub, entry);
AST_RWDLLIST_UNLOCK(&ast_event_subs[sub->type]);
if (ast_event_check_subscriber(AST_EVENT_UNSUB,
AST_EVENT_IE_EVENTTYPE, AST_EVENT_IE_PLTYPE_UINT, sub->type,
AST_EVENT_IE_END) != AST_EVENT_SUB_NONE) {
event = ast_event_new(AST_EVENT_UNSUB,
AST_EVENT_IE_UNIQUEID, AST_EVENT_IE_PLTYPE_UINT, sub->uniqueid,
AST_EVENT_IE_EVENTTYPE, AST_EVENT_IE_PLTYPE_UINT, sub->type,
AST_EVENT_IE_DESCRIPTION, AST_EVENT_IE_PLTYPE_STR, sub->description,
AST_EVENT_IE_END);
if (event && ast_event_queue(event)) {
ast_event_destroy(event);
}
}
ast_event_sub_destroy(sub);
return NULL;
}
int ast_event_iterator_init(struct ast_event_iterator *iterator, const struct ast_event *event)
{
int res = 0;
iterator->event_len = ast_event_get_size(event);
iterator->event = event;
if (iterator->event_len >= sizeof(*event) + sizeof(struct ast_event_ie)) {
iterator->ie = (struct ast_event_ie *) ( ((char *) event) + sizeof(*event) );
} else {
iterator->ie = NULL;
res = -1;
}
return res;
}
int ast_event_iterator_next(struct ast_event_iterator *iterator)
{
iterator->ie = (struct ast_event_ie *) ( ((char *) iterator->ie) + sizeof(*iterator->ie) + ntohs(iterator->ie->ie_payload_len));
return ((iterator->event_len <= (((char *) iterator->ie) - ((char *) iterator->event))) ? -1 : 0);
}
enum ast_event_ie_type ast_event_iterator_get_ie_type(struct ast_event_iterator *iterator)
{
return ntohs(iterator->ie->ie_type);
}
uint32_t ast_event_iterator_get_ie_uint(struct ast_event_iterator *iterator)
{
return ntohl(get_unaligned_uint32(iterator->ie->ie_payload));
}
uint32_t ast_event_iterator_get_ie_bitflags(struct ast_event_iterator *iterator)
{
return ntohl(get_unaligned_uint32(iterator->ie->ie_payload));
}
const char *ast_event_iterator_get_ie_str(struct ast_event_iterator *iterator)
{
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
const struct ast_event_ie_str_payload *str_payload;
str_payload = (struct ast_event_ie_str_payload *) iterator->ie->ie_payload;
return str_payload ? str_payload->str : NULL;
}
void *ast_event_iterator_get_ie_raw(struct ast_event_iterator *iterator)
{
return iterator->ie->ie_payload;
}
Merged revisions 295711 via svnmerge from https://origsvn.digium.com/svn/asterisk/branches/1.8 ................ r295711 | russell | 2010-11-19 18:50:00 -0600 (Fri, 19 Nov 2010) | 36 lines Merged revisions 295710 via svnmerge from https://origsvn.digium.com/svn/asterisk/branches/1.6.2 ........ r295710 | russell | 2010-11-19 18:45:51 -0600 (Fri, 19 Nov 2010) | 29 lines Fix cache of device state changes for multiple servers. This patch addresses a regression where device states across multiple servers were not being processing completely correctly. The code works to determine the overall state by looking at the last known state of a device on each server. However, there was a regression due to some invasive rewrites of how the cache works that led to the cache only storing the last device state change for a device, regardless of which server it was on. The code is set up to cache device state change events by ensuring that each event in the cache has a unique device name + entity ID (server ID). The code that was responsible for comparing raw information elements (which EID is) always returned a match due to a memcmp() with a length of 0. There isn't much code to fix the actual bug. This patch also introduces a new CLI command that was very useful for debugging this problem. The command allows you to dump the contents of the event cache. (closes issue #18284) Reported by: klaus3000 Patches: issue18284.rev1.txt uploaded by russell (license 2) Tested by: russell, klaus3000 (closes issue #18280) Reported by: klaus3000 Review: https://reviewboard.asterisk.org/r/1012/ ........ ................ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@295712 f38db490-d61c-443f-a65b-d21fe96a405b
2010-11-20 00:52:47 +00:00
uint16_t ast_event_iterator_get_ie_raw_payload_len(struct ast_event_iterator *iterator)
{
return ntohs(iterator->ie->ie_payload_len);
}
enum ast_event_type ast_event_get_type(const struct ast_event *event)
{
return ntohs(event->type);
}
uint32_t ast_event_get_ie_uint(const struct ast_event *event, enum ast_event_ie_type ie_type)
{
const uint32_t *ie_val;
ie_val = ast_event_get_ie_raw(event, ie_type);
return ie_val ? ntohl(get_unaligned_uint32(ie_val)) : 0;
}
uint32_t ast_event_get_ie_bitflags(const struct ast_event *event, enum ast_event_ie_type ie_type)
{
const uint32_t *ie_val;
ie_val = ast_event_get_ie_raw(event, ie_type);
return ie_val ? ntohl(get_unaligned_uint32(ie_val)) : 0;
}
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
uint32_t ast_event_get_ie_str_hash(const struct ast_event *event, enum ast_event_ie_type ie_type)
{
const struct ast_event_ie_str_payload *str_payload;
str_payload = ast_event_get_ie_raw(event, ie_type);
return str_payload ? str_payload->hash : 0;
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
}
const char *ast_event_get_ie_str(const struct ast_event *event, enum ast_event_ie_type ie_type)
{
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
const struct ast_event_ie_str_payload *str_payload;
str_payload = ast_event_get_ie_raw(event, ie_type);
return str_payload ? str_payload->str : NULL;
}
const void *ast_event_get_ie_raw(const struct ast_event *event, enum ast_event_ie_type ie_type)
{
struct ast_event_iterator iterator;
int res;
for (res = ast_event_iterator_init(&iterator, event); !res; res = ast_event_iterator_next(&iterator)) {
if (ast_event_iterator_get_ie_type(&iterator) == ie_type) {
return ast_event_iterator_get_ie_raw(&iterator);
}
}
return NULL;
}
Merged revisions 295711 via svnmerge from https://origsvn.digium.com/svn/asterisk/branches/1.8 ................ r295711 | russell | 2010-11-19 18:50:00 -0600 (Fri, 19 Nov 2010) | 36 lines Merged revisions 295710 via svnmerge from https://origsvn.digium.com/svn/asterisk/branches/1.6.2 ........ r295710 | russell | 2010-11-19 18:45:51 -0600 (Fri, 19 Nov 2010) | 29 lines Fix cache of device state changes for multiple servers. This patch addresses a regression where device states across multiple servers were not being processing completely correctly. The code works to determine the overall state by looking at the last known state of a device on each server. However, there was a regression due to some invasive rewrites of how the cache works that led to the cache only storing the last device state change for a device, regardless of which server it was on. The code is set up to cache device state change events by ensuring that each event in the cache has a unique device name + entity ID (server ID). The code that was responsible for comparing raw information elements (which EID is) always returned a match due to a memcmp() with a length of 0. There isn't much code to fix the actual bug. This patch also introduces a new CLI command that was very useful for debugging this problem. The command allows you to dump the contents of the event cache. (closes issue #18284) Reported by: klaus3000 Patches: issue18284.rev1.txt uploaded by russell (license 2) Tested by: russell, klaus3000 (closes issue #18280) Reported by: klaus3000 Review: https://reviewboard.asterisk.org/r/1012/ ........ ................ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@295712 f38db490-d61c-443f-a65b-d21fe96a405b
2010-11-20 00:52:47 +00:00
uint16_t ast_event_get_ie_raw_payload_len(const struct ast_event *event, enum ast_event_ie_type ie_type)
{
struct ast_event_iterator iterator;
int res;
for (res = ast_event_iterator_init(&iterator, event); !res; res = ast_event_iterator_next(&iterator)) {
if (ast_event_iterator_get_ie_type(&iterator) == ie_type) {
return ast_event_iterator_get_ie_raw_payload_len(&iterator);
}
}
return 0;
}
int ast_event_append_ie_str(struct ast_event **event, enum ast_event_ie_type ie_type,
const char *str)
{
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
struct ast_event_ie_str_payload *str_payload;
size_t payload_len;
payload_len = sizeof(*str_payload) + strlen(str);
str_payload = alloca(payload_len);
strcpy(str_payload->str, str);
str_payload->hash = ast_str_hash(str);
return ast_event_append_ie_raw(event, ie_type, str_payload, payload_len);
}
int ast_event_append_ie_uint(struct ast_event **event, enum ast_event_ie_type ie_type,
uint32_t data)
{
data = htonl(data);
return ast_event_append_ie_raw(event, ie_type, &data, sizeof(data));
}
int ast_event_append_ie_bitflags(struct ast_event **event, enum ast_event_ie_type ie_type,
uint32_t flags)
{
flags = htonl(flags);
return ast_event_append_ie_raw(event, ie_type, &flags, sizeof(flags));
}
int ast_event_append_ie_raw(struct ast_event **event, enum ast_event_ie_type ie_type,
const void *data, size_t data_len)
{
struct ast_event_ie *ie;
unsigned int extra_len;
uint16_t event_len;
event_len = ntohs((*event)->event_len);
extra_len = sizeof(*ie) + data_len;
if (!(*event = ast_realloc(*event, event_len + extra_len))) {
return -1;
}
ie = (struct ast_event_ie *) ( ((char *) *event) + event_len );
ie->ie_type = htons(ie_type);
ie->ie_payload_len = htons(data_len);
memcpy(ie->ie_payload, data, data_len);
(*event)->event_len = htons(event_len + extra_len);
return 0;
}
struct ast_event *ast_event_new(enum ast_event_type type, ...)
{
va_list ap;
struct ast_event *event;
enum ast_event_ie_type ie_type;
struct ast_event_ie_val *ie_val;
int has_ie = 0;
AST_LIST_HEAD_NOLOCK_STATIC(ie_vals, ast_event_ie_val);
/* Invalid type */
if (type >= AST_EVENT_TOTAL) {
ast_log(LOG_WARNING, "Someone tried to create an event of invalid "
"type '%d'!\n", type);
return NULL;
}
va_start(ap, type);
for (ie_type = va_arg(ap, enum ast_event_ie_type);
ie_type != AST_EVENT_IE_END;
ie_type = va_arg(ap, enum ast_event_ie_type))
{
struct ast_event_ie_val *ie_value = alloca(sizeof(*ie_value));
int insert = 0;
memset(ie_value, 0, sizeof(*ie_value));
ie_value->ie_type = ie_type;
ie_value->ie_pltype = va_arg(ap, enum ast_event_ie_pltype);
switch (ie_value->ie_pltype) {
case AST_EVENT_IE_PLTYPE_UINT:
ie_value->payload.uint = va_arg(ap, uint32_t);
insert = 1;
break;
case AST_EVENT_IE_PLTYPE_BITFLAGS:
ie_value->payload.uint = va_arg(ap, uint32_t);
insert = 1;
break;
case AST_EVENT_IE_PLTYPE_STR:
ie_value->payload.str = va_arg(ap, const char *);
insert = 1;
break;
case AST_EVENT_IE_PLTYPE_RAW:
{
void *data = va_arg(ap, void *);
size_t datalen = va_arg(ap, size_t);
ie_value->payload.raw = alloca(datalen);
memcpy(ie_value->payload.raw, data, datalen);
ie_value->raw_datalen = datalen;
insert = 1;
break;
}
case AST_EVENT_IE_PLTYPE_UNKNOWN:
case AST_EVENT_IE_PLTYPE_EXISTS:
break;
}
if (insert) {
AST_LIST_INSERT_TAIL(&ie_vals, ie_value, entry);
has_ie = 1;
} else {
ast_log(LOG_WARNING, "Unsupported PLTYPE(%d)\n", ie_value->ie_pltype);
}
}
va_end(ap);
if (!(event = ast_calloc(1, sizeof(*event)))) {
return NULL;
}
event->type = htons(type);
event->event_len = htons(sizeof(*event));
AST_LIST_TRAVERSE(&ie_vals, ie_val, entry) {
switch (ie_val->ie_pltype) {
case AST_EVENT_IE_PLTYPE_STR:
ast_event_append_ie_str(&event, ie_val->ie_type, ie_val->payload.str);
break;
case AST_EVENT_IE_PLTYPE_UINT:
ast_event_append_ie_uint(&event, ie_val->ie_type, ie_val->payload.uint);
break;
case AST_EVENT_IE_PLTYPE_BITFLAGS:
ast_event_append_ie_bitflags(&event, ie_val->ie_type, ie_val->payload.uint);
break;
case AST_EVENT_IE_PLTYPE_RAW:
ast_event_append_ie_raw(&event, ie_val->ie_type,
ie_val->payload.raw, ie_val->raw_datalen);
break;
case AST_EVENT_IE_PLTYPE_EXISTS:
case AST_EVENT_IE_PLTYPE_UNKNOWN:
break;
}
if (!event) {
break;
}
}
if (has_ie && !ast_event_get_ie_raw(event, AST_EVENT_IE_EID)) {
/* If the event is originating on this server, add the server's
* entity ID to the event. */
ast_event_append_eid(&event);
}
return event;
}
int ast_event_append_eid(struct ast_event **event)
{
return ast_event_append_ie_raw(event, AST_EVENT_IE_EID,
&ast_eid_default, sizeof(ast_eid_default));
}
void ast_event_destroy(struct ast_event *event)
{
ast_free(event);
}
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
static void ast_event_ref_destroy(void *obj)
{
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
struct ast_event_ref *event_ref = obj;
ast_event_destroy(event_ref->event);
}
static struct ast_event *ast_event_dup(const struct ast_event *event)
{
struct ast_event *dup_event;
uint16_t event_len;
event_len = ast_event_get_size(event);
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
if (!(dup_event = ast_calloc(1, event_len))) {
return NULL;
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
}
memcpy(dup_event, event, event_len);
return dup_event;
}
struct ast_event *ast_event_get_cached(enum ast_event_type type, ...)
{
va_list ap;
enum ast_event_ie_type ie_type;
struct ast_event *dup_event = NULL;
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
struct ast_event_ref *cached_event_ref;
struct ast_event *cache_arg_event;
struct ast_event_ref tmp_event_ref = {
.event = NULL,
};
struct ao2_container *container = NULL;
if (type >= AST_EVENT_TOTAL) {
ast_log(LOG_ERROR, "%u is an invalid type!\n", type);
return NULL;
}
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
if (!(container = ast_event_cache[type].container)) {
ast_log(LOG_ERROR, "%u is not a cached event type\n", type);
return NULL;
}
if (!(cache_arg_event = ast_event_new(type, AST_EVENT_IE_END))) {
return NULL;
}
va_start(ap, type);
for (ie_type = va_arg(ap, enum ast_event_ie_type);
ie_type != AST_EVENT_IE_END;
ie_type = va_arg(ap, enum ast_event_ie_type))
{
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
enum ast_event_ie_pltype ie_pltype;
ie_pltype = va_arg(ap, enum ast_event_ie_pltype);
switch (ie_pltype) {
case AST_EVENT_IE_PLTYPE_UINT:
ast_event_append_ie_uint(&cache_arg_event, ie_type, va_arg(ap, uint32_t));
break;
case AST_EVENT_IE_PLTYPE_BITFLAGS:
ast_event_append_ie_bitflags(&cache_arg_event, ie_type, va_arg(ap, uint32_t));
break;
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
case AST_EVENT_IE_PLTYPE_STR:
ast_event_append_ie_str(&cache_arg_event, ie_type, va_arg(ap, const char *));
break;
case AST_EVENT_IE_PLTYPE_RAW:
{
void *data = va_arg(ap, void *);
size_t datalen = va_arg(ap, size_t);
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
ast_event_append_ie_raw(&cache_arg_event, ie_type, data, datalen);
break;
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
}
case AST_EVENT_IE_PLTYPE_EXISTS:
ast_log(LOG_WARNING, "PLTYPE_EXISTS not supported by this function\n");
break;
case AST_EVENT_IE_PLTYPE_UNKNOWN:
break;
}
}
va_end(ap);
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
tmp_event_ref.event = cache_arg_event;
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
cached_event_ref = ao2_find(container, &tmp_event_ref, OBJ_POINTER);
ast_event_destroy(cache_arg_event);
cache_arg_event = NULL;
if (cached_event_ref) {
dup_event = ast_event_dup(cached_event_ref->event);
ao2_ref(cached_event_ref, -1);
cached_event_ref = NULL;
}
return dup_event;
}
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
static struct ast_event_ref *alloc_event_ref(void)
{
return ao2_alloc(sizeof(struct ast_event_ref), ast_event_ref_destroy);
}
/*!
* \internal
* \brief Update the given event cache with the new event.
* \since 1.8
*
* \param cache Event cache container to update.
* \param event New event to put in the cache.
*
* \return Nothing
*/
static void event_update_cache(struct ao2_container *cache, struct ast_event *event)
{
struct ast_event_ref tmp_event_ref = {
.event = event,
};
struct ast_event *dup_event;
struct ast_event_ref *event_ref;
/* Hold the cache container lock while it is updated. */
ao2_lock(cache);
/* Remove matches from the cache. */
ao2_callback(cache, OBJ_POINTER | OBJ_UNLINK | OBJ_MULTIPLE | OBJ_NODATA,
ast_event_cmp, &tmp_event_ref);
/* Save a copy of the event in the cache. */
dup_event = ast_event_dup(event);
if (dup_event) {
event_ref = alloc_event_ref();
if (event_ref) {
event_ref->event = dup_event;
ao2_link(cache, event_ref);
ao2_ref(event_ref, -1);
} else {
ast_event_destroy(dup_event);
}
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
}
ao2_unlock(cache);
}
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
int ast_event_queue_and_cache(struct ast_event *event)
{
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
struct ao2_container *container;
container = ast_event_cache[ast_event_get_type(event)].container;
if (!container) {
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
ast_log(LOG_WARNING, "cache requested for non-cached event type\n");
} else {
event_update_cache(container, event);
}
if (ast_event_queue(event)) {
ast_event_destroy(event);
}
return 0;
}
static int handle_event(void *data)
{
struct ast_event_ref *event_ref = data;
struct ast_event_sub *sub;
const enum ast_event_type event_types[] = {
ntohs(event_ref->event->type),
AST_EVENT_ALL
};
int i;
for (i = 0; i < ARRAY_LEN(event_types); i++) {
AST_RWDLLIST_RDLOCK(&ast_event_subs[event_types[i]]);
AST_RWDLLIST_TRAVERSE(&ast_event_subs[event_types[i]], sub, entry) {
struct ast_event_ie_val *ie_val;
AST_LIST_TRAVERSE(&sub->ie_vals, ie_val, entry) {
if (!match_ie_val(event_ref->event, ie_val, NULL)) {
/* The current subscription ie did not match an event ie. */
break;
}
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
}
if (ie_val) {
/* The event did not match this subscription. */
continue;
}
sub->cb(event_ref->event, sub->userdata);
}
AST_RWDLLIST_UNLOCK(&ast_event_subs[event_types[i]]);
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
}
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
ao2_ref(event_ref, -1);
return 0;
}
int ast_event_queue(struct ast_event *event)
{
struct ast_event_ref *event_ref;
uint16_t host_event_type;
int res;
host_event_type = ntohs(event->type);
/* Invalid type */
if (host_event_type >= AST_EVENT_TOTAL) {
ast_log(LOG_WARNING, "Someone tried to queue an event of invalid "
"type '%d'!\n", host_event_type);
return -1;
}
/* If nobody has subscribed to this event type, throw it away now */
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
if (ast_event_check_subscriber(host_event_type, AST_EVENT_IE_END)
== AST_EVENT_SUB_NONE) {
ast_event_destroy(event);
return 0;
}
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
if (!(event_ref = alloc_event_ref())) {
return -1;
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
}
event_ref->event = event;
res = ast_taskprocessor_push(event_dispatcher, handle_event, event_ref);
if (res) {
event_ref->event = NULL;
ao2_ref(event_ref, -1);
}
return res;
}
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
static int ast_event_hash_mwi(const void *obj, const int flags)
{
const struct ast_event *event = obj;
const char *mailbox = ast_event_get_ie_str(event, AST_EVENT_IE_MAILBOX);
const char *context = ast_event_get_ie_str(event, AST_EVENT_IE_CONTEXT);
return ast_str_hash_add(context, ast_str_hash(mailbox));
}
/*!
* \internal
* \brief Hash function for AST_EVENT_DEVICE_STATE
*
* \param[in] obj an ast_event
* \param[in] flags unused
*
* \return hash value
*/
static int ast_event_hash_devstate(const void *obj, const int flags)
{
const struct ast_event *event = obj;
return ast_str_hash(ast_event_get_ie_str(event, AST_EVENT_IE_DEVICE));
}
/*!
* \internal
* \brief Hash function for AST_EVENT_DEVICE_STATE_CHANGE
*
* \param[in] obj an ast_event
* \param[in] flags unused
*
* \return hash value
*/
static int ast_event_hash_devstate_change(const void *obj, const int flags)
{
const struct ast_event *event = obj;
return ast_str_hash(ast_event_get_ie_str(event, AST_EVENT_IE_DEVICE));
}
static int ast_event_hash(const void *obj, const int flags)
{
const struct ast_event_ref *event_ref;
const struct ast_event *event;
ao2_hash_fn *hash_fn;
event_ref = obj;
event = event_ref->event;
if (!(hash_fn = ast_event_cache[ast_event_get_type(event)].hash_fn)) {
return 0;
}
return hash_fn(event, flags);
}
/*!
* \internal
* \brief Compare two events
*
* \param[in] obj the first event, as an ast_event_ref
* \param[in] arg the second event, as an ast_event_ref
* \param[in] flags unused
*
* \pre Both events must be the same type.
* \pre The event type must be declared as a cached event type in ast_event_cache
*
* \details This function takes two events, and determines if they are considered
* equivalent. The values of information elements specified in the cache arguments
* for the event type are used to determine if the events are equivalent.
*
* \retval 0 No match
* \retval CMP_MATCH The events are considered equivalent based on the cache arguments
*/
static int ast_event_cmp(void *obj, void *arg, int flags)
{
struct ast_event_ref *event_ref, *event_ref2;
struct ast_event *event, *event2;
int res = CMP_MATCH;
int i;
enum ast_event_ie_type *cache_args;
event_ref = obj;
event = event_ref->event;
event_ref2 = arg;
event2 = event_ref2->event;
cache_args = ast_event_cache[ast_event_get_type(event)].cache_args;
for (i = 0; i < ARRAY_LEN(ast_event_cache[0].cache_args) && cache_args[i]; i++) {
struct ast_event_ie_val ie_val = {
.ie_pltype = ast_event_get_ie_pltype(cache_args[i]),
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
.ie_type = cache_args[i],
};
if (!match_ie_val(event, &ie_val, event2)) {
res = 0;
break;
}
}
return res;
}
Merged revisions 295711 via svnmerge from https://origsvn.digium.com/svn/asterisk/branches/1.8 ................ r295711 | russell | 2010-11-19 18:50:00 -0600 (Fri, 19 Nov 2010) | 36 lines Merged revisions 295710 via svnmerge from https://origsvn.digium.com/svn/asterisk/branches/1.6.2 ........ r295710 | russell | 2010-11-19 18:45:51 -0600 (Fri, 19 Nov 2010) | 29 lines Fix cache of device state changes for multiple servers. This patch addresses a regression where device states across multiple servers were not being processing completely correctly. The code works to determine the overall state by looking at the last known state of a device on each server. However, there was a regression due to some invasive rewrites of how the cache works that led to the cache only storing the last device state change for a device, regardless of which server it was on. The code is set up to cache device state change events by ensuring that each event in the cache has a unique device name + entity ID (server ID). The code that was responsible for comparing raw information elements (which EID is) always returned a match due to a memcmp() with a length of 0. There isn't much code to fix the actual bug. This patch also introduces a new CLI command that was very useful for debugging this problem. The command allows you to dump the contents of the event cache. (closes issue #18284) Reported by: klaus3000 Patches: issue18284.rev1.txt uploaded by russell (license 2) Tested by: russell, klaus3000 (closes issue #18280) Reported by: klaus3000 Review: https://reviewboard.asterisk.org/r/1012/ ........ ................ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@295712 f38db490-d61c-443f-a65b-d21fe96a405b
2010-11-20 00:52:47 +00:00
static void dump_raw_ie(struct ast_event_iterator *i, struct ast_cli_args *a)
{
char eid_buf[32];
enum ast_event_ie_type ie_type;
const char *ie_type_name;
ie_type = ast_event_iterator_get_ie_type(i);
ie_type_name = ast_event_get_ie_type_name(ie_type);
switch (ie_type) {
case AST_EVENT_IE_EID:
ast_eid_to_str(eid_buf, sizeof(eid_buf), ast_event_iterator_get_ie_raw(i));
ast_cli(a->fd, "%.30s: %s\n", ie_type_name, eid_buf);
break;
default:
ast_cli(a->fd, "%s\n", ie_type_name);
break;
}
}
static int event_dump_cli(void *obj, void *arg, int flags)
{
const struct ast_event_ref *event_ref = obj;
const struct ast_event *event = event_ref->event;
struct ast_cli_args *a = arg;
struct ast_event_iterator i;
if (ast_event_iterator_init(&i, event)) {
ast_cli(a->fd, "Failed to initialize event iterator. :-(\n");
return 0;
}
ast_cli(a->fd, "Event: %s\n", ast_event_get_type_name(event));
do {
enum ast_event_ie_type ie_type;
enum ast_event_ie_pltype ie_pltype;
const char *ie_type_name;
ie_type = ast_event_iterator_get_ie_type(&i);
ie_type_name = ast_event_get_ie_type_name(ie_type);
ie_pltype = ast_event_get_ie_pltype(ie_type);
switch (ie_pltype) {
case AST_EVENT_IE_PLTYPE_UNKNOWN:
case AST_EVENT_IE_PLTYPE_EXISTS:
ast_cli(a->fd, "%s\n", ie_type_name);
break;
case AST_EVENT_IE_PLTYPE_STR:
ast_cli(a->fd, "%.30s: %s\n", ie_type_name,
ast_event_iterator_get_ie_str(&i));
break;
case AST_EVENT_IE_PLTYPE_UINT:
ast_cli(a->fd, "%.30s: %u\n", ie_type_name,
ast_event_iterator_get_ie_uint(&i));
break;
case AST_EVENT_IE_PLTYPE_BITFLAGS:
ast_cli(a->fd, "%.30s: %u\n", ie_type_name,
ast_event_iterator_get_ie_bitflags(&i));
break;
case AST_EVENT_IE_PLTYPE_RAW:
dump_raw_ie(&i, a);
break;
}
} while (!ast_event_iterator_next(&i));
ast_cli(a->fd, "\n");
return 0;
}
static char *event_dump_cache(struct ast_cli_entry *e, int cmd, struct ast_cli_args *a)
{
enum ast_event_type event_type;
enum ast_event_ie_type *cache_args;
int i;
switch (cmd) {
case CLI_INIT:
e->command = "event dump cache";
e->usage =
"Usage: event dump cache <event type>\n"
" Dump all of the cached events for the given event type.\n"
" This is primarily intended for debugging.\n";
return NULL;
case CLI_GENERATE:
if (a->pos == 3) {
return ast_cli_complete(a->word, cached_event_types, a->n);
}
return NULL;
case CLI_HANDLER:
break;
}
if (a->argc != e->args + 1) {
return CLI_SHOWUSAGE;
}
if (ast_event_str_to_event_type(a->argv[e->args], &event_type)) {
ast_cli(a->fd, "Invalid cached event type: '%s'\n", a->argv[e->args]);
return CLI_SHOWUSAGE;
}
if (!ast_event_cache[event_type].container) {
ast_cli(a->fd, "Event type '%s' has no cache.\n", a->argv[e->args]);
return CLI_SUCCESS;
}
ast_cli(a->fd, "Event Type: %s\n", a->argv[e->args]);
ast_cli(a->fd, "Cache Unique Keys:\n");
cache_args = ast_event_cache[event_type].cache_args;
for (i = 0; i < ARRAY_LEN(ast_event_cache[0].cache_args) && cache_args[i]; i++) {
ast_cli(a->fd, "--> %s\n", ast_event_get_ie_type_name(cache_args[i]));
}
ast_cli(a->fd, "\n--- Begin Cache Dump ---\n\n");
ao2_callback(ast_event_cache[event_type].container, OBJ_NODATA, event_dump_cli, a);
ast_cli(a->fd, "--- End Cache Dump ---\n\n");
return CLI_SUCCESS;
}
static struct ast_cli_entry event_cli[] = {
AST_CLI_DEFINE(event_dump_cache, "Dump the internal event cache (for debugging)"),
};
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
int ast_event_init(void)
{
int i;
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
for (i = 0; i < AST_EVENT_TOTAL; i++) {
AST_RWDLLIST_HEAD_INIT(&ast_event_subs[i]);
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
}
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
for (i = 0; i < AST_EVENT_TOTAL; i++) {
if (!ast_event_cache[i].hash_fn) {
/* This event type is not cached. */
continue;
}
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
if (!(ast_event_cache[i].container = ao2_container_alloc(NUM_CACHE_BUCKETS,
ast_event_hash, ast_event_cmp))) {
return -1;
}
}
if (!(event_dispatcher = ast_taskprocessor_get("core_event_dispatcher", 0))) {
return -1;
}
Merged revisions 295711 via svnmerge from https://origsvn.digium.com/svn/asterisk/branches/1.8 ................ r295711 | russell | 2010-11-19 18:50:00 -0600 (Fri, 19 Nov 2010) | 36 lines Merged revisions 295710 via svnmerge from https://origsvn.digium.com/svn/asterisk/branches/1.6.2 ........ r295710 | russell | 2010-11-19 18:45:51 -0600 (Fri, 19 Nov 2010) | 29 lines Fix cache of device state changes for multiple servers. This patch addresses a regression where device states across multiple servers were not being processing completely correctly. The code works to determine the overall state by looking at the last known state of a device on each server. However, there was a regression due to some invasive rewrites of how the cache works that led to the cache only storing the last device state change for a device, regardless of which server it was on. The code is set up to cache device state change events by ensuring that each event in the cache has a unique device name + entity ID (server ID). The code that was responsible for comparing raw information elements (which EID is) always returned a match due to a memcmp() with a length of 0. There isn't much code to fix the actual bug. This patch also introduces a new CLI command that was very useful for debugging this problem. The command allows you to dump the contents of the event cache. (closes issue #18284) Reported by: klaus3000 Patches: issue18284.rev1.txt uploaded by russell (license 2) Tested by: russell, klaus3000 (closes issue #18280) Reported by: klaus3000 Review: https://reviewboard.asterisk.org/r/1012/ ........ ................ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@295712 f38db490-d61c-443f-a65b-d21fe96a405b
2010-11-20 00:52:47 +00:00
ast_cli_register_multiple(event_cli, ARRAY_LEN(event_cli));
Improve performance of the ast_event cache functionality. This code comes from svn/asterisk/team/russell/event_performance/. Here is a summary of the changes that have been made, in order of both invasiveness and performance impact, from smallest to largest. 1) Asterisk 1.6.1 introduces some additional logic to be able to handle distributed device state. This functionality comes at a cost. One relatively minor change in this patch is that the extra processing required for distributed device state is now completely bypassed if it's not needed. 2) One of the things that I noticed when profiling this code was that a _lot_ of time was spent doing string comparisons. I changed the way strings are represented in an event to include a hash value at the front. So, before doing a string comparison, we do an integer comparison on the hash. 3) Finally, the code that handles the event cache has been re-written. I tried to do this in a such a way that it had minimal impact on the API. I did have to change one API call, though - ast_event_queue_and_cache(). However, the way it works now is nicer, IMO. Each type of event that can be cached (MWI, device state) has its own hash table and rules for hashing and comparing objects. This by far made the biggest impact on performance. For additional details regarding this code and how it was tested, please see the review request. (closes issue #14738) Reported by: russell Review: http://reviewboard.digium.com/r/205/ git-svn-id: http://svn.digium.com/svn/asterisk/trunk@184339 f38db490-d61c-443f-a65b-d21fe96a405b
2009-03-25 21:57:19 +00:00
return 0;
}