*
* After a client completes the first request, the client can keep the
* connection open to send more requests with the same socket. This can save
- * signifigant overhead in creating TCP connections. However, the major
+ * significant overhead in creating TCP connections. However, the major
* disadvantage is that Apache traditionally keeps an entire child
* process/thread waiting for data from the client. To solve this problem,
- * this MPM has a dedicated thread for handling both the Listenting sockets,
+ * this MPM has a dedicated thread for handling both the Listening sockets,
* and all sockets that are in a Keep Alive status.
*
* The MPM assumes the underlying apr_pollset implementation is somewhat
* enables the MPM to avoid extra high level locking or having to wake up the
* listener thread when a keep-alive socket needs to be sent to it.
*
- * This MPM not preform well on older platforms that do not have very good
+ * This MPM does not perform well on older platforms that do not have very good
* threading, like Linux with a 2.4 kernel, but this does not matter, since we
* require EPoll or KQueue.
*
#include "apr_poll.h"
#include "apr_ring.h"
#include "apr_queue.h"
+#include "apr_atomic.h"
#define APR_WANT_STRFUNC
#include "apr_want.h"
#include "apr_version.h"
+#include <stdlib.h>
+
#if APR_HAVE_UNISTD_H
#include <unistd.h>
#endif
#include "http_config.h" /* for read_config */
#include "http_core.h" /* for get_remote_host */
#include "http_connection.h"
+#include "http_protocol.h"
#include "ap_mpm.h"
-#include "pod.h"
#include "mpm_common.h"
#include "ap_listen.h"
#include "scoreboard.h"
#include "mpm_default.h"
#include "http_vhost.h"
#include "unixd.h"
+#include "apr_skiplist.h"
#include <signal.h>
#include <limits.h> /* for INT_MAX */
-#if HAVE_SERF
-#include "mod_serf.h"
-#include "serf.h"
-#endif
-
/* Limit on the total --- clients will be locked out if more servers than
* this are needed. It is intended solely to keep the server from crashing
* when things get out of hand.
#define apr_time_from_msec(x) (x * 1000)
#endif
+#ifndef MAX_SECS_TO_LINGER
+#define MAX_SECS_TO_LINGER 30
+#endif
+#define SECONDS_TO_LINGER 2
+
/*
* Actual definitions of config globals
*/
-static int threads_per_child = 0; /* Worker threads per child */
-static int ap_daemons_to_start = 0;
-static int min_spare_threads = 0;
-static int max_spare_threads = 0;
-static int ap_daemons_limit = 0;
-static int max_clients = 0;
-static int server_limit = 0;
-static int thread_limit = 0;
+#ifndef DEFAULT_WORKER_FACTOR
+#define DEFAULT_WORKER_FACTOR 2
+#endif
+#define WORKER_FACTOR_SCALE 16 /* scale factor to allow fractional values */
+static unsigned int worker_factor = DEFAULT_WORKER_FACTOR * WORKER_FACTOR_SCALE;
+ /* AsyncRequestWorkerFactor * 16 */
+
+static int threads_per_child = 0; /* ThreadsPerChild */
+static int ap_daemons_to_start = 0; /* StartServers */
+static int min_spare_threads = 0; /* MinSpareThreads */
+static int max_spare_threads = 0; /* MaxSpareThreads */
+static int active_daemons_limit = 0; /* MaxRequestWorkers / ThreadsPerChild */
+static int active_daemons = 0; /* workers that still active, i.e. are
+ not shutting down gracefully */
+static int max_workers = 0; /* MaxRequestWorkers */
+static int server_limit = 0; /* ServerLimit */
+static int thread_limit = 0; /* ThreadLimit */
+static int had_healthy_child = 0;
static int dying = 0;
static int workers_may_exit = 0;
static int start_thread_may_exit = 0;
static int listener_may_exit = 0;
-static int requests_this_child;
+static int listener_is_wakeable = 0; /* Pollset supports APR_POLLSET_WAKEABLE */
static int num_listensocks = 0;
+static apr_int32_t conns_this_child; /* MaxConnectionsPerChild, only access
+ in listener thread */
+static apr_uint32_t connection_count = 0; /* Number of open connections */
+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;
-static int mpm_state = AP_MPMQ_STARTING;
-static int sick_child_detected;
-static ap_generation_t volatile my_generation = 0;
static apr_thread_mutex_t *timeout_mutex;
-APR_RING_HEAD(timeout_head_t, conn_state_t);
-static struct timeout_head_t timeout_head, keepalive_timeout_head;
+module AP_MODULE_DECLARE_DATA mpm_event_module;
+
+/* forward declare */
+struct event_srv_cfg_s;
+typedef struct event_srv_cfg_s event_srv_cfg;
+
+static apr_pollfd_t *listener_pollfd;
+
+/*
+ * The pollset for sockets that are in any of the timeout queues. Currently
+ * we use the timeout_mutex to make sure that connections are added/removed
+ * atomically to/from both event_pollset and a timeout queue. Otherwise
+ * some confusion can happen under high load if timeout queues and pollset
+ * get out of sync.
+ * XXX: It should be possible to make the lock unnecessary in many or even all
+ * XXX: cases.
+ */
static apr_pollset_t *event_pollset;
-#if HAVE_SERF
-typedef struct {
- apr_pollset_t *pollset;
- apr_pool_t *pool;
-} s_baton_t;
+/*
+ * The chain of connections to be shutdown by a worker thread (deferred),
+ * linked list updated atomically.
+ */
+static event_conn_state_t *volatile defer_linger_chain;
+
+struct event_conn_state_t {
+ /** APR_RING of expiration timeouts */
+ APR_RING_ENTRY(event_conn_state_t) timeout_list;
+ /** the time when the entry was queued */
+ apr_time_t queue_timestamp;
+ /** connection record this struct refers to */
+ conn_rec *c;
+ /** request record (if any) this struct refers to */
+ request_rec *r;
+ /** server config this struct refers to */
+ event_srv_cfg *sc;
+ /** is the current conn_rec suspended? (disassociated with
+ * a particular MPM thread; for suspend_/resume_connection
+ * hooks)
+ */
+ int suspended;
+ /** memory pool to allocate from */
+ apr_pool_t *p;
+ /** bucket allocator */
+ apr_bucket_alloc_t *bucket_alloc;
+ /** poll file descriptor information */
+ apr_pollfd_t pfd;
+ /** public parts of the connection state */
+ conn_state_t pub;
+ /** chaining in defer_linger_chain */
+ struct event_conn_state_t *chain;
+};
-static serf_context_t *g_serf;
-#endif
+APR_RING_HEAD(timeout_head_t, event_conn_state_t);
+
+struct timeout_queue {
+ struct timeout_head_t head;
+ apr_interval_time_t timeout;
+ apr_uint32_t count; /* for this queue */
+ apr_uint32_t *total; /* for all chained/related queues */
+ struct timeout_queue *next; /* chaining */
+};
+/*
+ * Several timeout queues that use different timeouts, so that we always can
+ * simply append to the end.
+ * write_completion_q uses vhost's TimeOut
+ * keepalive_q uses vhost's KeepAliveTimeOut
+ * linger_q uses MAX_SECS_TO_LINGER
+ * short_linger_q uses SECONDS_TO_LINGER
+ */
+static struct timeout_queue *write_completion_q,
+ *keepalive_q,
+ *linger_q,
+ *short_linger_q;
+static volatile apr_time_t queues_next_expiry;
+
+/* Prevent extra poll/wakeup calls for timeouts close in the future (queues
+ * have the granularity of a second anyway).
+ * XXX: Wouldn't 0.5s (instead of 0.1s) be "enough"?
+ */
+#define TIMEOUT_FUDGE_FACTOR apr_time_from_msec(100)
+
+/*
+ * Macros for accessing struct timeout_queue.
+ * For TO_QUEUE_APPEND and TO_QUEUE_REMOVE, timeout_mutex must be held.
+ */
+static void TO_QUEUE_APPEND(struct timeout_queue *q, event_conn_state_t *el)
+{
+ apr_time_t q_expiry;
+ apr_time_t next_expiry;
+
+ APR_RING_INSERT_TAIL(&q->head, el, event_conn_state_t, timeout_list);
+ apr_atomic_inc32(q->total);
+ ++q->count;
+
+ /* Cheaply update the overall queues' next expiry according to the
+ * first entry of this queue (oldest), if necessary.
+ */
+ el = APR_RING_FIRST(&q->head);
+ q_expiry = el->queue_timestamp + q->timeout;
+ next_expiry = queues_next_expiry;
+ if (!next_expiry || next_expiry > q_expiry + TIMEOUT_FUDGE_FACTOR) {
+ queues_next_expiry = q_expiry;
+ /* Unblock the poll()ing listener for it to update its timeout. */
+ if (listener_is_wakeable) {
+ apr_pollset_wakeup(event_pollset);
+ }
+ }
+}
+
+static void TO_QUEUE_REMOVE(struct timeout_queue *q, event_conn_state_t *el)
+{
+ APR_RING_REMOVE(el, timeout_list);
+ APR_RING_ELEM_INIT(el, timeout_list);
+ apr_atomic_dec32(q->total);
+ --q->count;
+}
+
+static struct timeout_queue *TO_QUEUE_MAKE(apr_pool_t *p, apr_time_t t,
+ struct timeout_queue *ref)
+{
+ struct timeout_queue *q;
+
+ q = apr_pcalloc(p, sizeof *q);
+ APR_RING_INIT(&q->head, event_conn_state_t, timeout_list);
+ q->total = (ref) ? ref->total : apr_pcalloc(p, sizeof *q->total);
+ q->timeout = t;
+
+ return q;
+}
+
+#define TO_QUEUE_ELEM_INIT(el) \
+ APR_RING_ELEM_INIT((el), timeout_list)
/* 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
{
PT_CSD,
PT_ACCEPT
-#if HAVE_SERF
- , PT_SERF
-#endif
} poll_type_e;
typedef struct
{
poll_type_e type;
- int bypass_push;
void *baton;
} listener_poll_type;
* subsequent calls to pre-config hook
*/
typedef struct event_retained_data {
+ ap_unixd_mpm_retained_data *mpm;
+
int first_server_limit;
int first_thread_limit;
- int module_loads;
+ int sick_child_detected;
+ int maxclients_reported;
+ /*
+ * The max child slot ever assigned, preserved across restarts. Necessary
+ * to deal with MaxRequestWorkers changes across AP_SIG_GRACEFUL restarts.
+ * We use this value to optimize routines that have to scan the entire
+ * scoreboard.
+ */
+ int max_daemons_limit;
+
+ /*
+ * All running workers, active and shutting down, including those that
+ * may be left from before a graceful restart.
+ * Not kept up-to-date when shutdown is pending.
+ */
+ int total_daemons;
+
+ /*
+ * idle_spawn_rate is the number of children that will be spawned on the
+ * next maintenance cycle if there aren't enough idle servers. It is
+ * maintained per listeners bucket, doubled up to MAX_SPAWN_RATE, and
+ * reset only when a cycle goes by without the need to spawn.
+ */
+ int *idle_spawn_rate;
+#ifndef MAX_SPAWN_RATE
+#define MAX_SPAWN_RATE (32)
+#endif
+ int hold_off_on_exponential_spawning;
} event_retained_data;
static event_retained_data *retained;
+
+typedef struct event_child_bucket {
+ ap_pod_t *pod;
+ ap_listen_rec *listeners;
+} event_child_bucket;
+static event_child_bucket *all_buckets, /* All listeners buckets */
+ *my_bucket; /* Current child bucket */
+
+struct event_srv_cfg_s {
+ struct timeout_queue *wc_q,
+ *ka_q;
+};
#define ID_FROM_CHILD_THREAD(c, t) ((c * thread_limit) + t)
-/*
- * The max child slot ever assigned, preserved across restarts. Necessary
- * to deal with MaxClients changes across AP_SIG_GRACEFUL restarts. We
- * use this value to optimize routines that have to scan the entire
- * scoreboard.
- */
-static int max_daemons_limit = -1;
-
-static ap_event_pod_t *pod;
-
/* The event MPM respects a couple of runtime flags that can aid
* in debugging. Setting the -DNO_DETACH flag will prevent the root process
* from detaching from its controlling terminal. Additionally, setting
*/
static apr_socket_t **worker_sockets;
+static void disable_listensocks(int process_slot)
+{
+ int i;
+ for (i = 0; i < num_listensocks; i++) {
+ apr_pollset_remove(event_pollset, &listener_pollfd[i]);
+ }
+ ap_scoreboard_image->parent[process_slot].not_accepting = 1;
+}
+
+static void enable_listensocks(int process_slot)
+{
+ int i;
+ if (listener_may_exit) {
+ return;
+ }
+ ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, ap_server_conf, APLOGNO(00457)
+ "Accepting new connections again: "
+ "%u active conns (%u lingering/%u clogged/%u suspended), "
+ "%u idle workers",
+ apr_atomic_read32(&connection_count),
+ apr_atomic_read32(&lingering_count),
+ apr_atomic_read32(&clogged_count),
+ apr_atomic_read32(&suspended_count),
+ ap_queue_info_get_idlers(worker_queue_info));
+ for (i = 0; i < num_listensocks; i++)
+ apr_pollset_add(event_pollset, &listener_pollfd[i]);
+ /*
+ * XXX: This is not yet optimal. If many workers suddenly become available,
+ * XXX: the parent may kill some processes off too soon.
+ */
+ ap_scoreboard_image->parent[process_slot].not_accepting = 0;
+}
+
+static void abort_socket_nonblocking(apr_socket_t *csd)
+{
+ apr_status_t rv;
+ apr_socket_timeout_set(csd, 0);
+#if defined(SOL_SOCKET) && defined(SO_LINGER)
+ /* This socket is over now, and we don't want to block nor linger
+ * anymore, so reset it. A normal close could still linger in the
+ * system, while RST is fast, nonblocking, and what the peer will
+ * get if it sends us further data anyway.
+ */
+ {
+ apr_os_sock_t osd = -1;
+ struct linger opt;
+ opt.l_onoff = 1;
+ opt.l_linger = 0; /* zero timeout is RST */
+ apr_os_sock_get(&osd, csd);
+ setsockopt(osd, SOL_SOCKET, SO_LINGER, (void *)&opt, sizeof opt);
+ }
+#endif
+ rv = apr_socket_close(csd);
+ if (rv != APR_SUCCESS) {
+ ap_log_error(APLOG_MARK, APLOG_ERR, rv, ap_server_conf, APLOGNO(00468)
+ "error closing socket");
+ AP_DEBUG_ASSERT(0);
+ }
+}
+
static void close_worker_sockets(void)
{
int i;
for (i = 0; i < threads_per_child; i++) {
- if (worker_sockets[i]) {
- apr_socket_close(worker_sockets[i]);
+ apr_socket_t *csd = worker_sockets[i];
+ if (csd) {
worker_sockets[i] = NULL;
+ abort_socket_nonblocking(csd);
}
}
+ for (;;) {
+ event_conn_state_t *cs = defer_linger_chain;
+ if (!cs) {
+ break;
+ }
+ if (apr_atomic_casptr((void *)&defer_linger_chain, cs->chain,
+ cs) != cs) {
+ /* Race lost, try again */
+ continue;
+ }
+ cs->chain = NULL;
+ abort_socket_nonblocking(cs->pfd.desc.s);
+ }
}
static void wakeup_listener(void)
return;
}
+ /* Unblock the listener if it's poll()ing */
+ if (listener_is_wakeable) {
+ apr_pollset_wakeup(event_pollset);
+ }
+
/* unblock the listener if it's waiting for a worker */
- ap_queue_info_term(worker_queue_info);
+ ap_queue_info_term(worker_queue_info);
/*
* we should just be able to "kill(ap_my_pid, LISTENER_SIGNAL)" on all
return;
}
terminate_mode = mode;
- mpm_state = AP_MPMQ_STOPPING;
+ retained->mpm->mpm_state = AP_MPMQ_STOPPING;
/* in case we weren't called from the listener thread, wake up the
* listener thread
*rv = APR_SUCCESS;
switch (query_code) {
case AP_MPMQ_MAX_DAEMON_USED:
- *result = max_daemons_limit;
+ *result = retained->max_daemons_limit;
break;
case AP_MPMQ_IS_THREADED:
*result = AP_MPMQ_STATIC;
case AP_MPMQ_IS_ASYNC:
*result = 1;
break;
- case AP_MPMQ_HAS_SERF:
- *result = 1;
- break;
case AP_MPMQ_HARD_LIMIT_DAEMONS:
*result = server_limit;
break;
*result = ap_max_requests_per_child;
break;
case AP_MPMQ_MAX_DAEMONS:
- *result = ap_daemons_limit;
+ *result = active_daemons_limit;
break;
case AP_MPMQ_MPM_STATE:
- *result = mpm_state;
+ *result = retained->mpm->mpm_state;
break;
case AP_MPMQ_GENERATION:
- *result = my_generation;
+ *result = retained->mpm->my_generation;
break;
default:
*rv = APR_ENOTIMPL;
return OK;
}
-static apr_status_t event_note_child_killed(int childnum)
+static void event_note_child_killed(int childnum, pid_t pid, ap_generation_t gen)
{
- ap_scoreboard_image->parent[childnum].pid = 0;
- return APR_SUCCESS;
+ if (childnum != -1) { /* child had a scoreboard slot? */
+ ap_run_child_status(ap_server_conf,
+ ap_scoreboard_image->parent[childnum].pid,
+ ap_scoreboard_image->parent[childnum].generation,
+ childnum, MPM_CHILD_EXITED);
+ ap_scoreboard_image->parent[childnum].pid = 0;
+ }
+ else {
+ ap_run_child_status(ap_server_conf, pid, gen, -1, MPM_CHILD_EXITED);
+ }
+}
+
+static void event_note_child_started(int slot, pid_t pid)
+{
+ ap_scoreboard_image->parent[slot].pid = pid;
+ ap_run_child_status(ap_server_conf,
+ ap_scoreboard_image->parent[slot].pid,
+ retained->mpm->my_generation, slot, MPM_CHILD_STARTED);
}
static const char *event_get_name(void)
static void clean_child_exit(int code) __attribute__ ((noreturn));
static void clean_child_exit(int code)
{
- mpm_state = AP_MPMQ_STOPPING;
+ retained->mpm->mpm_state = AP_MPMQ_STOPPING;
if (pchild) {
apr_pool_destroy(pchild);
}
+
+ if (one_process) {
+ event_note_child_killed(/* slot */ 0, 0, 0);
+ }
+
exit(code);
}
* Connection structures and accounting...
*/
-/* volatile just in case */
-static int volatile shutdown_pending;
-static int volatile restart_pending;
-static int volatile is_graceful;
-static volatile int child_fatal;
-
-/*
- * ap_start_shutdown() and ap_start_restart(), below, are a first stab at
- * functions to initiate shutdown or restart without relying on signals.
- * Previously this was initiated in sig_term() and restart() signal handlers,
- * but we want to be able to start a shutdown/restart from other sources --
- * e.g. on Win32, from the service manager. Now the service manager can
- * call ap_start_shutdown() or ap_start_restart() as appropiate. Note that
- * these functions can also be called by the child processes, since global
- * variables are no longer used to pass on the required action to the parent.
- *
- * These should only be called from the parent process itself, since the
- * parent process will use the shutdown_pending and restart_pending variables
- * to determine whether to shutdown or restart. The child process should
- * call signal_parent() directly to tell the parent to die -- this will
- * cause neither of those variable to be set, which the parent will
- * assume means something serious is wrong (which it will be, for the
- * child to force an exit) and so do an exit anyway.
- */
+static int child_fatal;
-static void ap_start_shutdown(int graceful)
+static apr_status_t decrement_connection_count(void *cs_)
{
- mpm_state = AP_MPMQ_STOPPING;
- if (shutdown_pending == 1) {
- /* Um, is this _probably_ not an error, if the user has
- * tried to do a shutdown twice quickly, so we won't
- * worry about reporting it.
- */
- return;
+ event_conn_state_t *cs = cs_;
+ switch (cs->pub.state) {
+ case CONN_STATE_LINGER_NORMAL:
+ case CONN_STATE_LINGER_SHORT:
+ apr_atomic_dec32(&lingering_count);
+ break;
+ case CONN_STATE_SUSPENDED:
+ apr_atomic_dec32(&suspended_count);
+ break;
+ default:
+ break;
+ }
+ /* Unblock the listener if it's waiting for connection_count = 0 */
+ if (!apr_atomic_dec32(&connection_count)
+ && listener_is_wakeable && listener_may_exit) {
+ apr_pollset_wakeup(event_pollset);
}
- shutdown_pending = 1;
- is_graceful = graceful;
+ return APR_SUCCESS;
}
-/* do a graceful restart if graceful == 1 */
-static void ap_start_restart(int graceful)
+static void notify_suspend(event_conn_state_t *cs)
{
- mpm_state = AP_MPMQ_STOPPING;
- if (restart_pending == 1) {
- /* Probably not an error - don't bother reporting it */
- return;
+ ap_run_suspend_connection(cs->c, cs->r);
+ cs->suspended = 1;
+ cs->c->sbh = NULL;
+}
+
+static void notify_resume(event_conn_state_t *cs, ap_sb_handle_t *sbh)
+{
+ cs->c->sbh = sbh;
+ cs->suspended = 0;
+ ap_run_resume_connection(cs->c, cs->r);
+}
+
+/*
+ * Close our side of the connection, flushing data to the client first.
+ * Pre-condition: cs is not in any timeout queue and not in the pollset,
+ * timeout_mutex is not locked
+ * return: 0 if connection is fully closed,
+ * 1 if connection is lingering
+ * May only be called by worker thread.
+ */
+static int start_lingering_close_blocking(event_conn_state_t *cs)
+{
+ apr_status_t rv;
+ struct timeout_queue *q;
+ apr_socket_t *csd = cs->pfd.desc.s;
+
+ if (ap_start_lingering_close(cs->c)) {
+ notify_suspend(cs);
+ apr_socket_close(csd);
+ ap_push_pool(worker_queue_info, cs->p);
+ return 0;
}
- restart_pending = 1;
- is_graceful = graceful;
+
+#ifdef AP_DEBUG
+ {
+ rv = apr_socket_timeout_set(csd, 0);
+ AP_DEBUG_ASSERT(rv == APR_SUCCESS);
+ }
+#else
+ apr_socket_timeout_set(csd, 0);
+#endif
+
+ cs->queue_timestamp = apr_time_now();
+ /*
+ * If some module requested a shortened waiting period, only wait for
+ * 2s (SECONDS_TO_LINGER). This is useful for mitigating certain
+ * DoS attacks.
+ */
+ if (apr_table_get(cs->c->notes, "short-lingering-close")) {
+ q = short_linger_q;
+ cs->pub.state = CONN_STATE_LINGER_SHORT;
+ }
+ else {
+ q = linger_q;
+ cs->pub.state = CONN_STATE_LINGER_NORMAL;
+ }
+ apr_atomic_inc32(&lingering_count);
+ notify_suspend(cs);
+
+ cs->pfd.reqevents = (
+ cs->pub.sense == CONN_SENSE_WANT_WRITE ? APR_POLLOUT :
+ APR_POLLIN) | APR_POLLHUP | APR_POLLERR;
+ cs->pub.sense = CONN_SENSE_DEFAULT;
+ apr_thread_mutex_lock(timeout_mutex);
+ TO_QUEUE_APPEND(q, cs);
+ rv = apr_pollset_add(event_pollset, &cs->pfd);
+ if (rv != APR_SUCCESS && !APR_STATUS_IS_EEXIST(rv)) {
+ TO_QUEUE_REMOVE(q, cs);
+ apr_thread_mutex_unlock(timeout_mutex);
+ ap_log_error(APLOG_MARK, APLOG_ERR, rv, ap_server_conf, APLOGNO(03092)
+ "start_lingering_close: apr_pollset_add failure");
+ apr_socket_close(cs->pfd.desc.s);
+ ap_push_pool(worker_queue_info, cs->p);
+ return 0;
+ }
+ apr_thread_mutex_unlock(timeout_mutex);
+ return 1;
}
-static void sig_term(int sig)
+/*
+ * Defer flush and close of the connection by adding it to defer_linger_chain,
+ * for a worker to grab it and do the job (should that be blocking).
+ * Pre-condition: cs is not in any timeout queue and not in the pollset,
+ * timeout_mutex is not locked
+ * return: 1 connection is alive (but aside and about to linger)
+ * May be called by listener thread.
+ */
+static int start_lingering_close_nonblocking(event_conn_state_t *cs)
{
- ap_start_shutdown(sig == AP_SIG_GRACEFUL_STOP);
+ event_conn_state_t *chain;
+ for (;;) {
+ cs->chain = chain = defer_linger_chain;
+ if (apr_atomic_casptr((void *)&defer_linger_chain, cs,
+ chain) != chain) {
+ /* Race lost, try again */
+ continue;
+ }
+ return 1;
+ }
}
-static void restart(int sig)
+/*
+ * forcibly close a lingering connection after the lingering period has
+ * expired
+ * Pre-condition: cs is not in any timeout queue and not in the pollset
+ * return: irrelevant (need same prototype as start_lingering_close)
+ */
+static int stop_lingering_close(event_conn_state_t *cs)
{
- ap_start_restart(sig == AP_SIG_GRACEFUL);
+ apr_socket_t *csd = ap_get_conn_socket(cs->c);
+ ap_log_error(APLOG_MARK, APLOG_TRACE4, 0, ap_server_conf,
+ "socket reached timeout in lingering-close state");
+ abort_socket_nonblocking(csd);
+ ap_push_pool(worker_queue_info, cs->p);
+ if (dying)
+ ap_queue_interrupt_one(worker_queue);
+ return 0;
}
-static void set_signals(void)
+/*
+ * This runs before any non-MPM cleanup code on the connection;
+ * if the connection is currently suspended as far as modules
+ * know, provide notification of resumption.
+ */
+static apr_status_t ptrans_pre_cleanup(void *dummy)
{
-#ifndef NO_USE_SIGACTION
- struct sigaction sa;
-#endif
+ event_conn_state_t *cs = dummy;
- if (!one_process) {
- ap_fatal_signal_setup(ap_server_conf, pconf);
+ if (cs->suspended) {
+ notify_resume(cs, NULL);
}
+ return APR_SUCCESS;
+}
-#ifndef NO_USE_SIGACTION
- sigemptyset(&sa.sa_mask);
- sa.sa_flags = 0;
-
- sa.sa_handler = sig_term;
- if (sigaction(SIGTERM, &sa, NULL) < 0)
- ap_log_error(APLOG_MARK, APLOG_WARNING, errno, ap_server_conf,
- "sigaction(SIGTERM)");
-#ifdef AP_SIG_GRACEFUL_STOP
- if (sigaction(AP_SIG_GRACEFUL_STOP, &sa, NULL) < 0)
- ap_log_error(APLOG_MARK, APLOG_WARNING, errno, ap_server_conf,
- "sigaction(" AP_SIG_GRACEFUL_STOP_STRING ")");
-#endif
-#ifdef SIGINT
- if (sigaction(SIGINT, &sa, NULL) < 0)
- ap_log_error(APLOG_MARK, APLOG_WARNING, errno, ap_server_conf,
- "sigaction(SIGINT)");
-#endif
-#ifdef SIGXCPU
- sa.sa_handler = SIG_DFL;
- if (sigaction(SIGXCPU, &sa, NULL) < 0)
- ap_log_error(APLOG_MARK, APLOG_WARNING, errno, ap_server_conf,
- "sigaction(SIGXCPU)");
-#endif
-#ifdef SIGXFSZ
- sa.sa_handler = SIG_DFL;
- if (sigaction(SIGXFSZ, &sa, NULL) < 0)
- ap_log_error(APLOG_MARK, APLOG_WARNING, errno, ap_server_conf,
- "sigaction(SIGXFSZ)");
-#endif
-#ifdef SIGPIPE
- sa.sa_handler = SIG_IGN;
- if (sigaction(SIGPIPE, &sa, NULL) < 0)
- ap_log_error(APLOG_MARK, APLOG_WARNING, errno, ap_server_conf,
- "sigaction(SIGPIPE)");
-#endif
+/*
+ * event_pre_read_request() and event_request_cleanup() track the
+ * current r for a given connection.
+ */
+static apr_status_t event_request_cleanup(void *dummy)
+{
+ conn_rec *c = dummy;
+ event_conn_state_t *cs = ap_get_module_config(c->conn_config,
+ &mpm_event_module);
- /* we want to ignore HUPs and AP_SIG_GRACEFUL while we're busy
- * processing one */
- sigaddset(&sa.sa_mask, SIGHUP);
- sigaddset(&sa.sa_mask, AP_SIG_GRACEFUL);
- sa.sa_handler = restart;
- if (sigaction(SIGHUP, &sa, NULL) < 0)
- ap_log_error(APLOG_MARK, APLOG_WARNING, errno, ap_server_conf,
- "sigaction(SIGHUP)");
- if (sigaction(AP_SIG_GRACEFUL, &sa, NULL) < 0)
- ap_log_error(APLOG_MARK, APLOG_WARNING, errno, ap_server_conf,
- "sigaction(" AP_SIG_GRACEFUL_STRING ")");
-#else
- if (!one_process) {
-#ifdef SIGXCPU
- apr_signal(SIGXCPU, SIG_DFL);
-#endif /* SIGXCPU */
-#ifdef SIGXFSZ
- apr_signal(SIGXFSZ, SIG_DFL);
-#endif /* SIGXFSZ */
- }
-
- apr_signal(SIGTERM, sig_term);
-#ifdef SIGHUP
- apr_signal(SIGHUP, restart);
-#endif /* SIGHUP */
-#ifdef AP_SIG_GRACEFUL
- apr_signal(AP_SIG_GRACEFUL, restart);
-#endif /* AP_SIG_GRACEFUL */
-#ifdef AP_SIG_GRACEFUL_STOP
- apr_signal(AP_SIG_GRACEFUL_STOP, sig_term);
-#endif /* AP_SIG_GRACEFUL_STOP */
-#ifdef SIGPIPE
- apr_signal(SIGPIPE, SIG_IGN);
-#endif /* SIGPIPE */
+ cs->r = NULL;
+ return APR_SUCCESS;
+}
-#endif
+static void event_pre_read_request(request_rec *r, conn_rec *c)
+{
+ event_conn_state_t *cs = ap_get_module_config(c->conn_config,
+ &mpm_event_module);
+
+ cs->r = r;
+ cs->sc = ap_get_module_config(ap_server_conf->module_config,
+ &mpm_event_module);
+ apr_pool_cleanup_register(r->pool, c, event_request_cleanup,
+ apr_pool_cleanup_null);
}
-/*****************************************************************
- * Child process main loop.
+/*
+ * event_post_read_request() tracks the current server config for a
+ * given request.
*/
+static int event_post_read_request(request_rec *r)
+{
+ conn_rec *c = r->connection;
+ event_conn_state_t *cs = ap_get_module_config(c->conn_config,
+ &mpm_event_module);
+
+ /* To preserve legacy behaviour (consistent with other MPMs), use
+ * the keepalive timeout from the base server (first on this IP:port)
+ * when none is explicitly configured on this server.
+ */
+ if (r->server->keep_alive_timeout_set) {
+ cs->sc = ap_get_module_config(r->server->module_config,
+ &mpm_event_module);
+ }
+ else {
+ cs->sc = ap_get_module_config(c->base_server->module_config,
+ &mpm_event_module);
+ }
+ return OK;
+}
-static int process_socket(apr_thread_t *thd, apr_pool_t * p, apr_socket_t * sock,
- conn_state_t * cs, int my_child_num,
+/*
+ * process one connection in the worker
+ */
+static void process_socket(apr_thread_t *thd, apr_pool_t * p, apr_socket_t * sock,
+ event_conn_state_t * cs, int my_child_num,
int my_thread_num)
{
conn_rec *c;
- listener_poll_type *pt;
long conn_id = ID_FROM_CHILD_THREAD(my_child_num, my_thread_num);
int rc;
ap_sb_handle_t *sbh;
+ /* XXX: This will cause unbounded mem usage for long lasting connections */
ap_create_sb_handle(&sbh, p, my_child_num, my_thread_num);
if (cs == NULL) { /* This is a new connection */
-
- cs = apr_pcalloc(p, sizeof(conn_state_t));
-
- pt = apr_pcalloc(p, sizeof(*pt));
-
+ listener_poll_type *pt = apr_pcalloc(p, sizeof(*pt));
+ cs = apr_pcalloc(p, sizeof(event_conn_state_t));
cs->bucket_alloc = apr_bucket_alloc_create(p);
c = ap_run_create_connection(p, ap_server_conf, sock,
conn_id, sbh, cs->bucket_alloc);
+ if (!c) {
+ ap_push_pool(worker_queue_info, p);
+ return;
+ }
+ apr_atomic_inc32(&connection_count);
+ apr_pool_cleanup_register(c->pool, cs, decrement_connection_count,
+ apr_pool_cleanup_null);
+ ap_set_module_config(c->conn_config, &mpm_event_module, cs);
c->current_thread = thd;
cs->c = c;
- c->cs = cs;
+ c->cs = &(cs->pub);
cs->p = p;
+ cs->sc = ap_get_module_config(ap_server_conf->module_config,
+ &mpm_event_module);
cs->pfd.desc_type = APR_POLL_SOCKET;
cs->pfd.reqevents = APR_POLLIN;
cs->pfd.desc.s = sock;
pt->type = PT_CSD;
- pt->bypass_push = 1;
pt->baton = cs;
cs->pfd.client_data = pt;
- APR_RING_ELEM_INIT(cs, timeout_list);
+ apr_pool_pre_cleanup_register(p, cs, ptrans_pre_cleanup);
+ TO_QUEUE_ELEM_INIT(cs);
ap_update_vhost_given_ip(c);
rc = ap_run_pre_connection(c, sock);
if (rc != OK && rc != DONE) {
- ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, ap_server_conf,
- "process_socket: connection aborted");
+ ap_log_cerror(APLOG_MARK, APLOG_DEBUG, 0, c, APLOGNO(00469)
+ "process_socket: connection aborted");
c->aborted = 1;
}
* When the accept filter is active, sockets are kept in the
* kernel until a HTTP request is received.
*/
- cs->state = CONN_STATE_READ_REQUEST_LINE;
+ cs->pub.state = CONN_STATE_READ_REQUEST_LINE;
+ cs->pub.sense = CONN_SENSE_DEFAULT;
}
else {
c = cs->c;
- c->sbh = sbh;
- pt = cs->pfd.client_data;
+ notify_resume(cs, sbh);
c->current_thread = thd;
+ /* Subsequent request on a conn, and thread number is part of ID */
+ c->id = conn_id;
+
+ if (c->aborted) {
+ cs->pub.state = CONN_STATE_LINGER;
+ }
}
- if (c->clogging_input_filters && !c->aborted) {
- /* Since we have an input filter which 'cloggs' the input stream,
- * like mod_ssl, lets just do the normal read from input filters,
- * like the Worker MPM does.
+ if (cs->pub.state == CONN_STATE_LINGER) {
+ /* do lingering close below */
+ }
+ else if (c->clogging_input_filters) {
+ /* Since we have an input filter which 'clogs' the input stream,
+ * like mod_ssl used to, lets just do the normal read from input
+ * filters, like the Worker MPM does. Filters that need to write
+ * where they would otherwise read, or read where they would
+ * otherwise write, should set the sense appropriately.
*/
+ apr_atomic_inc32(&clogged_count);
ap_run_process_connection(c);
- if (cs->state != CONN_STATE_SUSPENDED) {
- cs->state = CONN_STATE_LINGER;
+ if (cs->pub.state != CONN_STATE_SUSPENDED) {
+ cs->pub.state = CONN_STATE_LINGER;
}
+ apr_atomic_dec32(&clogged_count);
}
-
+ else if (cs->pub.state == CONN_STATE_READ_REQUEST_LINE) {
read_request:
- if (cs->state == CONN_STATE_READ_REQUEST_LINE) {
- if (!c->aborted) {
- ap_run_process_connection(c);
+ ap_run_process_connection(c);
- /* state will be updated upon return
- * fall thru to either wait for readability/timeout or
- * do lingering close
- */
- }
- else {
- cs->state = CONN_STATE_LINGER;
- }
+ /* state will be updated upon return
+ * fall thru to either wait for readability/timeout or
+ * do lingering close
+ */
}
- if (cs->state == CONN_STATE_WRITE_COMPLETION) {
+ if (cs->pub.state == CONN_STATE_WRITE_COMPLETION) {
ap_filter_t *output_filter = c->output_filters;
apr_status_t rv;
+ ap_update_child_status(sbh, SERVER_BUSY_WRITE, NULL);
while (output_filter->next != NULL) {
output_filter = output_filter->next;
}
rv = output_filter->frec->filter_func.out_func(output_filter, NULL);
if (rv != APR_SUCCESS) {
- ap_log_error(APLOG_MARK, APLOG_WARNING, rv, ap_server_conf,
- "network write failure in core output filter");
- cs->state = CONN_STATE_LINGER;
+ ap_log_cerror(APLOG_MARK, APLOG_DEBUG, rv, c, APLOGNO(00470)
+ "network write failure in core output filter");
+ cs->pub.state = CONN_STATE_LINGER;
}
else if (c->data_in_output_filters) {
/* Still in WRITE_COMPLETION_STATE:
* Set a write timeout for this connection, and let the
* event thread poll for writeability.
*/
- cs->expiration_time = ap_server_conf->timeout + apr_time_now();
+ cs->queue_timestamp = apr_time_now();
+ notify_suspend(cs);
+ cs->pfd.reqevents = (
+ cs->pub.sense == CONN_SENSE_WANT_READ ? APR_POLLIN :
+ APR_POLLOUT) | APR_POLLHUP | APR_POLLERR;
+ cs->pub.sense = CONN_SENSE_DEFAULT;
apr_thread_mutex_lock(timeout_mutex);
- APR_RING_INSERT_TAIL(&timeout_head, cs, conn_state_t, timeout_list);
- apr_thread_mutex_unlock(timeout_mutex);
- pt->bypass_push = 0;
- cs->pfd.reqevents = APR_POLLOUT | APR_POLLHUP | APR_POLLERR;
+ TO_QUEUE_APPEND(cs->sc->wc_q, cs);
rc = apr_pollset_add(event_pollset, &cs->pfd);
- return 1;
+ if (rc != APR_SUCCESS && !APR_STATUS_IS_EEXIST(rc)) {
+ TO_QUEUE_REMOVE(cs->sc->wc_q, cs);
+ apr_thread_mutex_unlock(timeout_mutex);
+ ap_log_error(APLOG_MARK, APLOG_ERR, rc, ap_server_conf, APLOGNO(03465)
+ "process_socket: apr_pollset_add failure for "
+ "write completion");
+ apr_socket_close(cs->pfd.desc.s);
+ ap_push_pool(worker_queue_info, cs->p);
+ }
+ else {
+ apr_thread_mutex_unlock(timeout_mutex);
+ }
+ return;
}
else if (c->keepalive != AP_CONN_KEEPALIVE || c->aborted ||
- listener_may_exit) {
- c->cs->state = CONN_STATE_LINGER;
+ listener_may_exit) {
+ cs->pub.state = CONN_STATE_LINGER;
}
else if (c->data_in_input_filters) {
- cs->state = CONN_STATE_READ_REQUEST_LINE;
+ cs->pub.state = CONN_STATE_READ_REQUEST_LINE;
goto read_request;
}
else {
- cs->state = CONN_STATE_CHECK_REQUEST_LINE_READABLE;
+ cs->pub.state = CONN_STATE_CHECK_REQUEST_LINE_READABLE;
}
}
- if (cs->state == CONN_STATE_LINGER) {
- ap_lingering_close(c);
- apr_pool_clear(p);
- ap_push_pool(worker_queue_info, p);
- return 0;
+ if (cs->pub.state == CONN_STATE_LINGER) {
+ start_lingering_close_blocking(cs);
}
- else if (cs->state == CONN_STATE_CHECK_REQUEST_LINE_READABLE) {
- apr_status_t rc;
- listener_poll_type *pt = (listener_poll_type *) cs->pfd.client_data;
-
- /* It greatly simplifies the logic to use a single timeout value here
+ else if (cs->pub.state == CONN_STATE_CHECK_REQUEST_LINE_READABLE) {
+ /* It greatly simplifies the logic to use a single timeout value per q
* because the new element can just be added to the end of the list and
* it will stay sorted in expiration time sequence. If brand new
* sockets are sent to the event thread for a readability check, this
* timeout today. With a normal client, the socket will be readable in
* a few milliseconds anyway.
*/
- cs->expiration_time = ap_server_conf->keep_alive_timeout +
- apr_time_now();
- apr_thread_mutex_lock(timeout_mutex);
- APR_RING_INSERT_TAIL(&keepalive_timeout_head, cs, conn_state_t, timeout_list);
- apr_thread_mutex_unlock(timeout_mutex);
+ cs->queue_timestamp = apr_time_now();
+ notify_suspend(cs);
- pt->bypass_push = 0;
/* Add work to pollset. */
cs->pfd.reqevents = APR_POLLIN;
+ apr_thread_mutex_lock(timeout_mutex);
+ TO_QUEUE_APPEND(cs->sc->ka_q, cs);
rc = apr_pollset_add(event_pollset, &cs->pfd);
-
- if (rc != APR_SUCCESS) {
- ap_log_error(APLOG_MARK, APLOG_ERR, rc, ap_server_conf,
- "process_socket: apr_pollset_add failure");
- AP_DEBUG_ASSERT(rc == APR_SUCCESS);
+ if (rc != APR_SUCCESS && !APR_STATUS_IS_EEXIST(rc)) {
+ TO_QUEUE_REMOVE(cs->sc->ka_q, cs);
+ apr_thread_mutex_unlock(timeout_mutex);
+ ap_log_error(APLOG_MARK, APLOG_ERR, rc, ap_server_conf, APLOGNO(03093)
+ "process_socket: apr_pollset_add failure for "
+ "keep alive");
+ apr_socket_close(cs->pfd.desc.s);
+ ap_push_pool(worker_queue_info, cs->p);
+ return;
}
+ apr_thread_mutex_unlock(timeout_mutex);
+ }
+ else if (cs->pub.state == CONN_STATE_SUSPENDED) {
+ apr_atomic_inc32(&suspended_count);
+ notify_suspend(cs);
}
- return 1;
}
-/* requests_this_child has gone to zero or below. See if the admin coded
- "MaxRequestsPerChild 0", and keep going in that case. Doing it this way
+/* conns_this_child has gone to zero or below. See if the admin coded
+ "MaxConnectionsPerChild 0", and keep going in that case. Doing it this way
simplifies the hot path in worker_thread */
static void check_infinite_requests(void)
{
if (ap_max_requests_per_child) {
+ ap_log_error(APLOG_MARK, APLOG_TRACE1, 0, ap_server_conf,
+ "Stopping process due to MaxConnectionsPerChild");
signal_threads(ST_GRACEFUL);
}
else {
- requests_this_child = INT_MAX; /* keep going */
+ /* keep going */
+ conns_this_child = APR_INT32_MAX;
+ }
+}
+
+static void close_listeners(int process_slot, int *closed)
+{
+ if (!*closed) {
+ int i;
+ disable_listensocks(process_slot);
+ ap_close_listeners_ex(my_bucket->listeners);
+ *closed = 1;
+ dying = 1;
+ ap_scoreboard_image->parent[process_slot].quiescing = 1;
+ for (i = 0; i < threads_per_child; ++i) {
+ 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);
}
}
}
-#if HAVE_SERF
-static apr_status_t s_socket_add(void *user_baton,
- apr_pollfd_t *pfd,
- void *serf_baton)
-{
- s_baton_t *s = (s_baton_t*)user_baton;
- /* XXXXX: recycle listener_poll_types */
- listener_poll_type *pt = malloc(sizeof(*pt));
- pt->type = PT_SERF;
- pt->baton = serf_baton;
- pfd->client_data = pt;
- return apr_pollset_add(s->pollset, pfd);
-}
-
-static apr_status_t s_socket_remove(void *user_baton,
- apr_pollfd_t *pfd,
- void *serf_baton)
-{
- s_baton_t *s = (s_baton_t*)user_baton;
- listener_poll_type *pt = pfd->client_data;
- free(pt);
- return apr_pollset_remove(s->pollset, pfd);
-}
-#endif
-
static apr_status_t init_pollset(apr_pool_t *p)
{
-#if HAVE_SERF
- s_baton_t *baton = NULL;
-#endif
ap_listen_rec *lr;
listener_poll_type *pt;
+ int i = 0;
- APR_RING_INIT(&timeout_head, conn_state_t, timeout_list);
- APR_RING_INIT(&keepalive_timeout_head, conn_state_t, timeout_list);
-
- for (lr = ap_listeners; lr != NULL; lr = lr->next) {
- apr_pollfd_t *pfd = apr_palloc(p, sizeof(*pfd));
+ listener_pollfd = apr_palloc(p, sizeof(apr_pollfd_t) * num_listensocks);
+ for (lr = my_bucket->listeners; lr != NULL; lr = lr->next, i++) {
+ apr_pollfd_t *pfd;
+ AP_DEBUG_ASSERT(i < num_listensocks);
+ pfd = &listener_pollfd[i];
pt = apr_pcalloc(p, sizeof(*pt));
pfd->desc_type = APR_POLL_SOCKET;
pfd->desc.s = lr->sd;
lr->accept_func = ap_unixd_accept;
}
-#if HAVE_SERF
- baton = apr_pcalloc(p, sizeof(*baton));
- baton->pollset = event_pollset;
- /* TODO: subpools, threads, reuse, etc. -- currently use malloc() inside :( */
- baton->pool = p;
-
- g_serf = serf_context_create_ex(baton,
- s_socket_add,
- s_socket_remove, p);
-
- ap_register_provider(p, "mpm_serf",
- "instance", "0", g_serf);
-
-#endif
-
return APR_SUCCESS;
}
return ap_queue_push_timer(worker_queue, te);
}
-static apr_status_t push2worker(const apr_pollfd_t * pfd,
- apr_pollset_t * pollset)
+/*
+ * Pre-condition: cs is neither in event_pollset nor a timeout queue
+ * this function may only be called by the listener
+ */
+static apr_status_t push2worker(event_conn_state_t *cs, apr_socket_t *csd,
+ apr_pool_t *ptrans)
{
- listener_poll_type *pt = (listener_poll_type *) pfd->client_data;
- conn_state_t *cs = (conn_state_t *) pt->baton;
apr_status_t rc;
- if (pt->bypass_push) {
- return APR_SUCCESS;
+ if (cs) {
+ csd = cs->pfd.desc.s;
+ ptrans = cs->p;
}
-
- pt->bypass_push = 1;
-
- rc = apr_pollset_remove(pollset, pfd);
-
- /*
- * Some of the pollset backends, like KQueue or Epoll
- * automagically remove the FD if the socket is closed,
- * therefore, we can accept _SUCCESS or _NOTFOUND,
- * and we still want to keep going
- */
- if (rc != APR_SUCCESS && !APR_STATUS_IS_NOTFOUND(rc)) {
- cs->state = CONN_STATE_LINGER;
- }
-
- rc = ap_queue_push(worker_queue, cs->pfd.desc.s, cs, cs->p);
+ rc = ap_queue_push(worker_queue, csd, cs, ptrans);
if (rc != APR_SUCCESS) {
+ ap_log_error(APLOG_MARK, APLOG_CRIT, rc, ap_server_conf, APLOGNO(00471)
+ "push2worker: ap_queue_push failed");
/* trash the connection; we couldn't queue the connected
* socket to a worker
*/
- apr_bucket_alloc_destroy(cs->bucket_alloc);
- apr_socket_close(cs->pfd.desc.s);
- ap_log_error(APLOG_MARK, APLOG_CRIT, rc,
- ap_server_conf, "push2worker: ap_queue_push failed");
- apr_pool_clear(cs->p);
- ap_push_pool(worker_queue_info, cs->p);
+ if (csd) {
+ abort_socket_nonblocking(csd);
+ }
+ if (ptrans) {
+ ap_push_pool(worker_queue_info, ptrans);
+ }
+ signal_threads(ST_GRACEFUL);
}
return rc;
}
/* get_worker:
- * reserve a worker thread, block if all are currently busy.
- * this prevents the worker queue from overflowing and lets
- * other processes accept new connections in the mean time.
+ * If *have_idle_worker_p == 0, reserve a worker thread, and set
+ * *have_idle_worker_p = 1.
+ * If *have_idle_worker_p is already 1, will do nothing.
+ * If blocking == 1, block if all workers are currently busy.
+ * If no worker was available immediately, will set *all_busy to 1.
+ * XXX: If there are no workers, we should not block immediately but
+ * XXX: close all keep-alive connections first.
*/
-static int get_worker(int *have_idle_worker_p)
+static void get_worker(int *have_idle_worker_p, int blocking, int *all_busy)
{
apr_status_t rc;
- if (!*have_idle_worker_p) {
- rc = ap_queue_info_wait_for_idler(worker_queue_info);
-
- if (rc == APR_SUCCESS) {
- *have_idle_worker_p = 1;
- return 1;
- }
- else {
- if (!APR_STATUS_IS_EOF(rc)) {
- ap_log_error(APLOG_MARK, APLOG_ERR, rc, ap_server_conf,
- "ap_queue_info_wait_for_idler failed. "
- "Attempting to shutdown process gracefully");
- signal_threads(ST_GRACEFUL);
- }
- return 0;
- }
- }
- else {
+ if (*have_idle_worker_p) {
/* already reserved a worker thread - must have hit a
* transient error on a previous pass
*/
- return 1;
+ return;
}
-}
-/* XXXXXX: Convert to skiplist or other better data structure
- * (yes, this is VERY VERY VERY VERY BAD)
- */
+ if (blocking)
+ rc = ap_queue_info_wait_for_idler(worker_queue_info, all_busy);
+ else
+ rc = ap_queue_info_try_get_idler(worker_queue_info);
+
+ if (rc == APR_SUCCESS || APR_STATUS_IS_EOF(rc)) {
+ *have_idle_worker_p = 1;
+ }
+ else if (!blocking && rc == APR_EAGAIN) {
+ *all_busy = 1;
+ }
+ else {
+ ap_log_error(APLOG_MARK, APLOG_ERR, rc, ap_server_conf, APLOGNO(00472)
+ "ap_queue_info_wait_for_idler failed. "
+ "Attempting to shutdown process gracefully");
+ signal_threads(ST_GRACEFUL);
+ }
+}
/* Structures to reuse */
static APR_RING_HEAD(timer_free_ring_t, timer_event_t) timer_free_ring;
-/* Active timers */
-static APR_RING_HEAD(timer_ring_t, timer_event_t) timer_ring;
-static apr_thread_mutex_t *g_timer_ring_mtx;
+static apr_skiplist *timer_skiplist;
+static volatile apr_time_t timers_next_expiry;
+
+/* Same goal as for TIMEOUT_FUDGE_FACTOR (avoid extra poll calls), but applied
+ * to timers. Since their timeouts are custom (user defined), we can't be too
+ * approximative here (hence using 0.01s).
+ */
+#define EVENT_FUDGE_FACTOR apr_time_from_msec(10)
+
+/* The following compare function is used by apr_skiplist_insert() to keep the
+ * elements (timers) sorted and provide O(log n) complexity (this is also true
+ * for apr_skiplist_{find,remove}(), but those are not used in MPM event where
+ * inserted timers are not searched nor removed, but with apr_skiplist_pop()
+ * which does use any compare function). It is meant to return 0 when a == b,
+ * <0 when a < b, and >0 when a > b. However apr_skiplist_insert() will not
+ * add duplicates (i.e. a == b), and apr_skiplist_add() is only available in
+ * APR 1.6, yet multiple timers could possibly be created in the same micro-
+ * second (duplicates with regard to apr_time_t); therefore we implement the
+ * compare function to return +1 instead of 0 when compared timers are equal,
+ * thus duplicates are still added after each other (in order of insertion).
+ */
+static int timer_comp(void *a, void *b)
+{
+ apr_time_t t1 = (apr_time_t) ((timer_event_t *)a)->when;
+ apr_time_t t2 = (apr_time_t) ((timer_event_t *)b)->when;
+ AP_DEBUG_ASSERT(t1);
+ AP_DEBUG_ASSERT(t2);
+ return ((t1 < t2) ? -1 : 1);
+}
+
+static apr_thread_mutex_t *g_timer_skiplist_mtx;
static apr_status_t event_register_timed_callback(apr_time_t t,
ap_mpm_callback_fn_t *cbfn,
void *baton)
{
- int inserted = 0;
- timer_event_t *ep;
timer_event_t *te;
/* oh yeah, and make locking smarter/fine grained. */
- apr_thread_mutex_lock(g_timer_ring_mtx);
+ apr_thread_mutex_lock(g_timer_skiplist_mtx);
if (!APR_RING_EMPTY(&timer_free_ring, timer_event_t, link)) {
te = APR_RING_FIRST(&timer_free_ring);
APR_RING_REMOVE(te, link);
}
else {
- /* XXXXX: lol, pool allocation without a context from any thread.Yeah. Right. MPMs Suck. */
- te = malloc(sizeof(timer_event_t));
+ te = apr_skiplist_alloc(timer_skiplist, sizeof(timer_event_t));
APR_RING_ELEM_INIT(te, link);
}
/* XXXXX: optimize */
te->when = t + apr_time_now();
- /* Okay, insert sorted by when.. */
- for (ep = APR_RING_FIRST(&timer_ring);
- ep != APR_RING_SENTINEL(&timer_ring,
- timer_event_t, link);
- ep = APR_RING_NEXT(ep, link))
- {
- if (ep->when > te->when) {
- inserted = 1;
- APR_RING_INSERT_BEFORE(ep, te, link);
- break;
+ {
+ apr_time_t next_expiry;
+
+ /* Okay, add sorted by when.. */
+ apr_skiplist_insert(timer_skiplist, te);
+
+ /* Cheaply update the overall timers' next expiry according to
+ * this event, if necessary.
+ */
+ next_expiry = timers_next_expiry;
+ if (!next_expiry || next_expiry > te->when + EVENT_FUDGE_FACTOR) {
+ timers_next_expiry = te->when;
+ /* Unblock the poll()ing listener for it to update its timeout. */
+ if (listener_is_wakeable) {
+ apr_pollset_wakeup(event_pollset);
+ }
}
}
-
- if (!inserted) {
- APR_RING_INSERT_TAIL(&timer_ring, te, timer_event_t, link);
- }
- apr_thread_mutex_unlock(g_timer_ring_mtx);
+ apr_thread_mutex_unlock(g_timer_skiplist_mtx);
return APR_SUCCESS;
}
+
+/*
+ * Close socket and clean up if remote closed its end while we were in
+ * lingering close.
+ * Only to be called in the listener thread;
+ * Pre-condition: cs is in one of the linger queues and in the pollset
+ */
+static void process_lingering_close(event_conn_state_t *cs, const apr_pollfd_t *pfd)
+{
+ apr_socket_t *csd = ap_get_conn_socket(cs->c);
+ char dummybuf[2048];
+ apr_size_t nbytes;
+ apr_status_t rv;
+ struct timeout_queue *q;
+ q = (cs->pub.state == CONN_STATE_LINGER_SHORT) ? short_linger_q : linger_q;
+
+ /* socket is already in non-blocking state */
+ do {
+ nbytes = sizeof(dummybuf);
+ rv = apr_socket_recv(csd, dummybuf, &nbytes);
+ } while (rv == APR_SUCCESS);
+
+ if (APR_STATUS_IS_EAGAIN(rv)) {
+ return;
+ }
+
+ apr_thread_mutex_lock(timeout_mutex);
+ TO_QUEUE_REMOVE(q, cs);
+ rv = apr_pollset_remove(event_pollset, pfd);
+ apr_thread_mutex_unlock(timeout_mutex);
+ AP_DEBUG_ASSERT(rv == APR_SUCCESS || APR_STATUS_IS_NOTFOUND(rv));
+
+ rv = apr_socket_close(csd);
+ AP_DEBUG_ASSERT(rv == APR_SUCCESS);
+
+ ap_push_pool(worker_queue_info, cs->p);
+ if (dying)
+ ap_queue_interrupt_one(worker_queue);
+}
+
+/* call 'func' for all elements of 'q' with timeout less than 'timeout_time'.
+ * Pre-condition: timeout_mutex must already be locked
+ * Post-condition: timeout_mutex will be locked again
+ */
+static void process_timeout_queue(struct timeout_queue *q,
+ apr_time_t timeout_time,
+ int (*func)(event_conn_state_t *))
+{
+ apr_uint32_t total = 0, count;
+ event_conn_state_t *first, *cs, *last;
+ struct timeout_head_t trash;
+ struct timeout_queue *qp;
+ apr_status_t rv;
+
+ if (!apr_atomic_read32(q->total)) {
+ return;
+ }
+
+ APR_RING_INIT(&trash, event_conn_state_t, timeout_list);
+ for (qp = q; qp; qp = qp->next) {
+ count = 0;
+ cs = first = last = APR_RING_FIRST(&qp->head);
+ while (cs != APR_RING_SENTINEL(&qp->head, event_conn_state_t,
+ timeout_list)) {
+ /* Trash the entry if:
+ * - no timeout_time was given (asked for all), or
+ * - it expired (according to the queue timeout), or
+ * - the system clock skewed in the past: no entry should be
+ * registered above the given timeout_time (~now) + the queue
+ * timeout, we won't keep any here (eg. for centuries).
+ *
+ * Otherwise stop, no following entry will match thanks to the
+ * single timeout per queue (entries are added to the end!).
+ * This allows maintenance in O(1).
+ */
+ if (timeout_time
+ && cs->queue_timestamp + qp->timeout > timeout_time
+ && cs->queue_timestamp < timeout_time + qp->timeout) {
+ /* Since this is the next expiring of this queue, update the
+ * overall queues' next expiry if it's later than this one.
+ */
+ apr_time_t q_expiry = cs->queue_timestamp + qp->timeout;
+ apr_time_t next_expiry = queues_next_expiry;
+ if (!next_expiry || next_expiry > q_expiry) {
+ queues_next_expiry = q_expiry;
+ }
+ break;
+ }
+
+ last = cs;
+ rv = apr_pollset_remove(event_pollset, &cs->pfd);
+ if (rv != APR_SUCCESS && !APR_STATUS_IS_NOTFOUND(rv)) {
+ ap_log_cerror(APLOG_MARK, APLOG_ERR, rv, cs->c, APLOGNO(00473)
+ "apr_pollset_remove failed");
+ }
+ cs = APR_RING_NEXT(cs, timeout_list);
+ count++;
+ }
+ if (!count)
+ continue;
+
+ APR_RING_UNSPLICE(first, last, timeout_list);
+ APR_RING_SPLICE_TAIL(&trash, first, last, event_conn_state_t,
+ timeout_list);
+ AP_DEBUG_ASSERT(apr_atomic_read32(q->total) >= count);
+ apr_atomic_sub32(q->total, count);
+ qp->count -= count;
+ total += count;
+ }
+ if (!total)
+ return;
+
+ apr_thread_mutex_unlock(timeout_mutex);
+ first = APR_RING_FIRST(&trash);
+ do {
+ cs = APR_RING_NEXT(first, timeout_list);
+ TO_QUEUE_ELEM_INIT(first);
+ func(first);
+ first = cs;
+ } while (--total);
+ apr_thread_mutex_lock(timeout_mutex);
+}
+
+static void process_keepalive_queue(apr_time_t timeout_time)
+{
+ /* If all workers are busy, we kill older keep-alive connections so
+ * that they may connect to another process.
+ */
+ if (!timeout_time) {
+ ap_log_error(APLOG_MARK, APLOG_TRACE1, 0, ap_server_conf,
+ "All workers are busy or dying, will close %u "
+ "keep-alive connections",
+ apr_atomic_read32(keepalive_q->total));
+ }
+ process_timeout_queue(keepalive_q, timeout_time,
+ start_lingering_close_nonblocking);
+}
+
static void * APR_THREAD_FUNC listener_thread(apr_thread_t * thd, void *dummy)
{
- timer_event_t *ep;
timer_event_t *te;
apr_status_t rc;
proc_info *ti = dummy;
- int process_slot = ti->pid;
+ int process_slot = ti->pslot;
+ struct process_score *ps = ap_get_scoreboard_process(process_slot);
apr_pool_t *tpool = apr_thread_pool_get(thd);
void *csd = NULL;
apr_pool_t *ptrans; /* Pool for per-transaction stuff */
ap_listen_rec *lr;
int have_idle_worker = 0;
- conn_state_t *cs;
const apr_pollfd_t *out_pfd;
apr_int32_t num = 0;
- apr_time_t time_now = 0;
apr_interval_time_t timeout_interval;
- apr_time_t timeout_time;
+ apr_time_t timeout_time = 0, now, last_log;
listener_poll_type *pt;
+ int closed = 0, listeners_disabled = 0;
+ last_log = apr_time_now();
free(ti);
- /* the following times out events that are really close in the future
- * to prevent extra poll calls
- *
- * current value is .1 second
- */
-#define TIMEOUT_FUDGE_FACTOR 100000
-#define EVENT_FUDGE_FACTOR 10000
-
rc = init_pollset(tpool);
if (rc != APR_SUCCESS) {
ap_log_error(APLOG_MARK, APLOG_ERR, rc, ap_server_conf,
unblock_signal(LISTENER_SIGNAL);
apr_signal(LISTENER_SIGNAL, dummy_signal_handler);
- while (!listener_may_exit) {
+ for (;;) {
+ int workers_were_busy = 0;
- if (requests_this_child <= 0) {
- check_infinite_requests();
+ if (listener_may_exit) {
+ close_listeners(process_slot, &closed);
+ if (terminate_mode == ST_UNGRACEFUL
+ || apr_atomic_read32(&connection_count) == 0)
+ break;
}
+ if (conns_this_child <= 0)
+ check_infinite_requests();
- {
- apr_time_t now = apr_time_now();
- apr_thread_mutex_lock(g_timer_ring_mtx);
+ now = apr_time_now();
+ if (APLOGtrace6(ap_server_conf)) {
+ /* trace log status every second */
+ if (now - last_log > apr_time_from_sec(1)) {
+ last_log = now;
+ apr_thread_mutex_lock(timeout_mutex);
+ ap_log_error(APLOG_MARK, APLOG_TRACE6, 0, ap_server_conf,
+ "connections: %u (clogged: %u write-completion: %d "
+ "keep-alive: %d lingering: %d suspended: %u)",
+ apr_atomic_read32(&connection_count),
+ apr_atomic_read32(&clogged_count),
+ apr_atomic_read32(write_completion_q->total),
+ apr_atomic_read32(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);
+ }
+ }
+
+ /* Start with an infinite poll() timeout and update it according to
+ * the next expiring timer or queue entry. If there are none, either
+ * the listener is wakeable and it can poll() indefinitely until a wake
+ * up occurs, otherwise periodic checks (maintenance, shutdown, ...)
+ * must be performed.
+ */
+ timeout_interval = -1;
- if (!APR_RING_EMPTY(&timer_ring, timer_event_t, link)) {
- te = APR_RING_FIRST(&timer_ring);
- if (te->when > now) {
+ /* Push expired timers to a worker, the first remaining one determines
+ * the maximum time to poll() below, if any.
+ */
+ timeout_time = timers_next_expiry;
+ if (timeout_time && timeout_time < now + EVENT_FUDGE_FACTOR) {
+ apr_thread_mutex_lock(g_timer_skiplist_mtx);
+ while ((te = apr_skiplist_peek(timer_skiplist))) {
+ if (te->when > now + EVENT_FUDGE_FACTOR) {
+ timers_next_expiry = te->when;
timeout_interval = te->when - now;
+ break;
}
- else {
- timeout_interval = 1;
- }
+ apr_skiplist_pop(timer_skiplist, NULL);
+ push_timer2worker(te);
}
- else {
- timeout_interval = apr_time_from_msec(100);
+ if (!te) {
+ timers_next_expiry = 0;
}
- apr_thread_mutex_unlock(g_timer_ring_mtx);
+ apr_thread_mutex_unlock(g_timer_skiplist_mtx);
}
-#if HAVE_SERF
- rc = serf_context_prerun(g_serf);
- if (rc != APR_SUCCESS) {
- /* TOOD: what should do here? ugh. */
+ /* Same for queues, use their next expiry, if any. */
+ timeout_time = queues_next_expiry;
+ if (timeout_time
+ && (timeout_interval < 0
+ || timeout_time <= now
+ || timeout_interval > timeout_time - now)) {
+ timeout_interval = timeout_time > now ? timeout_time - now : 1;
}
-
-#endif
- rc = apr_pollset_poll(event_pollset, timeout_interval, &num,
- &out_pfd);
+ /* When non-wakeable, don't wait more than 100 ms, in any case. */
+#define NON_WAKEABLE_POLL_TIMEOUT apr_time_from_msec(100)
+ if (!listener_is_wakeable
+ && (timeout_interval < 0
+ || timeout_interval > NON_WAKEABLE_POLL_TIMEOUT)) {
+ timeout_interval = NON_WAKEABLE_POLL_TIMEOUT;
+ }
+
+ rc = apr_pollset_poll(event_pollset, timeout_interval, &num, &out_pfd);
if (rc != APR_SUCCESS) {
if (APR_STATUS_IS_EINTR(rc)) {
- continue;
+ /* Woken up, if we are exiting we must fall through to kill
+ * kept-alive connections, otherwise we only need to update
+ * timeouts (logic is above, so restart the loop).
+ */
+ if (!listener_may_exit) {
+ continue;
+ }
+ timeout_time = 0;
}
- if (!APR_STATUS_IS_TIMEUP(rc)) {
- ap_log_error(APLOG_MARK, APLOG_ERR, rc, ap_server_conf,
+ else if (!APR_STATUS_IS_TIMEUP(rc)) {
+ ap_log_error(APLOG_MARK, APLOG_CRIT, rc, ap_server_conf,
"apr_pollset_poll failed. Attempting to "
"shutdown process gracefully");
signal_threads(ST_GRACEFUL);
}
+ num = 0;
}
- if (listener_may_exit)
- break;
-
- {
- apr_time_t now = apr_time_now();
- apr_thread_mutex_lock(g_timer_ring_mtx);
- for (ep = APR_RING_FIRST(&timer_ring);
- ep != APR_RING_SENTINEL(&timer_ring,
- timer_event_t, link);
- ep = APR_RING_FIRST(&timer_ring))
- {
- if (ep->when < now + EVENT_FUDGE_FACTOR) {
- APR_RING_REMOVE(ep, link);
- push_timer2worker(ep);
- }
- else {
- break;
- }
- }
- apr_thread_mutex_unlock(g_timer_ring_mtx);
+ if (listener_may_exit) {
+ close_listeners(process_slot, &closed);
+ if (terminate_mode == ST_UNGRACEFUL
+ || apr_atomic_read32(&connection_count) == 0)
+ break;
}
- while (num && get_worker(&have_idle_worker)) {
+ while (num) {
pt = (listener_poll_type *) out_pfd->client_data;
if (pt->type == PT_CSD) {
/* one of the sockets is readable */
- cs = (conn_state_t *) pt->baton;
- switch (cs->state) {
+ event_conn_state_t *cs = (event_conn_state_t *) pt->baton;
+ struct timeout_queue *remove_from_q = cs->sc->wc_q;
+ int blocking = 1;
+
+ switch (cs->pub.state) {
case CONN_STATE_CHECK_REQUEST_LINE_READABLE:
- cs->state = CONN_STATE_READ_REQUEST_LINE;
- break;
+ cs->pub.state = CONN_STATE_READ_REQUEST_LINE;
+ remove_from_q = cs->sc->ka_q;
+ /* don't wait for a worker for a keepalive request */
+ blocking = 0;
+ /* FALL THROUGH */
case CONN_STATE_WRITE_COMPLETION:
+ get_worker(&have_idle_worker, blocking,
+ &workers_were_busy);
+ apr_thread_mutex_lock(timeout_mutex);
+ TO_QUEUE_REMOVE(remove_from_q, cs);
+ rc = apr_pollset_remove(event_pollset, &cs->pfd);
+ apr_thread_mutex_unlock(timeout_mutex);
+
+ /*
+ * Some of the pollset backends, like KQueue or Epoll
+ * automagically remove the FD if the socket is closed,
+ * therefore, we can accept _SUCCESS or _NOTFOUND,
+ * and we still want to keep going
+ */
+ if (rc != APR_SUCCESS && !APR_STATUS_IS_NOTFOUND(rc)) {
+ ap_log_error(APLOG_MARK, APLOG_ERR, rc, ap_server_conf,
+ APLOGNO(03094) "pollset remove failed");
+ start_lingering_close_nonblocking(cs);
+ break;
+ }
+
+ /* If we didn't get a worker immediately for a keep-alive
+ * request, we close the connection, so that the client can
+ * re-connect to a different process.
+ */
+ if (!have_idle_worker) {
+ start_lingering_close_nonblocking(cs);
+ }
+ else if (push2worker(cs, NULL, NULL) == APR_SUCCESS) {
+ have_idle_worker = 0;
+ }
break;
- default:
- ap_log_error(APLOG_MARK, APLOG_ERR, rc,
- ap_server_conf,
- "event_loop: unexpected state %d",
- cs->state);
- AP_DEBUG_ASSERT(0);
- }
- apr_thread_mutex_lock(timeout_mutex);
- APR_RING_REMOVE(cs, timeout_list);
- apr_thread_mutex_unlock(timeout_mutex);
- APR_RING_ELEM_INIT(cs, timeout_list);
+ case CONN_STATE_LINGER_NORMAL:
+ case CONN_STATE_LINGER_SHORT:
+ process_lingering_close(cs, out_pfd);
+ break;
- rc = push2worker(out_pfd, event_pollset);
- if (rc != APR_SUCCESS) {
+ default:
ap_log_error(APLOG_MARK, APLOG_CRIT, rc,
- ap_server_conf, "push2worker failed");
- }
- else {
- have_idle_worker = 0;
+ ap_server_conf, APLOGNO(03096)
+ "event_loop: unexpected state %d",
+ cs->pub.state);
+ ap_assert(0);
}
}
else if (pt->type == PT_ACCEPT) {
/* A Listener Socket is ready for an accept() */
+ if (workers_were_busy) {
+ if (!listeners_disabled)
+ disable_listensocks(process_slot);
+ listeners_disabled = 1;
+ ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, ap_server_conf,
+ "All workers busy, not accepting new conns "
+ "in this process");
+ }
+ else if ( (int)apr_atomic_read32(&connection_count)
+ - (int)apr_atomic_read32(&lingering_count)
+ > threads_per_child
+ + ap_queue_info_get_idlers(worker_queue_info) *
+ worker_factor / WORKER_FACTOR_SCALE)
+ {
+ if (!listeners_disabled)
+ disable_listensocks(process_slot);
+ ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, ap_server_conf,
+ "Too many open connections (%u), "
+ "not accepting new conns in this process",
+ apr_atomic_read32(&connection_count));
+ ap_log_error(APLOG_MARK, APLOG_TRACE1, 0, ap_server_conf,
+ "Idle workers: %u",
+ ap_queue_info_get_idlers(worker_queue_info));
+ listeners_disabled = 1;
+ }
+ else if (listeners_disabled) {
+ listeners_disabled = 0;
+ enable_listensocks(process_slot);
+ }
+ if (!listeners_disabled) {
+ lr = (ap_listen_rec *) pt->baton;
+ ap_pop_pool(&ptrans, worker_queue_info);
- lr = (ap_listen_rec *) pt->baton;
-
- ap_pop_pool(&ptrans, worker_queue_info);
-
- if (ptrans == NULL) {
- /* create a new transaction pool for each accepted socket */
- apr_allocator_t *allocator;
-
- apr_allocator_create(&allocator);
- apr_allocator_max_free_set(allocator,
- ap_max_mem_free);
- apr_pool_create_ex(&ptrans, pconf, NULL, allocator);
- apr_allocator_owner_set(allocator, ptrans);
if (ptrans == NULL) {
- ap_log_error(APLOG_MARK, APLOG_CRIT, rc,
- ap_server_conf,
- "Failed to create transaction pool");
- signal_threads(ST_GRACEFUL);
- return NULL;
+ /* create a new transaction pool for each accepted socket */
+ apr_allocator_t *allocator;
+
+ apr_allocator_create(&allocator);
+ apr_allocator_max_free_set(allocator,
+ ap_max_mem_free);
+ apr_pool_create_ex(&ptrans, pconf, NULL, allocator);
+ apr_allocator_owner_set(allocator, ptrans);
+ if (ptrans == NULL) {
+ ap_log_error(APLOG_MARK, APLOG_CRIT, rc,
+ ap_server_conf, APLOGNO(03097)
+ "Failed to create transaction pool");
+ signal_threads(ST_GRACEFUL);
+ return NULL;
+ }
}
- }
- apr_pool_tag(ptrans, "transaction");
+ apr_pool_tag(ptrans, "transaction");
- rc = lr->accept_func(&csd, lr, ptrans);
+ get_worker(&have_idle_worker, 1, &workers_were_busy);
+ rc = lr->accept_func(&csd, lr, ptrans);
- /* later we trash rv and rely on csd to indicate
- * success/failure
- */
- AP_DEBUG_ASSERT(rc == APR_SUCCESS || !csd);
+ /* later we trash rv and rely on csd to indicate
+ * success/failure
+ */
+ AP_DEBUG_ASSERT(rc == APR_SUCCESS || !csd);
- if (rc == APR_EGENERAL) {
- /* E[NM]FILE, ENOMEM, etc */
- resource_shortage = 1;
- signal_threads(ST_GRACEFUL);
- }
+ if (rc == APR_EGENERAL) {
+ /* E[NM]FILE, ENOMEM, etc */
+ resource_shortage = 1;
+ signal_threads(ST_GRACEFUL);
+ }
- if (csd != NULL) {
- rc = ap_queue_push(worker_queue, csd, NULL, ptrans);
- if (rc != APR_SUCCESS) {
- /* trash the connection; we couldn't queue the connected
- * socket to a worker
- */
- apr_socket_close(csd);
- ap_log_error(APLOG_MARK, APLOG_CRIT, rc,
- ap_server_conf,
- "ap_queue_push failed");
- apr_pool_clear(ptrans);
- ap_push_pool(worker_queue_info, ptrans);
+ if (csd != NULL) {
+ conns_this_child--;
+ if (push2worker(NULL, csd, ptrans) == APR_SUCCESS) {
+ have_idle_worker = 0;
+ }
}
else {
- have_idle_worker = 0;
+ ap_push_pool(worker_queue_info, ptrans);
}
}
- else {
- apr_pool_clear(ptrans);
- ap_push_pool(worker_queue_info, ptrans);
- }
} /* if:else on pt->type */
-#if HAVE_SERF
- else if (pt->type == PT_SERF) {
- /* send socket to serf. */
- /* XXXX: this doesn't require get_worker(&have_idle_worker) */
- serf_event_trigger(g_serf, pt->baton, out_pfd);
- }
-#endif
out_pfd++;
num--;
} /* while for processing poll */
/* XXX possible optimization: stash the current time for use as
* r->request_time for new requests
*/
- time_now = apr_time_now();
-
- /* handle timed out sockets */
- apr_thread_mutex_lock(timeout_mutex);
-
- /* Step 1: keepalive timeouts */
- cs = APR_RING_FIRST(&keepalive_timeout_head);
- timeout_time = time_now + TIMEOUT_FUDGE_FACTOR;
- while (!APR_RING_EMPTY(&keepalive_timeout_head, conn_state_t, timeout_list)
- && cs->expiration_time < timeout_time) {
+ /* We process the timeout queues here only when their overall next
+ * expiry (read once above) is over. This happens accurately since
+ * adding to the queues (in workers) can only decrease this expiry,
+ * while latest ones are only taken into account here (in listener)
+ * during queues' processing, with the lock held. This works both
+ * with and without wake-ability.
+ */
+ if (timeout_time && timeout_time < (now = apr_time_now())) {
+ timeout_time = now + TIMEOUT_FUDGE_FACTOR;
- cs->state = CONN_STATE_LINGER;
+ /* handle timed out sockets */
+ apr_thread_mutex_lock(timeout_mutex);
- APR_RING_REMOVE(cs, timeout_list);
- apr_thread_mutex_unlock(timeout_mutex);
+ /* Processing all the queues below will recompute this. */
+ queues_next_expiry = 0;
- if (!get_worker(&have_idle_worker)) {
- apr_thread_mutex_lock(timeout_mutex);
- APR_RING_INSERT_HEAD(&keepalive_timeout_head, cs,
- conn_state_t, timeout_list);
- break;
+ /* Step 1: keepalive timeouts */
+ if (workers_were_busy || dying) {
+ process_keepalive_queue(0); /* kill'em all \m/ */
}
-
- rc = push2worker(&cs->pfd, event_pollset);
-
- if (rc != APR_SUCCESS) {
- return NULL;
- /* XXX return NULL looks wrong - not an init failure
- * that bypasses all the cleanup outside the main loop
- * break seems more like it
- * need to evaluate seriousness of push2worker failures
- */
+ else {
+ process_keepalive_queue(timeout_time);
}
- have_idle_worker = 0;
- apr_thread_mutex_lock(timeout_mutex);
- cs = APR_RING_FIRST(&keepalive_timeout_head);
- }
-
- /* Step 2: write completion timeouts */
- cs = APR_RING_FIRST(&timeout_head);
- while (!APR_RING_EMPTY(&timeout_head, conn_state_t, timeout_list)
- && cs->expiration_time < timeout_time) {
+ /* Step 2: write completion timeouts */
+ process_timeout_queue(write_completion_q, timeout_time,
+ start_lingering_close_nonblocking);
+ /* Step 3: (normal) lingering close completion timeouts */
+ process_timeout_queue(linger_q, timeout_time,
+ stop_lingering_close);
+ /* Step 4: (short) lingering close completion timeouts */
+ process_timeout_queue(short_linger_q, timeout_time,
+ stop_lingering_close);
- cs->state = CONN_STATE_LINGER;
- APR_RING_REMOVE(cs, timeout_list);
apr_thread_mutex_unlock(timeout_mutex);
- if (!get_worker(&have_idle_worker)) {
- apr_thread_mutex_lock(timeout_mutex);
- APR_RING_INSERT_HEAD(&timeout_head, cs,
- conn_state_t, timeout_list);
- break;
- }
-
- rc = push2worker(&cs->pfd, event_pollset);
- if (rc != APR_SUCCESS) {
- return NULL;
- }
- have_idle_worker = 0;
+ ps->keep_alive = apr_atomic_read32(keepalive_q->total);
+ ps->write_completion = apr_atomic_read32(write_completion_q->total);
+ ps->connections = apr_atomic_read32(&connection_count);
+ ps->suspended = apr_atomic_read32(&suspended_count);
+ ps->lingering_close = apr_atomic_read32(&lingering_count);
+ }
+ else if ((workers_were_busy || dying)
+ && apr_atomic_read32(keepalive_q->total)) {
apr_thread_mutex_lock(timeout_mutex);
- cs = APR_RING_FIRST(&timeout_head);
+ process_keepalive_queue(0); /* kill'em all \m/ */
+ apr_thread_mutex_unlock(timeout_mutex);
+ ps->keep_alive = 0;
}
- apr_thread_mutex_unlock(timeout_mutex);
+ /* If there are some lingering closes to defer (to a worker), schedule
+ * them now. We might wakeup a worker spuriously if another one empties
+ * defer_linger_chain in the meantime, but there also may be no active
+ * or all busy workers for an undefined time. In any case a deferred
+ * lingering close can't starve if we do that here since the chain is
+ * filled only above in the listener and it's emptied only in the
+ * worker(s); thus a NULL here means it will stay so while the listener
+ * waits (possibly indefinitely) in poll().
+ */
+ if (defer_linger_chain) {
+ get_worker(&have_idle_worker, 0, &workers_were_busy);
+ if (have_idle_worker
+ && defer_linger_chain /* re-test */
+ && push2worker(NULL, NULL, NULL) == APR_SUCCESS) {
+ have_idle_worker = 0;
+ }
+ }
+ if (listeners_disabled && !workers_were_busy
+ && (int)apr_atomic_read32(&connection_count)
+ - (int)apr_atomic_read32(&lingering_count)
+ < ((int)ap_queue_info_get_idlers(worker_queue_info) - 1)
+ * worker_factor / WORKER_FACTOR_SCALE + threads_per_child)
+ {
+ listeners_disabled = 0;
+ enable_listensocks(process_slot);
+ }
+ /*
+ * XXX: do we need to set some timeout that re-enables the listensocks
+ * XXX: in case no other event occurs?
+ */
} /* listener main loop */
- ap_close_listeners();
+ close_listeners(process_slot, &closed);
ap_queue_term(worker_queue);
- dying = 1;
- ap_scoreboard_image->parent[process_slot].quiescing = 1;
-
- /* wake up the main thread */
- kill(ap_my_pid, SIGTERM);
apr_thread_exit(thd, APR_SUCCESS);
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,
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;
- conn_state_t *cs;
+ event_conn_state_t *cs;
apr_pool_t *ptrans; /* Pool for per-transaction stuff */
apr_status_t rv;
int is_idle = 0;
timer_event_t *te = NULL;
-
+
free(ti);
ap_scoreboard_image->servers[process_slot][thread_slot].pid = ap_my_pid;
ap_scoreboard_image->servers[process_slot][thread_slot].tid = apr_os_thread_current();
- ap_scoreboard_image->servers[process_slot][thread_slot].generation = my_generation;
+ ap_scoreboard_image->servers[process_slot][thread_slot].generation = retained->mpm->my_generation;
ap_update_child_status_from_indexes(process_slot, thread_slot,
SERVER_STARTING, NULL);
}
ap_update_child_status_from_indexes(process_slot, thread_slot,
- SERVER_READY, NULL);
+ dying ? SERVER_GRACEFUL
+ : SERVER_READY, NULL);
worker_pop:
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);
if (rv != APR_SUCCESS) {
/* We got some other error. */
else if (!workers_may_exit) {
ap_log_error(APLOG_MARK, APLOG_CRIT, rv, ap_server_conf,
- "ap_queue_pop failed");
+ APLOGNO(03099) "ap_queue_pop failed");
}
continue;
}
if (te != NULL) {
-
te->cbfunc(te->baton);
{
- apr_thread_mutex_lock(g_timer_ring_mtx);
+ apr_thread_mutex_lock(g_timer_skiplist_mtx);
APR_RING_INSERT_TAIL(&timer_free_ring, te, timer_event_t, link);
- apr_thread_mutex_unlock(g_timer_ring_mtx);
+ apr_thread_mutex_unlock(g_timer_skiplist_mtx);
}
}
else {
is_idle = 0;
- worker_sockets[thread_slot] = csd;
- rv = process_socket(thd, ptrans, csd, cs, process_slot, thread_slot);
- if (!rv) {
- requests_this_child--;
+ if (csd != NULL) {
+ worker_sockets[thread_slot] = csd;
+ process_socket(thd, ptrans, csd, cs, process_slot, thread_slot);
+ worker_sockets[thread_slot] = NULL;
+ }
+ }
+
+ /* If there are deferred lingering closes, handle them now. */
+ while (!workers_may_exit) {
+ cs = defer_linger_chain;
+ if (!cs) {
+ break;
+ }
+ if (apr_atomic_casptr((void *)&defer_linger_chain, cs->chain,
+ cs) != cs) {
+ /* Race lost, try again */
+ continue;
}
+ cs->chain = NULL;
+
+ worker_sockets[thread_slot] = csd = cs->pfd.desc.s;
+#ifdef AP_DEBUG
+ rv = apr_socket_timeout_set(csd, SECONDS_TO_LINGER);
+ AP_DEBUG_ASSERT(rv == APR_SUCCESS);
+#else
+ apr_socket_timeout_set(csd, SECONDS_TO_LINGER);
+#endif
+ cs->pub.state = CONN_STATE_LINGER;
+ process_socket(thd, cs->p, csd, cs, process_slot, thread_slot);
worker_sockets[thread_slot] = NULL;
}
}
ap_update_child_status_from_indexes(process_slot, thread_slot,
- (dying) ? SERVER_DEAD :
- SERVER_GRACEFUL,
- (request_rec *) NULL);
+ dying ? SERVER_DEAD
+ : SERVER_GRACEFUL, NULL);
apr_thread_exit(thd, APR_SUCCESS);
return NULL;
proc_info *my_info;
apr_status_t rv;
- my_info = (proc_info *) 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 = (proc_info *) ap_malloc(sizeof(proc_info));
+ 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) {
- ap_log_error(APLOG_MARK, APLOG_ALERT, rv, ap_server_conf,
+ ap_log_error(APLOG_MARK, APLOG_ALERT, rv, ap_server_conf, APLOGNO(00474)
"apr_thread_create: unable to create listener thread");
/* let the parent decide how bad this really is */
clean_child_exit(APEXIT_CHILDSICK);
thread_starter *ts = dummy;
apr_thread_t **threads = ts->threads;
apr_threadattr_t *thread_attr = ts->threadattr;
- int child_num_arg = ts->child_num_arg;
- int my_child_num = child_num_arg;
+ int my_child_num = ts->child_num_arg;
proc_info *my_info;
apr_status_t rv;
int i;
int listener_started = 0;
int loops;
int prev_threads_created;
+ int max_recycled_pools = -1;
+ int good_methods[] = {APR_POLLSET_KQUEUE, APR_POLLSET_PORT, APR_POLLSET_EPOLL};
+ /* XXX don't we need more to handle K-A or lingering close? */
+ const apr_uint32_t pollset_size = threads_per_child * 2;
/* We must create the fd queues before we start up the listener
* and worker threads. */
worker_queue = apr_pcalloc(pchild, sizeof(*worker_queue));
rv = ap_queue_init(worker_queue, threads_per_child, pchild);
if (rv != APR_SUCCESS) {
- ap_log_error(APLOG_MARK, APLOG_ALERT, rv, ap_server_conf,
+ ap_log_error(APLOG_MARK, APLOG_ALERT, rv, ap_server_conf, APLOGNO(03100)
"ap_queue_init() failed");
clean_child_exit(APEXIT_CHILDFATAL);
}
+ if (ap_max_mem_free != APR_ALLOCATOR_MAX_FREE_UNLIMITED) {
+ /* If we want to conserve memory, let's not keep an unlimited number of
+ * pools & allocators.
+ * XXX: This should probably be a separate config directive
+ */
+ max_recycled_pools = threads_per_child * 3 / 4 ;
+ }
rv = ap_queue_info_create(&worker_queue_info, pchild,
- threads_per_child);
+ threads_per_child, max_recycled_pools);
if (rv != APR_SUCCESS) {
- ap_log_error(APLOG_MARK, APLOG_ALERT, rv, ap_server_conf,
+ ap_log_error(APLOG_MARK, APLOG_ALERT, rv, ap_server_conf, APLOGNO(03101)
"ap_queue_info_create() failed");
clean_child_exit(APEXIT_CHILDFATAL);
}
rv = apr_thread_mutex_create(&timeout_mutex, APR_THREAD_MUTEX_DEFAULT,
pchild);
if (rv != APR_SUCCESS) {
- ap_log_error(APLOG_MARK, APLOG_ERR, rv, ap_server_conf,
+ ap_log_error(APLOG_MARK, APLOG_ERR, rv, ap_server_conf, APLOGNO(03102)
"creation of the timeout mutex failed.");
clean_child_exit(APEXIT_CHILDFATAL);
}
/* Create the main pollset */
- rv = apr_pollset_create(&event_pollset,
- threads_per_child,
- pchild, APR_POLLSET_THREADSAFE | APR_POLLSET_NOCOPY);
+ for (i = 0; i < sizeof(good_methods) / sizeof(good_methods[0]); i++) {
+ apr_uint32_t flags = APR_POLLSET_THREADSAFE | APR_POLLSET_NOCOPY |
+ APR_POLLSET_NODEFAULT | APR_POLLSET_WAKEABLE;
+ rv = apr_pollset_create_ex(&event_pollset, pollset_size, pchild, flags,
+ good_methods[i]);
+ if (rv == APR_SUCCESS) {
+ listener_is_wakeable = 1;
+ break;
+ }
+ flags &= ~APR_POLLSET_WAKEABLE;
+ rv = apr_pollset_create_ex(&event_pollset, pollset_size, pchild, flags,
+ good_methods[i]);
+ if (rv == APR_SUCCESS) {
+ break;
+ }
+ }
if (rv != APR_SUCCESS) {
- ap_log_error(APLOG_MARK, APLOG_ERR, rv, ap_server_conf,
+ rv = apr_pollset_create(&event_pollset, pollset_size, pchild,
+ APR_POLLSET_THREADSAFE | APR_POLLSET_NOCOPY);
+ }
+ if (rv != APR_SUCCESS) {
+ ap_log_error(APLOG_MARK, APLOG_ERR, rv, ap_server_conf, APLOGNO(03103)
"apr_pollset_create with Thread Safety failed.");
clean_child_exit(APEXIT_CHILDFATAL);
}
+ ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, ap_server_conf, APLOGNO(02471)
+ "start_threads: Using %s (%swakeable)",
+ apr_pollset_method_name(event_pollset),
+ listener_is_wakeable ? "" : "not ");
worker_sockets = apr_pcalloc(pchild, threads_per_child
* sizeof(apr_socket_t *));
/* threads_per_child does not include the listener thread */
for (i = 0; i < threads_per_child; i++) {
int status =
- ap_scoreboard_image->servers[child_num_arg][i].status;
+ ap_scoreboard_image->servers[my_child_num][i].status;
- if (status != SERVER_GRACEFUL && status != SERVER_DEAD) {
+ if (status != SERVER_DEAD) {
continue;
}
- my_info = (proc_info *) malloc(sizeof(proc_info));
- if (my_info == NULL) {
- ap_log_error(APLOG_MARK, APLOG_ALERT, errno, ap_server_conf,
- "malloc: out of memory");
- clean_child_exit(APEXIT_CHILDFATAL);
- }
- my_info->pid = my_child_num;
- my_info->tid = i;
- my_info->sd = 0;
+ my_info = (proc_info *) ap_malloc(sizeof(proc_info));
+ 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,
worker_thread, my_info, pchild);
if (rv != APR_SUCCESS) {
ap_log_error(APLOG_MARK, APLOG_ALERT, rv, ap_server_conf,
+ APLOGNO(03104)
"apr_thread_create: unable to create worker thread");
/* let the parent decide how bad this really is */
clean_child_exit(APEXIT_CHILDSICK);
*/
iter = 0;
- while (iter < 10 &&
-#ifdef HAVE_PTHREAD_KILL
- pthread_kill(*listener_os_thread, 0)
-#else
- kill(ap_my_pid, 0)
-#endif
- == 0) {
- /* listener not dead yet */
+ while (iter < 10 && !dying) {
+ /* listener has not stopped accepting yet */
apr_sleep(apr_time_make(0, 500000));
wakeup_listener();
++iter;
}
if (iter >= 10) {
- ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, ap_server_conf,
- "the listener thread didn't exit");
+ ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, ap_server_conf, APLOGNO(00475)
+ "the listener thread didn't stop accepting");
}
else {
rv = apr_thread_join(&thread_rv, listener);
if (rv != APR_SUCCESS) {
- ap_log_error(APLOG_MARK, APLOG_CRIT, rv, ap_server_conf,
+ ap_log_error(APLOG_MARK, APLOG_CRIT, rv, ap_server_conf, APLOGNO(00476)
"apr_thread_join: unable to join listener thread");
}
}
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,
+ ap_log_error(APLOG_MARK, APLOG_CRIT, rv, ap_server_conf, APLOGNO(00477)
"apr_thread_join: unable to join worker "
"thread %d", i);
}
*/
rv = apr_thread_join(&thread_rv, start_thread_id);
if (rv != APR_SUCCESS) {
- ap_log_error(APLOG_MARK, APLOG_CRIT, rv, ap_server_conf,
+ ap_log_error(APLOG_MARK, APLOG_CRIT, rv, ap_server_conf, APLOGNO(00478)
"apr_thread_join: unable to join the start " "thread");
}
}
-static void child_main(int child_num_arg)
+static void child_main(int child_num_arg, int child_bucket)
{
apr_thread_t **threads;
apr_status_t rv;
thread_starter *ts;
apr_threadattr_t *thread_attr;
apr_thread_t *start_thread_id;
+ int i;
+
+ /* for benefit of any hooks that run as this child initializes */
+ retained->mpm->mpm_state = AP_MPMQ_STARTING;
- mpm_state = AP_MPMQ_STARTING; /* for benefit of any hooks that run as this
- * child initializes
- */
ap_my_pid = getpid();
ap_fatal_signal_child_setup(ap_server_conf);
apr_pool_create(&pchild, pconf);
+ /* close unused listeners and pods */
+ for (i = 0; i < retained->mpm->num_buckets; i++) {
+ if (i != child_bucket) {
+ ap_close_listeners_ex(all_buckets[i].listeners);
+ ap_mpm_podx_close(all_buckets[i].pod);
+ }
+ }
+
/*stuff to do before we switch id's, so we have permissions. */
ap_reopen_scoreboard(pchild, NULL, 0);
clean_child_exit(APEXIT_CHILDFATAL);
}
- apr_thread_mutex_create(&g_timer_ring_mtx, APR_THREAD_MUTEX_DEFAULT, pchild);
+ apr_thread_mutex_create(&g_timer_skiplist_mtx, APR_THREAD_MUTEX_DEFAULT, pchild);
APR_RING_INIT(&timer_free_ring, timer_event_t, link);
- APR_RING_INIT(&timer_ring, timer_event_t, link);
-
+ apr_skiplist_init(&timer_skiplist, pchild);
+ apr_skiplist_set_compare(timer_skiplist, timer_comp, timer_comp);
ap_run_child_init(pchild, ap_server_conf);
/* done with init critical section */
*/
rv = apr_setup_signal_thread();
if (rv != APR_SUCCESS) {
- ap_log_error(APLOG_MARK, APLOG_EMERG, rv, ap_server_conf,
+ ap_log_error(APLOG_MARK, APLOG_EMERG, rv, ap_server_conf, APLOGNO(00479)
"Couldn't initialize signal thread");
clean_child_exit(APEXIT_CHILDFATAL);
}
if (ap_max_requests_per_child) {
- requests_this_child = ap_max_requests_per_child;
+ conns_this_child = ap_max_requests_per_child;
}
else {
/* coding a value of zero means infinity */
- requests_this_child = INT_MAX;
+ conns_this_child = APR_INT32_MAX;
}
/* Setup worker threads */
/* clear the storage; we may not create all our threads immediately,
* and we want a 0 entry to indicate a thread which was not created
*/
- threads = (apr_thread_t **) calloc(1,
- sizeof(apr_thread_t *) *
- threads_per_child);
- if (threads == NULL) {
- ap_log_error(APLOG_MARK, APLOG_ALERT, errno, ap_server_conf,
- "malloc: out of memory");
- clean_child_exit(APEXIT_CHILDFATAL);
- }
-
- ts = (thread_starter *) apr_palloc(pchild, sizeof(*ts));
+ threads = ap_calloc(threads_per_child, sizeof(apr_thread_t *));
+ ts = apr_palloc(pchild, sizeof(*ts));
apr_threadattr_create(&thread_attr, pchild);
/* 0 means PTHREAD_CREATE_JOINABLE */
apr_threadattr_detach_set(thread_attr, 0);
if (ap_thread_stacksize != 0) {
- apr_threadattr_stacksize_set(thread_attr, ap_thread_stacksize);
+ rv = apr_threadattr_stacksize_set(thread_attr, ap_thread_stacksize);
+ if (rv != APR_SUCCESS && rv != APR_ENOTIMPL) {
+ ap_log_error(APLOG_MARK, APLOG_WARNING, rv, ap_server_conf, APLOGNO(02436)
+ "WARNING: ThreadStackSize of %" APR_SIZE_T_FMT " is "
+ "inappropriate, using default",
+ ap_thread_stacksize);
+ }
}
ts->threads = threads;
rv = apr_thread_create(&start_thread_id, thread_attr, start_threads,
ts, pchild);
if (rv != APR_SUCCESS) {
- ap_log_error(APLOG_MARK, APLOG_ALERT, rv, ap_server_conf,
+ ap_log_error(APLOG_MARK, APLOG_ALERT, rv, ap_server_conf, APLOGNO(00480)
"apr_thread_create: unable to create worker thread");
/* let the parent decide how bad this really is */
clean_child_exit(APEXIT_CHILDSICK);
}
- mpm_state = AP_MPMQ_RUNNING;
+ retained->mpm->mpm_state = AP_MPMQ_RUNNING;
/* If we are only running in one_process mode, we will want to
* still handle signals. */
else { /* !one_process */
/* remove SIGTERM from the set of blocked signals... if one of
* the other threads in the process needs to take us down
- * (e.g., for MaxRequestsPerChild) it will send us SIGTERM
+ * (e.g., for MaxConnectionsPerChild) it will send us SIGTERM
*/
unblock_signal(SIGTERM);
apr_signal(SIGTERM, dummy_signal_handler);
/* Watch for any messages from the parent over the POD */
while (1) {
- rv = ap_event_pod_check(pod);
- if (rv == AP_NORESTART) {
+ rv = ap_mpm_podx_check(my_bucket->pod);
+ if (rv == AP_MPM_PODX_NORESTART) {
/* see if termination was triggered while we slept */
switch (terminate_mode) {
case ST_GRACEFUL:
- rv = AP_GRACEFUL;
+ rv = AP_MPM_PODX_GRACEFUL;
break;
case ST_UNGRACEFUL:
- rv = AP_RESTART;
+ rv = AP_MPM_PODX_RESTART;
break;
}
}
- if (rv == AP_GRACEFUL || rv == AP_RESTART) {
+ if (rv == AP_MPM_PODX_GRACEFUL || rv == AP_MPM_PODX_RESTART) {
/* make sure the start thread has finished;
* signal_threads() and join_workers depend on that
*/
join_start_thread(start_thread_id);
signal_threads(rv ==
- AP_GRACEFUL ? ST_GRACEFUL : ST_UNGRACEFUL);
+ AP_MPM_PODX_GRACEFUL ? ST_GRACEFUL : ST_UNGRACEFUL);
break;
}
}
clean_child_exit(resource_shortage ? APEXIT_CHILDSICK : 0);
}
-static int make_child(server_rec * s, int slot)
+static int make_child(server_rec * s, int slot, int bucket)
{
int pid;
- if (slot + 1 > max_daemons_limit) {
- max_daemons_limit = slot + 1;
+ if (slot + 1 > retained->max_daemons_limit) {
+ retained->max_daemons_limit = slot + 1;
+ }
+
+ if (ap_scoreboard_image->parent[slot].pid != 0) {
+ /* XXX replace with assert or remove ? */
+ ap_log_error(APLOG_MARK, APLOG_ERR, 0, ap_server_conf, APLOGNO(03455)
+ "BUG: Scoreboard slot %d should be empty but is "
+ "in use by pid %" APR_PID_T_FMT,
+ slot, ap_scoreboard_image->parent[slot].pid);
+ return -1;
}
if (one_process) {
- set_signals();
- ap_scoreboard_image->parent[slot].pid = getpid();
- child_main(slot);
+ my_bucket = &all_buckets[0];
+
+ event_note_child_started(slot, getpid());
+ child_main(slot, 0);
+ /* NOTREACHED */
+ ap_assert(0);
+ return -1;
}
if ((pid = fork()) == -1) {
- ap_log_error(APLOG_MARK, APLOG_ERR, errno, s,
+ ap_log_error(APLOG_MARK, APLOG_ERR, errno, s, APLOGNO(00481)
"fork: Unable to fork new process");
/* fork didn't succeed. There's no need to touch the scoreboard;
}
if (!pid) {
+ my_bucket = &all_buckets[bucket];
+
#ifdef HAVE_BINDPROCESSOR
/* By default, AIX binds to a single processor. This bit unbinds
* children which will then bind to another CPU.
PROCESSOR_CLASS_ANY);
if (status != OK)
ap_log_error(APLOG_MARK, APLOG_DEBUG, errno,
- ap_server_conf,
+ ap_server_conf, APLOGNO(00482)
"processor unbind failed");
#endif
RAISE_SIGSTOP(MAKE_CHILD);
apr_signal(SIGTERM, just_die);
- child_main(slot);
-
- clean_child_exit(0);
- }
- /* else */
- if (ap_scoreboard_image->parent[slot].pid != 0) {
- /* This new child process is squatting on the scoreboard
- * entry owned by an exiting child process, which cannot
- * exit until all active requests complete.
- * Don't forget about this exiting child process, or we
- * won't be able to kill it if it doesn't exit by the
- * time the server is shut down.
- */
- ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, ap_server_conf,
- "taking over scoreboard slot from %" APR_PID_T_FMT "%s",
- ap_scoreboard_image->parent[slot].pid,
- ap_scoreboard_image->parent[slot].quiescing ?
- " (quiescing)" : "");
- ap_register_extra_mpm_process(ap_scoreboard_image->parent[slot].pid);
+ child_main(slot, bucket);
+ /* NOTREACHED */
+ ap_assert(0);
+ return -1;
}
+
ap_scoreboard_image->parent[slot].quiescing = 0;
- ap_scoreboard_image->parent[slot].pid = pid;
+ ap_scoreboard_image->parent[slot].not_accepting = 0;
+ ap_scoreboard_image->parent[slot].bucket = bucket;
+ event_note_child_started(slot, pid);
+ active_daemons++;
+ retained->total_daemons++;
return 0;
}
{
int i;
- for (i = 0; number_to_start && i < ap_daemons_limit; ++i) {
+ for (i = 0; number_to_start && i < server_limit; ++i) {
if (ap_scoreboard_image->parent[i].pid != 0) {
continue;
}
- if (make_child(ap_server_conf, i) < 0) {
+ if (make_child(ap_server_conf, i, i % retained->mpm->num_buckets) < 0) {
break;
}
--number_to_start;
}
}
-
-/*
- * idle_spawn_rate is the number of children that will be spawned on the
- * next maintenance cycle if there aren't enough idle servers. It is
- * doubled up to MAX_SPAWN_RATE, and reset only when a cycle goes by
- * without the need to spawn.
- */
-static int idle_spawn_rate = 1;
-#ifndef MAX_SPAWN_RATE
-#define MAX_SPAWN_RATE (32)
-#endif
-static int hold_off_on_exponential_spawning;
-
-static void perform_idle_server_maintenance(void)
+static void perform_idle_server_maintenance(int child_bucket, int num_buckets)
{
int i, j;
- int idle_thread_count;
+ int idle_thread_count = 0;
worker_score *ws;
process_score *ps;
- int free_length;
- int totally_free_length = 0;
+ int free_length = 0;
int free_slots[MAX_SPAWN_RATE];
- int last_non_dead;
- int total_non_dead;
+ int last_non_dead = -1;
int active_thread_count = 0;
- /* initialize the free_list */
- free_length = 0;
-
- idle_thread_count = 0;
- last_non_dead = -1;
- total_non_dead = 0;
-
- for (i = 0; i < ap_daemons_limit; ++i) {
+ for (i = 0; i < server_limit; ++i) {
/* Initialization to satisfy the compiler. It doesn't know
* that threads_per_child is always > 0 */
int status = SERVER_DEAD;
- int any_dying_threads = 0;
- int any_dead_threads = 0;
- int all_dead_threads = 1;
+ int child_threads_active = 0;
- if (i >= max_daemons_limit
- && totally_free_length == idle_spawn_rate)
+ if (i >= retained->max_daemons_limit &&
+ free_length == retained->idle_spawn_rate[child_bucket]) {
/* short cut if all active processes have been examined and
* enough empty scoreboard slots have been found
*/
-
+
break;
+ }
ps = &ap_scoreboard_image->parent[i];
- for (j = 0; j < threads_per_child; j++) {
- ws = &ap_scoreboard_image->servers[i][j];
- status = ws->status;
-
- /* XXX any_dying_threads is probably no longer needed GLA */
- any_dying_threads = any_dying_threads ||
- (status == SERVER_GRACEFUL);
- any_dead_threads = any_dead_threads || (status == SERVER_DEAD);
- all_dead_threads = all_dead_threads &&
- (status == SERVER_DEAD || status == SERVER_GRACEFUL);
-
- /* We consider a starting server as idle because we started it
- * at least a cycle ago, and if it still hasn't finished starting
- * then we're just going to swamp things worse by forking more.
- * So we hopefully won't need to fork more if we count it.
- * This depends on the ordering of SERVER_READY and SERVER_STARTING.
- */
- if (ps->pid != 0) { /* XXX just set all_dead_threads in outer
- for loop if no pid? not much else matters */
- if (status <= SERVER_READY &&
- !ps->quiescing && ps->generation == my_generation) {
+ if (ps->pid != 0) {
+ for (j = 0; j < threads_per_child; j++) {
+ ws = &ap_scoreboard_image->servers[i][j];
+ status = ws->status;
+
+ /* We consider a starting server as idle because we started it
+ * at least a cycle ago, and if it still hasn't finished starting
+ * then we're just going to swamp things worse by forking more.
+ * So we hopefully won't need to fork more if we count it.
+ * This depends on the ordering of SERVER_READY and SERVER_STARTING.
+ */
+ if (status <= SERVER_READY && !ps->quiescing && !ps->not_accepting
+ && ps->generation == retained->mpm->my_generation
+ && ps->bucket == child_bucket)
+ {
++idle_thread_count;
}
if (status >= SERVER_READY && status < SERVER_GRACEFUL) {
- ++active_thread_count;
+ ++child_threads_active;
}
}
- }
- if (any_dead_threads
- && totally_free_length < idle_spawn_rate
- && free_length < MAX_SPAWN_RATE
- && (!ps->pid /* no process in the slot */
- || ps->quiescing)) { /* or at least one is going away */
- if (all_dead_threads) {
- /* great! we prefer these, because the new process can
- * start more threads sooner. So prioritize this slot
- * by putting it ahead of any slots with active threads.
- *
- * first, make room by moving a slot that's potentially still
- * in use to the end of the array
- */
- free_slots[free_length] = free_slots[totally_free_length];
- free_slots[totally_free_length++] = i;
- }
- else {
- /* slot is still in use - back of the bus
- */
- free_slots[free_length] = i;
- }
- ++free_length;
- }
- /* XXX if (!ps->quiescing) is probably more reliable GLA */
- if (!any_dying_threads) {
last_non_dead = i;
- ++total_non_dead;
}
+ active_thread_count += child_threads_active;
+ if (!ps->pid && free_length < retained->idle_spawn_rate[child_bucket])
+ free_slots[free_length++] = i;
+ else if (child_threads_active == threads_per_child)
+ had_healthy_child = 1;
}
- if (sick_child_detected) {
- if (active_thread_count > 0) {
- /* some child processes appear to be working. don't kill the
- * whole server.
+ if (retained->sick_child_detected) {
+ if (had_healthy_child) {
+ /* Assume this is a transient error, even though it may not be. Leave
+ * the server up in case it is able to serve some requests or the
+ * problem will be resolved.
*/
- sick_child_detected = 0;
+ retained->sick_child_detected = 0;
}
else {
- /* looks like a basket case. give up.
+ /* looks like a basket case, as no child ever fully initialized; give up.
*/
- shutdown_pending = 1;
+ retained->mpm->shutdown_pending = 1;
child_fatal = 1;
ap_log_error(APLOG_MARK, APLOG_ALERT, 0,
- ap_server_conf,
- "No active workers found..."
- " Apache is exiting!");
+ ap_server_conf, APLOGNO(02324)
+ "A resource shortage or other unrecoverable failure "
+ "was encountered before any child process initialized "
+ "successfully... httpd is exiting!");
/* the child already logged the failure details */
return;
}
}
- max_daemons_limit = last_non_dead + 1;
+ retained->max_daemons_limit = last_non_dead + 1;
- if (idle_thread_count > max_spare_threads) {
- /* Kill off one child */
- ap_event_pod_signal(pod, TRUE);
- idle_spawn_rate = 1;
+ if (idle_thread_count > max_spare_threads / num_buckets)
+ {
+ /*
+ * Child processes that we ask to shut down won't die immediately
+ * but may stay around for a long time when they finish their
+ * requests. If the server load changes many times, many such
+ * gracefully finishing processes may accumulate, filling up the
+ * scoreboard. To avoid running out of scoreboard entries, we
+ * don't shut down more processes when the total number of processes
+ * is high.
+ *
+ * XXX It would be nice if we could
+ * XXX - kill processes without keepalive connections first
+ * XXX - tell children to stop accepting new connections, and
+ * XXX depending on server load, later be able to resurrect them
+ * or kill them
+ */
+ if (retained->total_daemons <= active_daemons_limit &&
+ retained->total_daemons < server_limit) {
+ /* Kill off one child */
+ ap_mpm_podx_signal(all_buckets[child_bucket].pod,
+ AP_MPM_PODX_GRACEFUL);
+ retained->idle_spawn_rate[child_bucket] = 1;
+ active_daemons--;
+ } else {
+ ap_log_error(APLOG_MARK, APLOG_TRACE5, 0, ap_server_conf,
+ "Not shutting down child: total daemons %d / "
+ "active limit %d / ServerLimit %d",
+ retained->total_daemons, active_daemons_limit,
+ server_limit);
+ }
}
- else if (idle_thread_count < min_spare_threads) {
- /* terminate the free list */
- if (free_length == 0) { /* scoreboard is full, can't fork */
-
- if (active_thread_count >= ap_daemons_limit * threads_per_child) {
- static int reported = 0;
- if (!reported) {
- /* only report this condition once */
- ap_log_error(APLOG_MARK, APLOG_ERR, 0,
- ap_server_conf,
- "server reached MaxClients setting, consider"
- " raising the MaxClients setting");
- reported = 1;
- }
- }
- else {
- ap_log_error(APLOG_MARK, APLOG_ERR, 0,
- ap_server_conf,
- "scoreboard is full, not at MaxClients");
+ else if (idle_thread_count < min_spare_threads / num_buckets) {
+ if (active_thread_count >= max_workers) {
+ if (!retained->maxclients_reported) {
+ /* only report this condition once */
+ ap_log_error(APLOG_MARK, APLOG_ERR, 0, ap_server_conf, APLOGNO(00484)
+ "server reached MaxRequestWorkers setting, "
+ "consider raising the MaxRequestWorkers "
+ "setting");
+ retained->maxclients_reported = 1;
}
- idle_spawn_rate = 1;
+ retained->idle_spawn_rate[child_bucket] = 1;
+ }
+ else if (free_length == 0) { /* scoreboard is full, can't fork */
+ ap_log_error(APLOG_MARK, APLOG_ERR, 0, ap_server_conf, APLOGNO(03490)
+ "scoreboard is full, not at MaxRequestWorkers."
+ "Increase ServerLimit.");
+ retained->idle_spawn_rate[child_bucket] = 1;
}
else {
- if (free_length > idle_spawn_rate) {
- free_length = idle_spawn_rate;
+ if (free_length > retained->idle_spawn_rate[child_bucket]) {
+ free_length = retained->idle_spawn_rate[child_bucket];
}
- if (idle_spawn_rate >= 8) {
- ap_log_error(APLOG_MARK, APLOG_INFO, 0,
- ap_server_conf,
+ if (retained->idle_spawn_rate[child_bucket] >= 8) {
+ ap_log_error(APLOG_MARK, APLOG_INFO, 0, ap_server_conf, APLOGNO(00486)
"server seems busy, (you may need "
"to increase StartServers, ThreadsPerChild "
"or Min/MaxSpareThreads), "
"spawning %d children, there are around %d idle "
- "threads, and %d total children", free_length,
- idle_thread_count, total_non_dead);
+ "threads, %d active children, and %d children "
+ "that are shutting down", free_length,
+ idle_thread_count, active_daemons,
+ retained->total_daemons);
}
for (i = 0; i < free_length; ++i) {
- make_child(ap_server_conf, free_slots[i]);
+ ap_log_error(APLOG_MARK, APLOG_TRACE5, 0, ap_server_conf,
+ "Spawning new child: slot %d active / "
+ "total daemons: %d/%d",
+ free_slots[i], active_daemons,
+ retained->total_daemons);
+ make_child(ap_server_conf, free_slots[i], child_bucket);
}
/* the next time around we want to spawn twice as many if this
* wasn't good enough, but not if we've just done a graceful
*/
- if (hold_off_on_exponential_spawning) {
- --hold_off_on_exponential_spawning;
+ if (retained->hold_off_on_exponential_spawning) {
+ --retained->hold_off_on_exponential_spawning;
}
- else if (idle_spawn_rate < MAX_SPAWN_RATE) {
- idle_spawn_rate *= 2;
+ else if (retained->idle_spawn_rate[child_bucket]
+ < MAX_SPAWN_RATE / num_buckets) {
+ retained->idle_spawn_rate[child_bucket] *= 2;
}
}
}
else {
- idle_spawn_rate = 1;
+ retained->idle_spawn_rate[child_bucket] = 1;
}
}
-static void server_main_loop(int remaining_children_to_start)
+static void server_main_loop(int remaining_children_to_start, int num_buckets)
{
int child_slot;
apr_exit_why_e exitwhy;
apr_proc_t pid;
int i;
- while (!restart_pending && !shutdown_pending) {
+ while (!retained->mpm->restart_pending && !retained->mpm->shutdown_pending) {
ap_wait_or_timeout(&exitwhy, &status, &pid, pconf, ap_server_conf);
if (pid.pid != -1) {
processed_status = ap_process_child_status(&pid, exitwhy, status);
+ child_slot = ap_find_child_by_pid(&pid);
if (processed_status == APEXIT_CHILDFATAL) {
- shutdown_pending = 1;
- child_fatal = 1;
- return;
+ /* fix race condition found in PR 39311
+ * A child created at the same time as a graceful happens
+ * can find the lock missing and create a fatal error.
+ * It is not fatal for the last generation to be in this state.
+ */
+ if (child_slot < 0
+ || ap_get_scoreboard_process(child_slot)->generation
+ == retained->mpm->my_generation) {
+ retained->mpm->shutdown_pending = 1;
+ child_fatal = 1;
+ /*
+ * total_daemons counting will be off now, but as we
+ * are shutting down, that is not an issue anymore.
+ */
+ return;
+ }
+ else {
+ ap_log_error(APLOG_MARK, APLOG_WARNING, 0, ap_server_conf, APLOGNO(00487)
+ "Ignoring fatal error in child of previous "
+ "generation (pid %ld).",
+ (long)pid.pid);
+ retained->sick_child_detected = 1;
+ }
}
else if (processed_status == APEXIT_CHILDSICK) {
/* tell perform_idle_server_maintenance to check into this
* on the next timer pop
*/
- sick_child_detected = 1;
+ retained->sick_child_detected = 1;
}
/* non-fatal death... note that it's gone in the scoreboard. */
- child_slot = ap_find_child_by_pid(&pid);
if (child_slot >= 0) {
+ process_score *ps;
+
for (i = 0; i < threads_per_child; i++)
ap_update_child_status_from_indexes(child_slot, i,
- SERVER_DEAD,
- (request_rec *) NULL);
-
- ap_scoreboard_image->parent[child_slot].pid = 0;
- ap_scoreboard_image->parent[child_slot].quiescing = 0;
+ SERVER_DEAD, NULL);
+
+ event_note_child_killed(child_slot, 0, 0);
+ ps = &ap_scoreboard_image->parent[child_slot];
+ if (!ps->quiescing)
+ active_daemons--;
+ ps->quiescing = 0;
+ /* NOTE: We don't dec in the (child_slot < 0) case! */
+ retained->total_daemons--;
if (processed_status == APEXIT_CHILDSICK) {
/* resource shortage, minimize the fork rate */
- idle_spawn_rate = 1;
+ retained->idle_spawn_rate[ps->bucket] = 1;
}
- else if (remaining_children_to_start
- && child_slot < ap_daemons_limit) {
+ else if (remaining_children_to_start) {
/* we're still doing a 1-for-1 replacement of dead
* children with new children
*/
- make_child(ap_server_conf, child_slot);
+ make_child(ap_server_conf, child_slot, ps->bucket);
--remaining_children_to_start;
}
}
- else if (ap_unregister_extra_mpm_process(pid.pid) == 1) {
- /* handled */
#if APR_HAS_OTHER_CHILD
- }
else if (apr_proc_other_child_alert(&pid, APR_OC_REASON_DEATH,
status) == 0) {
/* handled */
-#endif
}
- else if (is_graceful) {
+#endif
+ else if (retained->mpm->was_graceful) {
/* Great, we've probably just lost a slot in the
* scoreboard. Somehow we don't know about this child.
*/
ap_log_error(APLOG_MARK, APLOG_WARNING, 0,
- ap_server_conf,
+ ap_server_conf, APLOGNO(00488)
"long lost child came home! (pid %ld)",
(long) pid.pid);
}
continue;
}
- perform_idle_server_maintenance();
+ for (i = 0; i < num_buckets; i++) {
+ perform_idle_server_maintenance(i, num_buckets);
+ }
}
}
static int event_run(apr_pool_t * _pconf, apr_pool_t * plog, server_rec * s)
{
+ int num_buckets = retained->mpm->num_buckets;
int remaining_children_to_start;
+ int i;
ap_log_pid(pconf, ap_pid_fname);
- if (!is_graceful) {
+ if (!retained->mpm->was_graceful) {
if (ap_run_pre_mpm(s->process->pool, SB_SHARED) != OK) {
- mpm_state = AP_MPMQ_STOPPING;
- return DONE;
+ retained->mpm->mpm_state = AP_MPMQ_STOPPING;
+ return !OK;
}
/* fix the generation number in the global score; we just got a new,
* cleared scoreboard
*/
- ap_scoreboard_image->global->running_generation = my_generation;
+ ap_scoreboard_image->global->running_generation = retained->mpm->my_generation;
+ }
+
+ if (!one_process) {
+ ap_fatal_signal_setup(ap_server_conf, pconf);
}
+ ap_unixd_mpm_set_signals(pconf, one_process);
- set_signals();
- /* Don't thrash... */
- if (max_spare_threads < min_spare_threads + threads_per_child)
- max_spare_threads = min_spare_threads + threads_per_child;
+ /* Don't thrash since num_buckets depends on the
+ * system and the number of online CPU cores...
+ */
+ if (active_daemons_limit < num_buckets)
+ active_daemons_limit = num_buckets;
+ if (ap_daemons_to_start < num_buckets)
+ ap_daemons_to_start = num_buckets;
+ /* We want to create as much children at a time as the number of buckets,
+ * so to optimally accept connections (evenly distributed across buckets).
+ * Thus min_spare_threads should at least maintain num_buckets children,
+ * and max_spare_threads allow num_buckets more children w/o triggering
+ * immediately (e.g. num_buckets idle threads margin, one per bucket).
+ */
+ if (min_spare_threads < threads_per_child * (num_buckets - 1) + num_buckets)
+ min_spare_threads = threads_per_child * (num_buckets - 1) + num_buckets;
+ if (max_spare_threads < min_spare_threads + (threads_per_child + 1) * num_buckets)
+ max_spare_threads = min_spare_threads + (threads_per_child + 1) * num_buckets;
/* If we're doing a graceful_restart then we're going to see a lot
* of children exiting immediately when we get into the main loop
* supposed to start up without the 1 second penalty between each fork.
*/
remaining_children_to_start = ap_daemons_to_start;
- if (remaining_children_to_start > ap_daemons_limit) {
- remaining_children_to_start = ap_daemons_limit;
+ if (remaining_children_to_start > active_daemons_limit) {
+ remaining_children_to_start = active_daemons_limit;
}
- if (!is_graceful) {
+ if (!retained->mpm->was_graceful) {
startup_children(remaining_children_to_start);
remaining_children_to_start = 0;
}
else {
/* give the system some time to recover before kicking into
* exponential mode */
- hold_off_on_exponential_spawning = 10;
+ retained->hold_off_on_exponential_spawning = 10;
}
- ap_log_error(APLOG_MARK, APLOG_NOTICE, 0, ap_server_conf,
+ ap_log_error(APLOG_MARK, APLOG_NOTICE, 0, ap_server_conf, APLOGNO(00489)
"%s configured -- resuming normal operations",
ap_get_server_description());
- ap_log_error(APLOG_MARK, APLOG_INFO, 0, ap_server_conf,
+ ap_log_error(APLOG_MARK, APLOG_INFO, 0, ap_server_conf, APLOGNO(00490)
"Server built: %s", ap_get_server_built());
ap_log_command_line(plog, s);
+ ap_log_mpm_common(s);
- restart_pending = shutdown_pending = 0;
- mpm_state = AP_MPMQ_RUNNING;
+ retained->mpm->mpm_state = AP_MPMQ_RUNNING;
- server_main_loop(remaining_children_to_start);
- mpm_state = AP_MPMQ_STOPPING;
+ server_main_loop(remaining_children_to_start, num_buckets);
+ retained->mpm->mpm_state = AP_MPMQ_STOPPING;
- if (shutdown_pending && !is_graceful) {
+ if (retained->mpm->shutdown_pending && retained->mpm->is_ungraceful) {
/* Time to shut down:
* Kill child processes, tell them to call child_exit, etc...
*/
- ap_event_pod_killpg(pod, ap_daemons_limit, FALSE);
- ap_reclaim_child_processes(1); /* Start with SIGTERM */
+ for (i = 0; i < num_buckets; i++) {
+ ap_mpm_podx_killpg(all_buckets[i].pod, active_daemons_limit,
+ AP_MPM_PODX_RESTART);
+ }
+ ap_reclaim_child_processes(1, /* Start with SIGTERM */
+ event_note_child_killed);
if (!child_fatal) {
/* cleanup pid file on normal shutdown */
- const char *pidfile = NULL;
- pidfile = ap_server_root_relative(pconf, ap_pid_fname);
- if (pidfile != NULL && unlink(pidfile) == 0)
- ap_log_error(APLOG_MARK, APLOG_INFO, 0,
- ap_server_conf,
- "removed PID file %s (pid=%ld)",
- pidfile, (long) getpid());
-
+ ap_remove_pid(pconf, ap_pid_fname);
ap_log_error(APLOG_MARK, APLOG_NOTICE, 0,
- ap_server_conf, "caught SIGTERM, shutting down");
+ ap_server_conf, APLOGNO(00491) "caught SIGTERM, shutting down");
}
+
return DONE;
- } else if (shutdown_pending) {
+ }
+
+ if (retained->mpm->shutdown_pending) {
/* Time to gracefully shut down:
* Kill child processes, tell them to call child_exit, etc...
*/
/* Close our listeners, and then ask our children to do same */
ap_close_listeners();
- ap_event_pod_killpg(pod, ap_daemons_limit, TRUE);
- ap_relieve_child_processes();
+ for (i = 0; i < num_buckets; i++) {
+ ap_mpm_podx_killpg(all_buckets[i].pod, active_daemons_limit,
+ AP_MPM_PODX_GRACEFUL);
+ }
+ ap_relieve_child_processes(event_note_child_killed);
if (!child_fatal) {
/* cleanup pid file on normal shutdown */
- const char *pidfile = NULL;
- pidfile = ap_server_root_relative (pconf, ap_pid_fname);
- if ( pidfile != NULL && unlink(pidfile) == 0)
- ap_log_error(APLOG_MARK, APLOG_INFO, 0,
- ap_server_conf,
- "removed PID file %s (pid=%ld)",
- pidfile, (long)getpid());
-
- ap_log_error(APLOG_MARK, APLOG_NOTICE, 0, ap_server_conf,
+ ap_remove_pid(pconf, ap_pid_fname);
+ ap_log_error(APLOG_MARK, APLOG_NOTICE, 0, ap_server_conf, APLOGNO(00492)
"caught " AP_SIG_GRACEFUL_STOP_STRING
", shutting down gracefully");
}
}
/* Don't really exit until each child has finished */
- shutdown_pending = 0;
+ retained->mpm->shutdown_pending = 0;
do {
/* Pause for a second */
apr_sleep(apr_time_from_sec(1));
/* Relieve any children which have now exited */
- ap_relieve_child_processes();
+ ap_relieve_child_processes(event_note_child_killed);
active_children = 0;
- for (index = 0; index < ap_daemons_limit; ++index) {
+ for (index = 0; index < retained->max_daemons_limit; ++index) {
if (ap_mpm_safe_kill(MPM_CHILD_PID(index), 0) == APR_SUCCESS) {
active_children = 1;
/* Having just one child is enough to stay around */
break;
}
}
- } while (!shutdown_pending && active_children &&
+ } while (!retained->mpm->shutdown_pending && active_children &&
(!ap_graceful_shutdown_timeout || apr_time_now() < cutoff));
/* We might be here because we received SIGTERM, either
* way, try and make sure that all of our processes are
* really dead.
*/
- ap_event_pod_killpg(pod, ap_daemons_limit, FALSE);
- ap_reclaim_child_processes(1);
+ for (i = 0; i < num_buckets; i++) {
+ ap_mpm_podx_killpg(all_buckets[i].pod, active_daemons_limit,
+ AP_MPM_PODX_RESTART);
+ }
+ ap_reclaim_child_processes(1, event_note_child_killed);
return DONE;
}
/* we've been told to restart */
- apr_signal(SIGHUP, SIG_IGN);
-
if (one_process) {
/* not worth thinking about */
return DONE;
/* XXX: we really need to make sure this new generation number isn't in
* use by any of the children.
*/
- ++my_generation;
- ap_scoreboard_image->global->running_generation = my_generation;
+ ++retained->mpm->my_generation;
+ ap_scoreboard_image->global->running_generation = retained->mpm->my_generation;
- if (is_graceful) {
- ap_log_error(APLOG_MARK, APLOG_NOTICE, 0, ap_server_conf,
+ if (!retained->mpm->is_ungraceful) {
+ ap_log_error(APLOG_MARK, APLOG_NOTICE, 0, ap_server_conf, APLOGNO(00493)
AP_SIG_GRACEFUL_STRING
" received. Doing graceful restart");
/* wake up the children...time to die. But we'll have more soon */
- ap_event_pod_killpg(pod, ap_daemons_limit, TRUE);
-
+ for (i = 0; i < num_buckets; i++) {
+ ap_mpm_podx_killpg(all_buckets[i].pod, active_daemons_limit,
+ AP_MPM_PODX_GRACEFUL);
+ }
/* This is mostly for debugging... so that we know what is still
* gracefully dealing with existing request.
* and a SIGHUP, we may as well use the same signal, because some user
* pthreads are stealing signals from us left and right.
*/
- ap_event_pod_killpg(pod, ap_daemons_limit, FALSE);
+ for (i = 0; i < num_buckets; i++) {
+ ap_mpm_podx_killpg(all_buckets[i].pod, active_daemons_limit,
+ AP_MPM_PODX_RESTART);
+ }
- ap_reclaim_child_processes(1); /* Start with SIGTERM */
- ap_log_error(APLOG_MARK, APLOG_NOTICE, 0, ap_server_conf,
+ ap_reclaim_child_processes(1, /* Start with SIGTERM */
+ event_note_child_killed);
+ ap_log_error(APLOG_MARK, APLOG_NOTICE, 0, ap_server_conf, APLOGNO(00494)
"SIGHUP received. Attempting to restart");
}
+ active_daemons = 0;
+
+ return OK;
+}
+
+static void setup_slave_conn(conn_rec *c, void *csd)
+{
+ event_conn_state_t *mcs;
+ event_conn_state_t *cs;
+
+ mcs = ap_get_module_config(c->master->conn_config, &mpm_event_module);
+
+ cs = apr_pcalloc(c->pool, sizeof(*cs));
+ cs->c = c;
+ cs->r = NULL;
+ cs->sc = mcs->sc;
+ cs->suspended = 0;
+ cs->p = c->pool;
+ cs->bucket_alloc = c->bucket_alloc;
+ cs->pfd = mcs->pfd;
+ cs->pub = mcs->pub;
+ cs->pub.state = CONN_STATE_READ_REQUEST_LINE;
+ cs->pub.sense = CONN_SENSE_DEFAULT;
+
+ c->cs = &(cs->pub);
+ ap_set_module_config(c->conn_config, &mpm_event_module, cs);
+}
+
+static int event_pre_connection(conn_rec *c, void *csd)
+{
+ if (c->master && (!c->cs || c->cs == c->master->cs)) {
+ setup_slave_conn(c, csd);
+ }
return OK;
}
+static int event_protocol_switch(conn_rec *c, request_rec *r, server_rec *s,
+ const char *protocol)
+{
+ if (!r && s) {
+ /* connection based switching of protocol, set the correct server
+ * configuration, so that timeouts, keepalives and such are used
+ * for the server that the connection was switched on.
+ * Normally, we set this on post_read_request, but on a protocol
+ * other than http/1.1, this might never happen.
+ */
+ event_conn_state_t *cs;
+
+ cs = ap_get_module_config(c->conn_config, &mpm_event_module);
+ cs->sc = ap_get_module_config(s->module_config, &mpm_event_module);
+ }
+ return DECLINED;
+}
+
/* This really should be a post_config hook, but the error log is already
* redirected by that point, so we need to do this in the open_logs phase.
*/
{
int startup = 0;
int level_flags = 0;
+ int num_buckets = 0;
+ ap_listen_rec **listen_buckets;
apr_status_t rv;
+ int i;
pconf = p;
/* the reverse of pre_config, we want this only the first time around */
- if (retained->module_loads == 1) {
+ if (retained->mpm->module_loads == 1) {
startup = 1;
level_flags |= APLOG_STARTUP;
}
ap_log_error(APLOG_MARK, APLOG_ALERT | level_flags, 0,
(startup ? NULL : s),
"no listening sockets available, shutting down");
- return DONE;
+ return !OK;
}
- if (!one_process) {
- if ((rv = ap_event_pod_open(pconf, &pod))) {
+ if (one_process) {
+ num_buckets = 1;
+ }
+ else if (retained->mpm->was_graceful) {
+ /* Preserve the number of buckets on graceful restarts. */
+ num_buckets = retained->mpm->num_buckets;
+ }
+ if ((rv = ap_duplicate_listeners(pconf, ap_server_conf,
+ &listen_buckets, &num_buckets))) {
+ ap_log_error(APLOG_MARK, APLOG_CRIT | level_flags, rv,
+ (startup ? NULL : s),
+ "could not duplicate listeners");
+ return !OK;
+ }
+
+ all_buckets = apr_pcalloc(pconf, num_buckets * sizeof(*all_buckets));
+ for (i = 0; i < num_buckets; i++) {
+ if (!one_process && /* no POD in one_process mode */
+ (rv = ap_mpm_podx_open(pconf, &all_buckets[i].pod))) {
ap_log_error(APLOG_MARK, APLOG_CRIT | level_flags, rv,
(startup ? NULL : s),
"could not open pipe-of-death");
- return DONE;
+ return !OK;
+ }
+ all_buckets[i].listeners = listen_buckets[i];
+ }
+
+ if (retained->mpm->max_buckets < num_buckets) {
+ int new_max, *new_ptr;
+ new_max = retained->mpm->max_buckets * 2;
+ if (new_max < num_buckets) {
+ new_max = num_buckets;
+ }
+ new_ptr = (int *)apr_palloc(ap_pglobal, new_max * sizeof(int));
+ memcpy(new_ptr, retained->idle_spawn_rate,
+ retained->mpm->num_buckets * sizeof(int));
+ retained->idle_spawn_rate = new_ptr;
+ retained->mpm->max_buckets = new_max;
+ }
+ if (retained->mpm->num_buckets < num_buckets) {
+ int rate_max = 1;
+ /* If new buckets are added, set their idle spawn rate to
+ * the highest so far, so that they get filled as quickly
+ * as the existing ones.
+ */
+ for (i = 0; i < retained->mpm->num_buckets; i++) {
+ if (rate_max < retained->idle_spawn_rate[i]) {
+ rate_max = retained->idle_spawn_rate[i];
+ }
+ }
+ for (/* up to date i */; i < num_buckets; i++) {
+ retained->idle_spawn_rate[i] = rate_max;
}
}
+ retained->mpm->num_buckets = num_buckets;
+
+ /* for skiplist */
+ srand((unsigned int)apr_time_now());
return OK;
}
int no_detach, debug, foreground;
apr_status_t rv;
const char *userdata_key = "mpm_event_module";
-
- mpm_state = AP_MPMQ_STARTING;
+ int test_atomics = 0;
debug = ap_exists_config_define("DEBUG");
foreground = ap_exists_config_define("FOREGROUND");
}
- /* sigh, want this only the second time around */
retained = ap_retained_data_get(userdata_key);
if (!retained) {
retained = ap_retained_data_create(userdata_key, sizeof(*retained));
+ retained->mpm = ap_unixd_mpm_get_retained_data();
+ retained->max_daemons_limit = -1;
+ if (retained->mpm->module_loads) {
+ test_atomics = 1;
+ }
+ }
+ retained->mpm->mpm_state = AP_MPMQ_STARTING;
+ if (retained->mpm->baton != retained) {
+ retained->mpm->was_graceful = 0;
+ retained->mpm->baton = retained;
+ }
+ ++retained->mpm->module_loads;
+
+ /* test once for correct operation of fdqueue */
+ if (test_atomics || retained->mpm->module_loads == 2) {
+ static apr_uint32_t foo1, foo2;
+
+ apr_atomic_set32(&foo1, 100);
+ foo2 = apr_atomic_add32(&foo1, -10);
+ if (foo2 != 100 || foo1 != 90) {
+ ap_log_error(APLOG_MARK, APLOG_CRIT, 0, NULL, APLOGNO(02405)
+ "atomics not working as expected - add32 of negative number");
+ return HTTP_INTERNAL_SERVER_ERROR;
+ }
}
- ++retained->module_loads;
- if (retained->module_loads == 2) {
- is_graceful = 0;
+
+ /* sigh, want this only the second time around */
+ if (retained->mpm->module_loads == 2) {
rv = apr_pollset_create(&event_pollset, 1, plog,
APR_POLLSET_THREADSAFE | APR_POLLSET_NOCOPY);
if (rv != APR_SUCCESS) {
- ap_log_error(APLOG_MARK, APLOG_CRIT, rv, NULL,
+ ap_log_error(APLOG_MARK, APLOG_CRIT, rv, NULL, APLOGNO(00495)
"Couldn't create a Thread Safe Pollset. "
"Is it supported on your platform?"
"Also check system or user limits!");
apr_pollset_destroy(event_pollset);
if (!one_process && !foreground) {
+ /* before we detach, setup crash handlers to log to errorlog */
+ ap_fatal_signal_setup(ap_server_conf, pconf);
rv = apr_proc_detach(no_detach ? APR_PROC_DETACH_FOREGROUND
: APR_PROC_DETACH_DAEMONIZE);
if (rv != APR_SUCCESS) {
- ap_log_error(APLOG_MARK, APLOG_CRIT, rv, NULL,
+ ap_log_error(APLOG_MARK, APLOG_CRIT, rv, NULL, APLOGNO(00496)
"apr_proc_detach failed");
return HTTP_INTERNAL_SERVER_ERROR;
}
}
- parent_pid = ap_my_pid = getpid();
}
+ parent_pid = ap_my_pid = getpid();
+
ap_listen_pre_config();
ap_daemons_to_start = DEFAULT_START_DAEMON;
min_spare_threads = DEFAULT_MIN_FREE_DAEMON * DEFAULT_THREADS_PER_CHILD;
max_spare_threads = DEFAULT_MAX_FREE_DAEMON * DEFAULT_THREADS_PER_CHILD;
server_limit = DEFAULT_SERVER_LIMIT;
thread_limit = DEFAULT_THREAD_LIMIT;
- ap_daemons_limit = server_limit;
+ active_daemons_limit = server_limit;
threads_per_child = DEFAULT_THREADS_PER_CHILD;
- max_clients = ap_daemons_limit * threads_per_child;
- ap_pid_fname = DEFAULT_PIDLOG;
- ap_max_requests_per_child = DEFAULT_MAX_REQUESTS_PER_CHILD;
+ max_workers = active_daemons_limit * threads_per_child;
+ defer_linger_chain = NULL;
+ had_healthy_child = 0;
ap_extended_status = 0;
- ap_max_mem_free = APR_ALLOCATOR_MAX_FREE_UNLIMITED;
- apr_cpystrn(ap_coredump_dir, ap_server_root, sizeof(ap_coredump_dir));
+ return OK;
+}
+
+static int event_post_config(apr_pool_t *pconf, apr_pool_t *plog,
+ apr_pool_t *ptemp, server_rec *s)
+{
+ struct {
+ struct timeout_queue *tail, *q;
+ apr_hash_t *hash;
+ } wc, ka;
+
+ /* Not needed in pre_config stage */
+ if (ap_state_query(AP_SQ_MAIN_STATE) == AP_SQ_MS_CREATE_PRE_CONFIG) {
+ return OK;
+ }
+
+ wc.tail = ka.tail = NULL;
+ wc.hash = apr_hash_make(ptemp);
+ ka.hash = apr_hash_make(ptemp);
+
+ linger_q = TO_QUEUE_MAKE(pconf, apr_time_from_sec(MAX_SECS_TO_LINGER),
+ NULL);
+ short_linger_q = TO_QUEUE_MAKE(pconf, apr_time_from_sec(SECONDS_TO_LINGER),
+ NULL);
+
+ for (; s; s = s->next) {
+ event_srv_cfg *sc = apr_pcalloc(pconf, sizeof *sc);
+
+ ap_set_module_config(s->module_config, &mpm_event_module, sc);
+ if (!wc.tail) {
+ /* The main server uses the global queues */
+ wc.q = TO_QUEUE_MAKE(pconf, s->timeout, NULL);
+ apr_hash_set(wc.hash, &s->timeout, sizeof s->timeout, wc.q);
+ wc.tail = write_completion_q = wc.q;
+
+ ka.q = TO_QUEUE_MAKE(pconf, s->keep_alive_timeout, NULL);
+ apr_hash_set(ka.hash, &s->keep_alive_timeout,
+ sizeof s->keep_alive_timeout, ka.q);
+ ka.tail = keepalive_q = ka.q;
+ }
+ else {
+ /* The vhosts use any existing queue with the same timeout,
+ * or their own queue(s) if there isn't */
+ wc.q = apr_hash_get(wc.hash, &s->timeout, sizeof s->timeout);
+ if (!wc.q) {
+ wc.q = TO_QUEUE_MAKE(pconf, s->timeout, wc.tail);
+ apr_hash_set(wc.hash, &s->timeout, sizeof s->timeout, wc.q);
+ wc.tail = wc.tail->next = wc.q;
+ }
+
+ ka.q = apr_hash_get(ka.hash, &s->keep_alive_timeout,
+ sizeof s->keep_alive_timeout);
+ if (!ka.q) {
+ ka.q = TO_QUEUE_MAKE(pconf, s->keep_alive_timeout, ka.tail);
+ apr_hash_set(ka.hash, &s->keep_alive_timeout,
+ sizeof s->keep_alive_timeout, ka.q);
+ ka.tail = ka.tail->next = ka.q;
+ }
+ }
+ sc->wc_q = wc.q;
+ sc->ka_q = ka.q;
+ }
return OK;
}
int startup = 0;
/* the reverse of pre_config, we want this only the first time around */
- if (retained->module_loads == 1) {
+ if (retained->mpm->module_loads == 1) {
startup = 1;
}
if (server_limit > MAX_SERVER_LIMIT) {
if (startup) {
- ap_log_error(APLOG_MARK, APLOG_WARNING | APLOG_STARTUP, 0, NULL,
+ ap_log_error(APLOG_MARK, APLOG_WARNING | APLOG_STARTUP, 0, NULL, APLOGNO(00497)
"WARNING: ServerLimit of %d exceeds compile-time "
- "limit of", server_limit);
- ap_log_error(APLOG_MARK, APLOG_WARNING | APLOG_STARTUP, 0, NULL,
- " %d servers, decreasing to %d.",
- MAX_SERVER_LIMIT, MAX_SERVER_LIMIT);
+ "limit of %d servers, decreasing to %d.",
+ server_limit, MAX_SERVER_LIMIT, MAX_SERVER_LIMIT);
} else {
- ap_log_error(APLOG_MARK, APLOG_WARNING, 0, s,
+ ap_log_error(APLOG_MARK, APLOG_WARNING, 0, s, APLOGNO(00498)
"ServerLimit of %d exceeds compile-time limit "
"of %d, decreasing to match",
server_limit, MAX_SERVER_LIMIT);
}
else if (server_limit < 1) {
if (startup) {
- ap_log_error(APLOG_MARK, APLOG_WARNING | APLOG_STARTUP, 0, NULL,
+ ap_log_error(APLOG_MARK, APLOG_WARNING | APLOG_STARTUP, 0, NULL, APLOGNO(00499)
"WARNING: ServerLimit of %d not allowed, "
"increasing to 1.", server_limit);
} else {
- ap_log_error(APLOG_MARK, APLOG_WARNING, 0, s,
+ ap_log_error(APLOG_MARK, APLOG_WARNING, 0, s, APLOGNO(00500)
"ServerLimit of %d not allowed, increasing to 1",
server_limit);
}
}
else if (server_limit != retained->first_server_limit) {
/* don't need a startup console version here */
- ap_log_error(APLOG_MARK, APLOG_WARNING, 0, s,
+ ap_log_error(APLOG_MARK, APLOG_WARNING, 0, s, APLOGNO(00501)
"changing ServerLimit to %d from original value of %d "
"not allowed during restart",
server_limit, retained->first_server_limit);
if (thread_limit > MAX_THREAD_LIMIT) {
if (startup) {
- ap_log_error(APLOG_MARK, APLOG_WARNING | APLOG_STARTUP, 0, NULL,
+ ap_log_error(APLOG_MARK, APLOG_WARNING | APLOG_STARTUP, 0, NULL, APLOGNO(00502)
"WARNING: ThreadLimit of %d exceeds compile-time "
- "limit of", thread_limit);
- ap_log_error(APLOG_MARK, APLOG_WARNING | APLOG_STARTUP, 0, NULL,
- " %d threads, decreasing to %d.",
- MAX_THREAD_LIMIT, MAX_THREAD_LIMIT);
+ "limit of %d threads, decreasing to %d.",
+ thread_limit, MAX_THREAD_LIMIT, MAX_THREAD_LIMIT);
} else {
- ap_log_error(APLOG_MARK, APLOG_WARNING, 0, s,
+ ap_log_error(APLOG_MARK, APLOG_WARNING, 0, s, APLOGNO(00503)
"ThreadLimit of %d exceeds compile-time limit "
"of %d, decreasing to match",
thread_limit, MAX_THREAD_LIMIT);
}
else if (thread_limit < 1) {
if (startup) {
- ap_log_error(APLOG_MARK, APLOG_WARNING | APLOG_STARTUP, 0, NULL,
+ ap_log_error(APLOG_MARK, APLOG_WARNING | APLOG_STARTUP, 0, NULL, APLOGNO(00504)
"WARNING: ThreadLimit of %d not allowed, "
"increasing to 1.", thread_limit);
} else {
- ap_log_error(APLOG_MARK, APLOG_WARNING, 0, s,
+ ap_log_error(APLOG_MARK, APLOG_WARNING, 0, s, APLOGNO(00505)
"ThreadLimit of %d not allowed, increasing to 1",
thread_limit);
}
}
else if (thread_limit != retained->first_thread_limit) {
/* don't need a startup console version here */
- ap_log_error(APLOG_MARK, APLOG_WARNING, 0, s,
+ ap_log_error(APLOG_MARK, APLOG_WARNING, 0, s, APLOGNO(00506)
"changing ThreadLimit to %d from original value of %d "
"not allowed during restart",
thread_limit, retained->first_thread_limit);
if (threads_per_child > thread_limit) {
if (startup) {
- ap_log_error(APLOG_MARK, APLOG_WARNING | APLOG_STARTUP, 0, NULL,
+ ap_log_error(APLOG_MARK, APLOG_WARNING | APLOG_STARTUP, 0, NULL, APLOGNO(00507)
"WARNING: ThreadsPerChild of %d exceeds ThreadLimit "
- "of", threads_per_child);
- ap_log_error(APLOG_MARK, APLOG_WARNING | APLOG_STARTUP, 0, NULL,
- " %d threads, decreasing to %d.",
- thread_limit, thread_limit);
- ap_log_error(APLOG_MARK, APLOG_WARNING | APLOG_STARTUP, 0, NULL,
- " To increase, please see the ThreadLimit "
- "directive.");
+ "of %d threads, decreasing to %d. "
+ "To increase, please see the ThreadLimit directive.",
+ threads_per_child, thread_limit, thread_limit);
} else {
- ap_log_error(APLOG_MARK, APLOG_WARNING, 0, s,
+ ap_log_error(APLOG_MARK, APLOG_WARNING, 0, s, APLOGNO(00508)
"ThreadsPerChild of %d exceeds ThreadLimit "
"of %d, decreasing to match",
threads_per_child, thread_limit);
}
else if (threads_per_child < 1) {
if (startup) {
- ap_log_error(APLOG_MARK, APLOG_WARNING | APLOG_STARTUP, 0, NULL,
+ ap_log_error(APLOG_MARK, APLOG_WARNING | APLOG_STARTUP, 0, NULL, APLOGNO(00509)
"WARNING: ThreadsPerChild of %d not allowed, "
"increasing to 1.", threads_per_child);
} else {
- ap_log_error(APLOG_MARK, APLOG_WARNING, 0, s,
+ ap_log_error(APLOG_MARK, APLOG_WARNING, 0, s, APLOGNO(00510)
"ThreadsPerChild of %d not allowed, increasing to 1",
threads_per_child);
}
threads_per_child = 1;
}
- if (max_clients < threads_per_child) {
+ if (max_workers < threads_per_child) {
if (startup) {
- ap_log_error(APLOG_MARK, APLOG_WARNING | APLOG_STARTUP, 0, NULL,
- "WARNING: MaxClients of %d is less than "
- "ThreadsPerChild of", max_clients);
- ap_log_error(APLOG_MARK, APLOG_WARNING | APLOG_STARTUP, 0, NULL,
- " %d, increasing to %d. MaxClients must be at "
- "least as large",
- threads_per_child, threads_per_child);
- ap_log_error(APLOG_MARK, APLOG_WARNING | APLOG_STARTUP, 0, NULL,
- " as the number of threads in a single server.");
+ ap_log_error(APLOG_MARK, APLOG_WARNING | APLOG_STARTUP, 0, NULL, APLOGNO(00511)
+ "WARNING: MaxRequestWorkers of %d is less than "
+ "ThreadsPerChild of %d, increasing to %d. "
+ "MaxRequestWorkers must be at least as large "
+ "as the number of threads in a single server.",
+ max_workers, threads_per_child, threads_per_child);
} else {
- ap_log_error(APLOG_MARK, APLOG_WARNING, 0, s,
- "MaxClients of %d is less than ThreadsPerChild "
+ ap_log_error(APLOG_MARK, APLOG_WARNING, 0, s, APLOGNO(00512)
+ "MaxRequestWorkers of %d is less than ThreadsPerChild "
"of %d, increasing to match",
- max_clients, threads_per_child);
+ max_workers, threads_per_child);
}
- max_clients = threads_per_child;
+ max_workers = threads_per_child;
}
- ap_daemons_limit = max_clients / threads_per_child;
+ active_daemons_limit = max_workers / threads_per_child;
- if (max_clients % threads_per_child) {
- int tmp_max_clients = ap_daemons_limit * threads_per_child;
+ if (max_workers % threads_per_child) {
+ int tmp_max_workers = active_daemons_limit * threads_per_child;
if (startup) {
- ap_log_error(APLOG_MARK, APLOG_WARNING | APLOG_STARTUP, 0, NULL,
- "WARNING: MaxClients of %d is not an integer "
- "multiple of", max_clients);
- ap_log_error(APLOG_MARK, APLOG_WARNING | APLOG_STARTUP, 0, NULL,
- " ThreadsPerChild of %d, decreasing to nearest "
- "multiple %d,", threads_per_child,
- tmp_max_clients);
- ap_log_error(APLOG_MARK, APLOG_WARNING | APLOG_STARTUP, 0, NULL,
- " for a maximum of %d servers.",
- ap_daemons_limit);
+ ap_log_error(APLOG_MARK, APLOG_WARNING | APLOG_STARTUP, 0, NULL, APLOGNO(00513)
+ "WARNING: MaxRequestWorkers of %d is not an integer "
+ "multiple of ThreadsPerChild of %d, decreasing to nearest "
+ "multiple %d, for a maximum of %d servers.",
+ max_workers, threads_per_child, tmp_max_workers,
+ active_daemons_limit);
} else {
- ap_log_error(APLOG_MARK, APLOG_WARNING, 0, s,
- "MaxClients of %d is not an integer multiple of "
- "ThreadsPerChild of %d, decreasing to nearest "
- "multiple %d", max_clients, threads_per_child,
- tmp_max_clients);
+ ap_log_error(APLOG_MARK, APLOG_WARNING, 0, s, APLOGNO(00514)
+ "MaxRequestWorkers of %d is not an integer multiple "
+ "of ThreadsPerChild of %d, decreasing to nearest "
+ "multiple %d", max_workers, threads_per_child,
+ tmp_max_workers);
}
- max_clients = tmp_max_clients;
+ max_workers = tmp_max_workers;
}
- if (ap_daemons_limit > server_limit) {
+ if (active_daemons_limit > server_limit) {
if (startup) {
- ap_log_error(APLOG_MARK, APLOG_WARNING | APLOG_STARTUP, 0, NULL,
- "WARNING: MaxClients of %d would require %d "
- "servers and ", max_clients, ap_daemons_limit);
- ap_log_error(APLOG_MARK, APLOG_WARNING | APLOG_STARTUP, 0, NULL,
- " would exceed ServerLimit of %d, decreasing to %d.",
- server_limit, server_limit * threads_per_child);
- ap_log_error(APLOG_MARK, APLOG_WARNING | APLOG_STARTUP, 0, NULL,
- " To increase, please see the ServerLimit "
- "directive.");
+ ap_log_error(APLOG_MARK, APLOG_WARNING | APLOG_STARTUP, 0, NULL, APLOGNO(00515)
+ "WARNING: MaxRequestWorkers of %d would require %d servers "
+ "and would exceed ServerLimit of %d, decreasing to %d. "
+ "To increase, please see the ServerLimit directive.",
+ max_workers, active_daemons_limit, server_limit,
+ server_limit * threads_per_child);
} else {
- ap_log_error(APLOG_MARK, APLOG_WARNING, 0, s,
- "MaxClients of %d would require %d servers and "
+ ap_log_error(APLOG_MARK, APLOG_WARNING, 0, s, APLOGNO(00516)
+ "MaxRequestWorkers of %d would require %d servers and "
"exceed ServerLimit of %d, decreasing to %d",
- max_clients, ap_daemons_limit, server_limit,
+ max_workers, active_daemons_limit, server_limit,
server_limit * threads_per_child);
}
- ap_daemons_limit = server_limit;
+ active_daemons_limit = server_limit;
}
- /* ap_daemons_to_start > ap_daemons_limit checked in ap_mpm_run() */
- if (ap_daemons_to_start < 0) {
+ /* ap_daemons_to_start > active_daemons_limit checked in ap_mpm_run() */
+ if (ap_daemons_to_start < 1) {
if (startup) {
- ap_log_error(APLOG_MARK, APLOG_WARNING | APLOG_STARTUP, 0, NULL,
+ ap_log_error(APLOG_MARK, APLOG_WARNING | APLOG_STARTUP, 0, NULL, APLOGNO(00517)
"WARNING: StartServers of %d not allowed, "
"increasing to 1.", ap_daemons_to_start);
} else {
- ap_log_error(APLOG_MARK, APLOG_WARNING, 0, s,
+ ap_log_error(APLOG_MARK, APLOG_WARNING, 0, s, APLOGNO(00518)
"StartServers of %d not allowed, increasing to 1",
ap_daemons_to_start);
}
if (min_spare_threads < 1) {
if (startup) {
- ap_log_error(APLOG_MARK, APLOG_WARNING | APLOG_STARTUP, 0, NULL,
+ ap_log_error(APLOG_MARK, APLOG_WARNING | APLOG_STARTUP, 0, NULL, APLOGNO(00519)
"WARNING: MinSpareThreads of %d not allowed, "
- "increasing to 1", min_spare_threads);
- ap_log_error(APLOG_MARK, APLOG_WARNING | APLOG_STARTUP, 0, NULL,
- " to avoid almost certain server failure.");
- ap_log_error(APLOG_MARK, APLOG_WARNING | APLOG_STARTUP, 0, NULL,
- " Please read the documentation.");
+ "increasing to 1 to avoid almost certain server "
+ "failure. Please read the documentation.",
+ min_spare_threads);
} else {
- ap_log_error(APLOG_MARK, APLOG_WARNING, 0, s,
+ ap_log_error(APLOG_MARK, APLOG_WARNING, 0, s, APLOGNO(00520)
"MinSpareThreads of %d not allowed, increasing to 1",
min_spare_threads);
}
* to retrieve it, so register as REALLY_FIRST
*/
ap_hook_pre_config(event_pre_config, NULL, NULL, APR_HOOK_REALLY_FIRST);
+ ap_hook_post_config(event_post_config, NULL, NULL, APR_HOOK_MIDDLE);
ap_hook_check_config(event_check_config, NULL, NULL, APR_HOOK_MIDDLE);
ap_hook_mpm(event_run, NULL, NULL, APR_HOOK_MIDDLE);
ap_hook_mpm_query(event_query, NULL, NULL, APR_HOOK_MIDDLE);
- ap_hook_mpm_note_child_killed(event_note_child_killed, NULL, NULL, APR_HOOK_MIDDLE);
ap_hook_mpm_register_timed_callback(event_register_timed_callback, NULL, NULL,
APR_HOOK_MIDDLE);
+ ap_hook_pre_read_request(event_pre_read_request, NULL, NULL, APR_HOOK_MIDDLE);
+ ap_hook_post_read_request(event_post_read_request, NULL, NULL, APR_HOOK_MIDDLE);
ap_hook_mpm_get_name(event_get_name, NULL, NULL, APR_HOOK_MIDDLE);
+
+ ap_hook_pre_connection(event_pre_connection, NULL, NULL, APR_HOOK_REALLY_FIRST);
+ ap_hook_protocol_switch(event_protocol_switch, NULL, NULL, APR_HOOK_REALLY_FIRST);
}
static const char *set_daemons_to_start(cmd_parms *cmd, void *dummy,
return NULL;
}
-static const char *set_max_clients(cmd_parms * cmd, void *dummy,
+static const char *set_max_workers(cmd_parms * cmd, void *dummy,
const char *arg)
{
const char *err = ap_check_cmd_context(cmd, GLOBAL_ONLY);
if (err != NULL) {
return err;
}
-
- max_clients = atoi(arg);
+ if (!strcasecmp(cmd->cmd->name, "MaxClients")) {
+ ap_log_error(APLOG_MARK, APLOG_INFO, 0, NULL, APLOGNO(00521)
+ "MaxClients is deprecated, use MaxRequestWorkers "
+ "instead.");
+ }
+ max_workers = atoi(arg);
return NULL;
}
return NULL;
}
+static const char *set_worker_factor(cmd_parms * cmd, void *dummy,
+ const char *arg)
+{
+ double val;
+ char *endptr;
+ const char *err = ap_check_cmd_context(cmd, GLOBAL_ONLY);
+ if (err != NULL) {
+ return err;
+ }
+
+ val = strtod(arg, &endptr);
+ if (*endptr)
+ return "error parsing value";
+
+ if (val <= 0)
+ return "AsyncRequestWorkerFactor argument must be a positive number";
+
+ worker_factor = val * WORKER_FACTOR_SCALE;
+ if (worker_factor == 0)
+ worker_factor = 1;
+ return NULL;
+}
+
+
static const command_rec event_cmds[] = {
LISTEN_COMMANDS,
AP_INIT_TAKE1("StartServers", set_daemons_to_start, NULL, RSRC_CONF,
"Minimum number of idle threads, to handle request spikes"),
AP_INIT_TAKE1("MaxSpareThreads", set_max_spare_threads, NULL, RSRC_CONF,
"Maximum number of idle threads"),
- AP_INIT_TAKE1("MaxClients", set_max_clients, NULL, RSRC_CONF,
+ AP_INIT_TAKE1("MaxClients", set_max_workers, NULL, RSRC_CONF,
+ "Deprecated name of MaxRequestWorkers"),
+ AP_INIT_TAKE1("MaxRequestWorkers", set_max_workers, NULL, RSRC_CONF,
"Maximum number of threads alive at the same time"),
AP_INIT_TAKE1("ThreadsPerChild", set_threads_per_child, NULL, RSRC_CONF,
"Number of threads each child creates"),
AP_INIT_TAKE1("ThreadLimit", set_thread_limit, NULL, RSRC_CONF,
"Maximum number of worker threads per child process for this "
"run of Apache - Upper limit for ThreadsPerChild"),
+ AP_INIT_TAKE1("AsyncRequestWorkerFactor", set_worker_factor, NULL, RSRC_CONF,
+ "How many additional connects will be accepted per idle "
+ "worker thread"),
AP_GRACEFUL_SHUTDOWN_TIMEOUT_COMMAND,
{NULL}
};