threadpool.c

Apache 2.0.63 is the current stable version of the 2.0 series, and is recommended over any previous

    worker_score *ws;
    process_score *ps;
    int free_length;
    int totally_free_length = 0;
    int free_slots[MAX_SPAWN_RATE];
    int last_non_dead;
    int total_non_dead;

    /* 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) {
        /* Initialization to satisfy the compiler. It doesn't know
         * that ap_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;

        if (i >= ap_max_daemons_limit && totally_free_length == idle_spawn_rate)
            break;
        ps = &ap_scoreboard_image->parent[i];
        for (j = 0; j < ap_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 (status <= SERVER_READY && status != SERVER_DEAD &&
                    !ps->quiescing &&
                    ps->generation == ap_my_generation &&
                 /* XXX the following shouldn't be necessary if we clean up 
                  *     properly after seg faults, but we're not yet    GLA 
                  */     
                    ps->pid != 0) {
                ++idle_thread_count;
            }
        }
        if (any_dead_threads && totally_free_length < idle_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;
        }
    }
    ap_max_daemons_limit = last_non_dead + 1;

    if (idle_thread_count > max_spare_threads) {
        /* Kill off one child */
        ap_mpm_pod_signal(pod, TRUE);
        idle_spawn_rate = 1;
    }
    else if (idle_thread_count < min_spare_threads) {
        /* terminate the free list */
        if (free_length == 0) {
            /* only report this condition once */
            static int reported = 0;
            
            if (!reported) {
                ap_log_error(APLOG_MARK, APLOG_ERR, 0, 
                             ap_server_conf,
                             "server reached MaxClients setting, consider"
                             " raising the MaxClients setting");
                reported = 1;
            }
            idle_spawn_rate = 1;
        }
        else {
            if (free_length > idle_spawn_rate) {
                free_length = idle_spawn_rate;
            }
            if (idle_spawn_rate >= 8) {
                ap_log_error(APLOG_MARK, APLOG_INFO, 0, 
                             ap_server_conf,
                             "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);
            }
            for (i = 0; i < free_length; ++i) {
                make_child(ap_server_conf, free_slots[i]);
            }
            /* 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;
            }
            else if (idle_spawn_rate < MAX_SPAWN_RATE) {
                idle_spawn_rate *= 2;
            }
        }
    }
    else {
      idle_spawn_rate = 1;
    }
}

static void server_main_loop(int remaining_children_to_start)
{
    int child_slot;
    apr_exit_why_e exitwhy;
    int status, processed_status;
    apr_proc_t pid;
    int i;

    while (!restart_pending && !shutdown_pending) {
        ap_wait_or_timeout(&exitwhy, &status, &pid, pconf);
        
        if (pid.pid != -1) {
            processed_status = ap_process_child_status(&pid, exitwhy, status);
            if (processed_status == APEXIT_CHILDFATAL) {
                shutdown_pending = 1;
                child_fatal = 1;
                return;
            }
            /* non-fatal death... note that it's gone in the scoreboard. */
            child_slot = find_child_by_pid(&pid);
            if (child_slot >= 0) {
                for (i = 0; i < ap_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;
                if (processed_status == APEXIT_CHILDSICK) {
                    /* resource shortage, minimize the fork rate */
                    idle_spawn_rate = 1;
                }
                else if (remaining_children_to_start
                    && child_slot < ap_daemons_limit) {
                    /* we're still doing a 1-for-1 replacement of dead
                     * children with new children
                     */
                    make_child(ap_server_conf, child_slot);
                    --remaining_children_to_start;
                }
#if APR_HAS_OTHER_CHILD
            }
            else if (apr_proc_other_child_read(&pid, status) == 0) {
                /* handled */
#endif
            }
            else if (is_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,
                             "long lost child came home! (pid %ld)",
                             (long)pid.pid);
            }
            /* Don't perform idle maintenance when a child dies,
             * only do it when there's a timeout.  Remember only a
             * finite number of children can die, and it's pretty
             * pathological for a lot to die suddenly.
             */
            continue;
        }
        else if (remaining_children_to_start) {
            /* we hit a 1 second timeout in which none of the previous
             * generation of children needed to be reaped... so assume
             * they're all done, and pick up the slack if any is left.
             */
            startup_children(remaining_children_to_start);
            remaining_children_to_start = 0;
            /* In any event we really shouldn't do the code below because
             * few of the servers we just started are in the IDLE state
             * yet, so we'd mistakenly create an extra server.
             */
            continue;
        }

        perform_idle_server_maintenance();
    }
}

int ap_mpm_run(apr_pool_t *_pconf, apr_pool_t *plog, server_rec *s)
{
    int remaining_children_to_start;
    apr_status_t rv;

    ap_log_pid(pconf, ap_pid_fname);

    first_server_limit = server_limit;
    first_thread_limit = thread_limit;
    if (changed_limit_at_restart) {
        ap_log_error(APLOG_MARK, APLOG_WARNING, 0, s,
                     "WARNING: Attempt to change ServerLimit or ThreadLimit "
                     "ignored during restart");
        changed_limit_at_restart = 0;
    }
    
    /* Initialize cross-process accept lock */
    ap_lock_fname = apr_psprintf(_pconf, "%s.%" APR_PID_T_FMT,
                                 ap_server_root_relative(_pconf, ap_lock_fname),
                                 ap_my_pid);

    rv = apr_proc_mutex_create(&accept_mutex, ap_lock_fname, 
                               ap_accept_lock_mech, _pconf);
    if (rv != APR_SUCCESS) {
        ap_log_error(APLOG_MARK, APLOG_EMERG, rv, s,
                     "Couldn't create accept lock");
        mpm_state = AP_MPMQ_STOPPING;
        return 1;
    }

#if APR_USE_SYSVSEM_SERIALIZE
    if (ap_accept_lock_mech == APR_LOCK_DEFAULT || 
        ap_accept_lock_mech == APR_LOCK_SYSVSEM) {
#else
    if (ap_accept_lock_mech == APR_LOCK_SYSVSEM) {
#endif
        rv = unixd_set_proc_mutex_perms(accept_mutex);
        if (rv != APR_SUCCESS) {
            ap_log_error(APLOG_MARK, APLOG_EMERG, rv, s,
                         "Couldn't set permissions on cross-process lock; "
                         "check User and Group directives");
            mpm_state = AP_MPMQ_STOPPING;
            return 1;
        }
    }

    if (!is_graceful) {
        if (ap_run_pre_mpm(s->process->pool, SB_SHARED) != OK) {
            mpm_state = AP_MPMQ_STOPPING;
            return 1;
        }
        /* fix the generation number in the global score; we just got a new,
         * cleared scoreboard
         */
        ap_scoreboard_image->global->running_generation = ap_my_generation;
    }

    set_signals();
    /* Don't thrash... */
    if (max_spare_threads < min_spare_threads + ap_threads_per_child)
        max_spare_threads = min_spare_threads + ap_threads_per_child;

    /* 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
     * below (because we just sent them AP_SIG_GRACEFUL).  This happens pretty
     * rapidly... and for each one that exits we'll start a new one until
     * we reach at least daemons_min_free.  But we may be permitted to
     * start more than that, so we'll just keep track of how many we're
     * 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 (!is_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;
    }

    ap_log_error(APLOG_MARK, APLOG_NOTICE, 0, ap_server_conf,
                "%s configured -- resuming normal operations",
                ap_get_server_version());
    ap_log_error(APLOG_MARK, APLOG_INFO, 0, ap_server_conf,
                "Server built: %s", ap_get_server_built());
#ifdef AP_MPM_WANT_SET_ACCEPT_LOCK_MECH
    ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, ap_server_conf,
		"AcceptMutex: %s (default: %s)",
		apr_proc_mutex_name(accept_mutex),
		apr_proc_mutex_defname());
#endif
    restart_pending = shutdown_pending = 0;
    mpm_state = AP_MPMQ_RUNNING;

    server_main_loop(remaining_children_to_start);
    mpm_state = AP_MPMQ_STOPPING;

    if (shutdown_pending) {
        /* Time to gracefully shut down:
         * Kill child processes, tell them to call child_exit, etc...
         * (By "gracefully" we don't mean graceful in the same sense as 
         * "apachectl graceful" where we allow old connections to finish.)
         */
        ap_mpm_pod_killpg(pod, ap_daemons_limit, FALSE);
        ap_reclaim_child_processes(1);                /* Start with SIGTERM */

        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, "caught SIGTERM, shutting down");
        }
        return 1;
    }

    /* we've been told to restart */
    apr_signal(SIGHUP, SIG_IGN);

    if (one_process) {
        /* not worth thinking about */
        return 1;
    }

    /* advance to the next generation */
    /* XXX: we really need to make sure this new generation number isn't in
     * use by any of the children.
     */
    ++ap_my_generation;
    ap_scoreboard_image->global->running_generation = ap_my_generation;
    
    if (is_graceful) {
        ap_log_error(APLOG_MARK, APLOG_NOTICE, 0, ap_server_conf,
                     AP_SIG_GRACEFUL_STRING " received.  Doing graceful restart");
        /* wake up the children...time to die.  But we'll have more so