]> git.ipfire.org Git - thirdparty/asterisk.git/commitdiff
res_pjsip: Deny requests when threadpool queue is backed up. 07/1607/2
authorMark Michelson <mmichelson@digium.com>
Wed, 11 Nov 2015 23:11:53 +0000 (17:11 -0600)
committerMark Michelson <mmichelson@digium.com>
Thu, 12 Nov 2015 16:41:27 +0000 (11:41 -0500)
We have observed situations where the SIP threadpool may become
deadlocked. However, because incoming traffic is still arriving, the SIP
threadpool's queue can continue to grow, eventually running the system
out of memory.

This change makes it so that incoming traffic gets rejected with a 503
response if the queue is backed up too much.

Change-Id: I4e736d48a2ba79fd1f8056c0dcd330e38e6a3816

include/asterisk/res_pjsip.h
include/asterisk/taskprocessor.h
include/asterisk/threadpool.h
main/taskprocessor.c
main/threadpool.c
res/res_pjsip.c
res/res_pjsip/pjsip_distributor.c

index 459082901f82308a150b7f88c4836b74eda276d8..37b7662119b372e8f5a8c6b19bc42c036f23d661 100644 (file)
@@ -2116,4 +2116,10 @@ int ast_sip_get_host_ip(int af, pj_sockaddr *addr);
  */
 const char *ast_sip_get_host_ip_string(int af);
 
+/*!
+ * \brief Return the size of the SIP threadpool's task queue
+ * \since 13.7.0
+ */
+long ast_sip_threadpool_queue_size(void);
+
 #endif /* _RES_PJSIP_H */
index f16f144cb558ef9fcdc130a4ce91da89d8987dfc..06368867a8f309d24666b1a7941d6fa3380a4108 100644 (file)
@@ -262,4 +262,10 @@ int ast_taskprocessor_is_task(struct ast_taskprocessor *tps);
  */
 const char *ast_taskprocessor_name(struct ast_taskprocessor *tps);
 
+/*!
+ * \brief Return the current size of the taskprocessor queue
+ * \since 13.7.0
+ */
+long ast_taskprocessor_size(struct ast_taskprocessor *tps);
+
 #endif /* __AST_TASKPROCESSOR_H__ */
index 75ce0e4e4ef114901e48ef3062015601c9a79653..0f360c7a4a49539d567fd19b04544c470e2d7360 100644 (file)
@@ -292,4 +292,10 @@ struct ast_taskprocessor *ast_threadpool_serializer(const char *name, struct ast
 struct ast_taskprocessor *ast_threadpool_serializer_group(const char *name,
        struct ast_threadpool *pool, struct ast_serializer_shutdown_group *shutdown_group);
 
+/*!
+ * \brief Return the size of the threadpool's task queue
+ * \since 13.7.0
+ */
+long ast_threadpool_queue_size(struct ast_threadpool *pool);
+
 #endif /* ASTERISK_THREADPOOL_H */
index f382814af9796f4beb13321f98ea2814c9a793ab..eefa85f6162cb2165d33758f7d6fd366b6975c96 100644 (file)
@@ -130,9 +130,6 @@ static int tps_ping_handler(void *datap);
 /*! \brief Remove the front task off the taskprocessor queue */
 static struct tps_task *tps_taskprocessor_pop(struct ast_taskprocessor *tps);
 
-/*! \brief Return the size of the taskprocessor queue */
-static int tps_taskprocessor_depth(struct ast_taskprocessor *tps);
-
 static char *cli_tps_ping(struct ast_cli_entry *e, int cmd, struct ast_cli_args *a);
 static char *cli_tps_report(struct ast_cli_entry *e, int cmd, struct ast_cli_args *a);
 
@@ -508,7 +505,7 @@ static struct tps_task *tps_taskprocessor_pop(struct ast_taskprocessor *tps)
        return task;
 }
 
-static int tps_taskprocessor_depth(struct ast_taskprocessor *tps)
+long ast_taskprocessor_size(struct ast_taskprocessor *tps)
 {
        return (tps) ? tps->tps_queue_size : -1;
 }
@@ -765,7 +762,7 @@ int ast_taskprocessor_execute(struct ast_taskprocessor *tps)
 {
        struct ast_taskprocessor_local local;
        struct tps_task *t;
-       int size;
+       long size;
 
        ao2_lock(tps);
        t = tps_taskprocessor_pop(tps);
@@ -797,7 +794,7 @@ int ast_taskprocessor_execute(struct ast_taskprocessor *tps)
         * after we pop an empty stack.
         */
        tps->executing = 0;
-       size = tps_taskprocessor_depth(tps);
+       size = ast_taskprocessor_size(tps);
        /* If we executed a task, bump the stats */
        if (tps->stats) {
                tps->stats->_tasks_processed_count++;
index 46de9b7f805266f4c8bdb186ee4f7214c1c9301b..60e1e9a3b474cf8d1471f7d800d1196f67ba445e 100644 (file)
@@ -1397,3 +1397,8 @@ struct ast_taskprocessor *ast_threadpool_serializer(const char *name, struct ast
 {
        return ast_threadpool_serializer_group(name, pool, NULL);
 }
+
+long ast_threadpool_queue_size(struct ast_threadpool *pool)
+{
+       return ast_taskprocessor_size(pool->tps);
+}
index d2b393fccf930633e91362d24d377438676fd666..babbe7aaaa943bbb9f7018571baaa310fce218d6 100644 (file)
@@ -3755,6 +3755,11 @@ static void remove_request_headers(pjsip_endpoint *endpt)
        }
 }
 
+long ast_sip_threadpool_queue_size(void)
+{
+       return ast_threadpool_queue_size(sip_threadpool);
+}
+
 AST_TEST_DEFINE(xml_sanitization_end_null)
 {
        char sanitized[8];
index 9b052603a954d756fa04ff86dedd8864f809e689..1d39e0fd2802f230014e79c07322392872ce5e66 100644 (file)
@@ -246,6 +246,8 @@ static pjsip_module endpoint_mod = {
        .on_rx_request = endpoint_lookup,
 };
 
+#define SIP_MAX_QUEUE 500L
+
 static pj_bool_t distributor(pjsip_rx_data *rdata)
 {
        pjsip_dialog *dlg = find_dialog(rdata);
@@ -280,7 +282,17 @@ static pj_bool_t distributor(pjsip_rx_data *rdata)
                clone->endpt_info.mod_data[endpoint_mod.id] = ao2_bump(dist->endpoint);
        }
 
-       ast_sip_push_task(serializer, distribute, clone);
+       if (ast_sip_threadpool_queue_size() > SIP_MAX_QUEUE) {
+               /* When the threadpool is backed up this much, there is a good chance that we have encountered
+                * some sort of terrible condition and don't need to be adding more work to the threadpool.
+                * It's in our best interest to send back a 503 response and be done with it.
+                */
+               pjsip_endpt_respond_stateless(ast_sip_get_pjsip_endpoint(), rdata, 503, NULL, NULL, NULL);
+               ao2_cleanup(clone->endpt_info.mod_data[endpoint_mod.id]);
+               pjsip_rx_data_free_cloned(clone);
+       } else {
+               ast_sip_push_task(serializer, distribute, clone);
+       }
 
 end:
        if (dlg) {