[patch] exit some threads early during gracful shutdown of a process

event-exit-threads-early.diff (text/plain), 10.33 KB, created by Stefan Fritsch on 2015-10-03 15:31:46 UTC

(hide)

Description: exit some threads early during gracful shutdown of a process

Filename:

MIME Type:

Creator: Stefan Fritsch

Created: 2015-10-03 15:31:46 UTC

Size: 10.33 KB

patch

obsolete

>Index: server/mpm/event/event.c >=================================================================== >--- server/mpm/event/event.c (revision 1706598) >+++ server/mpm/event/event.c (working copy) >@@ -181,6 +181,8 @@ > static apr_uint32_t lingering_count = 0; /* Number of connections in lingering close */ > static apr_uint32_t suspended_count = 0; /* Number of suspended connections */ > static apr_uint32_t clogged_count = 0; /* Number of threads processing ssl conns */ >+static apr_uint32_t threads_shutdown = 0; /* Number of threads that have shutdown >+ early during graceful termination */ > static int resource_shortage = 0; > static fd_queue_t *worker_queue; > static fd_queue_info_t *worker_queue_info; >@@ -288,9 +290,8 @@ > /* The structure used to pass unique initialization info to each thread */ > typedef struct > { >- int pid; >- int tid; >- int sd; >+ int pslot; /* process slot */ >+ int tslot; /* worker slot of the thread */ > } proc_info; > > /* Structure used to pass information to the thread responsible for >@@ -1207,6 +1208,7 @@ > { > if (!*closed) { > int i; >+ worker_score *ws; > disable_listensocks(process_slot); > ap_close_listeners_ex(my_bucket->listeners); > *closed = 1; >@@ -1213,11 +1215,19 @@ > dying = 1; > ap_scoreboard_image->parent[process_slot].quiescing = 1; > for (i = 0; i < threads_per_child; ++i) { >+ ws = ap_get_scoreboard_worker_from_indexes(process_slot, i); >+ if (ws->pid != ap_my_pid) { >+ /* scoreboard slot still in use by previous generation */ >+ continue; >+ } > ap_update_child_status_from_indexes(process_slot, i, > SERVER_GRACEFUL, NULL); > } > /* wake up the main thread */ > kill(ap_my_pid, SIGTERM); >+ >+ ap_free_idle_pools(worker_queue_info); >+ ap_queue_interrupt_all(worker_queue); > } > } > >@@ -1517,7 +1527,7 @@ > timer_event_t *te; > apr_status_t rc; > proc_info *ti = dummy; >- int process_slot = ti->pid; >+ int process_slot = ti->pslot; > apr_pool_t *tpool = apr_thread_pool_get(thd); > void *csd = NULL; > apr_pool_t *ptrans; /* Pool for per-transaction stuff */ >@@ -1583,6 +1593,12 @@ > *keepalive_q->total, > apr_atomic_read32(&lingering_count), > apr_atomic_read32(&suspended_count)); >+ if (dying) { >+ ap_log_error(APLOG_MARK, APLOG_TRACE6, 0, ap_server_conf, >+ "%u/%u workers shutdown", >+ apr_atomic_read32(&threads_shutdown), >+ threads_per_child); >+ } > apr_thread_mutex_unlock(timeout_mutex); > } > } >@@ -1816,11 +1832,12 @@ > /* If all workers are busy, we kill older keep-alive connections so that they > * may connect to another process. > */ >- if (workers_were_busy && *keepalive_q->total) { >- ap_log_error(APLOG_MARK, APLOG_TRACE1, 0, ap_server_conf, >- "All workers are busy, will close %d keep-alive " >- "connections", >- *keepalive_q->total); >+ if ((workers_were_busy || dying) && *keepalive_q->total) { >+ if (!dying) >+ ap_log_error(APLOG_MARK, APLOG_TRACE1, 0, ap_server_conf, >+ "All workers are busy, will close %d keep-alive " >+ "connections", >+ *keepalive_q->total); > process_timeout_queue(keepalive_q, 0, > start_lingering_close_nonblocking); > } >@@ -1867,6 +1884,34 @@ > return NULL; > } > >+/* >+ * During graceful shutdown, if there are more running worker threads than >+ * open connections, exit one worker thread. >+ * >+ * return 1 if thread should exit, 0 if it should continue running. >+ */ >+static int worker_thread_should_exit_early(void) >+{ >+ for (;;) { >+ apr_uint32_t conns = apr_atomic_read32(&connection_count); >+ apr_uint32_t dead = apr_atomic_read32(&threads_shutdown); >+ apr_uint32_t newdead; >+ >+ AP_DEBUG_ASSERT(dead <= threads_per_child); >+ if (conns >= threads_per_child - dead) >+ return 0; >+ >+ newdead = dead + 1; >+ if (apr_atomic_cas32(&threads_shutdown, newdead, dead) == dead) { >+ /* >+ * No other thread has exited in the mean time, safe to exit >+ * this one. >+ */ >+ return 1; >+ } >+ } >+} >+ > /* XXX For ungraceful termination/restart, we definitely don't want to > * wait for active connections to finish but we may want to wait > * for idle workers to get out of the queue code and release mutexes, >@@ -1877,8 +1922,8 @@ > static void *APR_THREAD_FUNC worker_thread(apr_thread_t * thd, void *dummy) > { > proc_info *ti = dummy; >- int process_slot = ti->pid; >- int thread_slot = ti->tid; >+ int process_slot = ti->pslot; >+ int thread_slot = ti->tslot; > apr_socket_t *csd = NULL; > event_conn_state_t *cs; > apr_pool_t *ptrans; /* Pool for per-transaction stuff */ >@@ -1913,6 +1958,9 @@ > if (workers_may_exit) { > break; > } >+ if (dying && worker_thread_should_exit_early()) { >+ break; >+ } > > te = NULL; > rv = ap_queue_pop_something(worker_queue, &csd, &cs, &ptrans, &te); >@@ -1991,9 +2039,8 @@ > apr_status_t rv; > > my_info = (proc_info *) ap_malloc(sizeof(proc_info)); >- my_info->pid = my_child_num; >- my_info->tid = -1; /* listener thread doesn't have a thread slot */ >- my_info->sd = 0; >+ my_info->pslot = my_child_num; >+ my_info->tslot = -1; /* listener thread doesn't have a thread slot */ > rv = apr_thread_create(&ts->listener, thread_attr, listener_thread, > my_info, pchild); > if (rv != APR_SUCCESS) { >@@ -2102,14 +2149,13 @@ > int status = > ap_scoreboard_image->servers[child_num_arg][i].status; > >- if (status != SERVER_GRACEFUL && status != SERVER_DEAD) { >+ if (status != SERVER_DEAD) { > continue; > } > > my_info = (proc_info *) ap_malloc(sizeof(proc_info)); >- my_info->pid = my_child_num; >- my_info->tid = i; >- my_info->sd = 0; >+ my_info->pslot = my_child_num; >+ my_info->tslot = i; > > /* We are creating threads right now */ > ap_update_child_status_from_indexes(my_child_num, i, >@@ -2167,7 +2213,8 @@ > return NULL; > } > >-static void join_workers(apr_thread_t * listener, apr_thread_t ** threads) >+static void join_workers(apr_thread_t * listener, apr_thread_t ** threads, >+ int pslot) > { > int i; > apr_status_t rv, thread_rv; >@@ -2203,16 +2250,35 @@ > } > } > >+ /* first join threads which we know have already exited */ > for (i = 0; i < threads_per_child; i++) { >- if (threads[i]) { /* if we ever created this thread */ >- rv = apr_thread_join(&thread_rv, threads[i]); >- if (rv != APR_SUCCESS) { >- ap_log_error(APLOG_MARK, APLOG_CRIT, rv, ap_server_conf, APLOGNO(00477) >- "apr_thread_join: unable to join worker " >- "thread %d", i); >- } >+ worker_score *ws; >+ if (threads[i] == NULL) { >+ /* we have never created this thread */ >+ continue; > } >+ ws = &ap_scoreboard_image->servers[pslot][i]; >+ if (ws->pid != ap_my_pid || ws->status != SERVER_DEAD) { >+ /* not dead yet */ >+ /* XXX wrong pid should not happen? */ >+ continue; >+ } >+ rv = apr_thread_join(&thread_rv, threads[i]); >+ if (rv != APR_SUCCESS) { >+ ap_log_error(APLOG_MARK, APLOG_CRIT, rv, ap_server_conf, APLOGNO(02967) >+ "apr_thread_join: unable to join worker thread %d", i); >+ } >+ threads[i] = NULL; > } >+ for (i = 0; i < threads_per_child; i++) { >+ if (threads[i] == NULL) >+ continue; >+ rv = apr_thread_join(&thread_rv, threads[i]); >+ if (rv != APR_SUCCESS) { >+ ap_log_error(APLOG_MARK, APLOG_CRIT, rv, ap_server_conf, APLOGNO(00477) >+ "apr_thread_join: unable to join worker thread %d", i); >+ } >+ } > } > > static void join_start_thread(apr_thread_t * start_thread_id) >@@ -2352,7 +2418,7 @@ > * If the worker hasn't exited, then this blocks until > * they have (then cleans up). > */ >- join_workers(ts->listener, threads); >+ join_workers(ts->listener, threads, child_num_arg); > } > else { /* !one_process */ > /* remove SIGTERM from the set of blocked signals... if one of >@@ -2393,7 +2459,7 @@ > * If the worker hasn't exited, then this blocks until > * they have (then cleans up). > */ >- join_workers(ts->listener, threads); >+ join_workers(ts->listener, threads, child_num_arg); > } > > free(threads); >Index: server/mpm/event/fdqueue.c >=================================================================== >--- server/mpm/event/fdqueue.c (revision 1706598) >+++ server/mpm/event/fdqueue.c (working copy) >@@ -280,6 +280,19 @@ > } > } > >+void ap_free_idle_pools(fd_queue_info_t *queue_info) >+{ >+ apr_pool_t *p; >+ >+ queue_info->max_recycled_pools = 0; >+ do { >+ ap_pop_pool(&p, queue_info); >+ if (p != NULL) >+ apr_pool_destroy(p); >+ } while (p != NULL); >+} >+ >+ > apr_status_t ap_queue_info_term(fd_queue_info_t * queue_info) > { > apr_status_t rv; >Index: server/mpm/event/fdqueue.h >=================================================================== >--- server/mpm/event/fdqueue.h (revision 1706598) >+++ server/mpm/event/fdqueue.h (working copy) >@@ -52,6 +52,7 @@ > int *had_to_block); > apr_status_t ap_queue_info_term(fd_queue_info_t * queue_info); > apr_uint32_t ap_queue_info_get_idlers(fd_queue_info_t * queue_info); >+void ap_free_idle_pools(fd_queue_info_t *queue_info); > > struct fd_queue_elem_t > { View Attachment As Diff View Attachment As Raw

This is ASF Bugzilla: the Apache Software Foundation bug system. In case of problems with the functioning of ASF Bugzilla, please contact bugzilla-admin@apache.org. Please Note: this e-mail address is only for reporting problems with ASF Bugzilla. Mail about any other subject will be silently ignored.

• • Home
  • | New
  • | Browse
  • | Search
  • |
    [?]
  • | Reports
  • | Help
  • | New Account
  • | Log In
    Remember [x]
  • | Forgot Password
    Login: [x]