The Event MPM.

Designed to minimize Apache's KeepAlive overhead.

This MPM depends on the current APR-trunk for new features added to 
the apr_pollset interface. Currently the underlying operating
system must support KQueue or EPoll.

Status:
  Should work as a drop in replacement for all non-ssl servers.
  SSL Requests that use HTTP 1.1 Pipelining do not currently work.

Testing:
  I have tested it with Linux 2.6, FreeBSD 5.2.1, and OS X 10.3.
  
Originally based on the patch by Greg Ames.


git-svn-id: https://svn.apache.org/repos/asf/httpd/httpd/trunk@105919 13f79535-47bb-0310-9956-ffa450edef68

include/ap_mpm.h

@@ -139,6 +139,7 @@ AP_DECLARE(apr_status_t) ap_os_create_privileged_process(
 #define AP_MPMQ_MAX_REQUESTS_DAEMON  11  /* Max # of requests per daemon */
 #define AP_MPMQ_MAX_DAEMONS          12  /* Max # of daemons by config   */
 #define AP_MPMQ_MPM_STATE            13  /* starting, running, stopping  */
+#define AP_MPMQ_IS_ASYNC             14  /* MPM can process async connections  */
 
 /**
  * Query a property of the current MPM.  

include/httpd.h

@@ -37,6 +37,7 @@
 #include "apr_time.h"
 #include "apr_network_io.h"
 #include "apr_buckets.h"
+#include "apr_poll.h"
 
 #include "os.h"
 
@@ -675,6 +676,8 @@ typedef struct server_rec server_rec;
 typedef struct conn_rec conn_rec;
 /** A structure that represents the current request */
 typedef struct request_rec request_rec;
+/** A structure that represents the status of the current connection */
+typedef struct conn_state_t conn_state_t;
 
 /* ### would be nice to not include this from httpd.h ... */
 /* This comes after we have defined the request_rec type */
@@ -1011,6 +1014,26 @@ struct conn_rec {
     void *sbh;
     /** The bucket allocator to use for all bucket/brigade creations */
     struct apr_bucket_alloc_t *bucket_alloc;
+    /** The current state of this connection */
+    conn_state_t *cs;
+    /** Is there data pending in the input filters? */ 
+    int data_in_input_filters;
+};
+
+typedef enum  {
+    CONN_STATE_CHECK_REQUEST_LINE_READABLE,
+    CONN_STATE_READ_REQUEST_LINE,
+    CONN_STATE_LINGER,
+} conn_state_e;
+
+struct conn_state_t {
+    APR_RING_ENTRY(conn_state_t) timeout_list; 
+    apr_time_t expiration_time; 
+    conn_state_e state;
+    conn_rec *c;
+    apr_pool_t *p;
+    apr_bucket_alloc_t *bucket_alloc;
+    apr_pollfd_t pfd;
 };
 
 /* Per-vhost config... */

modules/http/http_core.c

@@ -231,6 +231,53 @@ static const char *http_method(const request_rec *r)
 static apr_port_t http_port(const request_rec *r)
     { return DEFAULT_HTTP_PORT; }
 
+static int ap_process_http_async_connection(conn_rec *c)
+{
+    request_rec *r;
+    conn_state_t *cs = c->cs;
+
+    switch (cs->state) {
+    case CONN_STATE_READ_REQUEST_LINE:
+        ap_update_child_status(c->sbh, SERVER_BUSY_READ, NULL);
+        while ((r = ap_read_request(c)) != NULL) {
+            c->keepalive = AP_CONN_UNKNOWN;
+
+            /* process the request if it was read without error */
+            ap_update_child_status(c->sbh, SERVER_BUSY_WRITE, r);
+
+            if (r->status == HTTP_OK)
+                ap_process_request(r);
+
+            if (ap_extended_status)
+                ap_increment_counts(c->sbh, r);
+
+            if (c->keepalive != AP_CONN_KEEPALIVE || c->aborted) {
+                cs->state = CONN_STATE_LINGER;
+                break;
+            }
+            else {
+                cs->state = CONN_STATE_CHECK_REQUEST_LINE_READABLE;
+            }
+
+            apr_pool_destroy(r->pool);
+
+            if (ap_graceful_stop_signalled())
+                break;
+
+            if (c->data_in_input_filters) {
+                continue;
+            }
+            break;
+        }
+	break;
+    default:
+        AP_DEBUG_ASSERT(0);
+        cs->state = CONN_STATE_LINGER;
+    }
+
+    return OK;
+}
+
 static int ap_process_http_connection(conn_rec *c)
 {
     request_rec *r;
@@ -290,8 +337,21 @@ static int http_create_request(request_rec *r)
 
 static void register_hooks(apr_pool_t *p)
 {
+    /**
+     * If we ae using an MPM That Supports Async Connections,
+     * use a different processing function
+     */
+    int async_mpm = 0;
+    if(ap_mpm_query(AP_MPMQ_IS_ASYNC, &async_mpm) == APR_SUCCESS 
+       && async_mpm == 1) {
+        ap_hook_process_connection(ap_process_http_async_connection,NULL,
+            NULL,APR_HOOK_REALLY_LAST);
+    }
+    else {
     ap_hook_process_connection(ap_process_http_connection,NULL,NULL,
 			       APR_HOOK_REALLY_LAST);
+    }
+
     ap_hook_map_to_storage(ap_send_http_trace,NULL,NULL,APR_HOOK_MIDDLE);
     ap_hook_http_method(http_method,NULL,NULL,APR_HOOK_REALLY_LAST);
     ap_hook_default_port(http_port,NULL,NULL,APR_HOOK_REALLY_LAST);

modules/http/http_request.c

@@ -205,9 +205,17 @@ static void check_pipeline_flush(request_rec *r)
      */
     /* ### shouldn't this read from the connection input filters? */
     /* ### is zero correct? that means "read one line" */
-    if (r->connection->keepalive == AP_CONN_CLOSE || 
+    if (r->connection->keepalive != AP_CONN_CLOSE) {
-        ap_get_brigade(r->input_filters, bb, AP_MODE_EATCRLF, 
+        if (ap_get_brigade(r->input_filters, bb, AP_MODE_EATCRLF, 
                        APR_NONBLOCK_READ, 0) != APR_SUCCESS) {
+            c->data_in_input_filters = 0;  /* we got APR_EOF or an error */
+        }
+        else {
+            c->data_in_input_filters = 1;
+            return;    /* don't flush */
+        }
+    }
+
         apr_bucket *e = apr_bucket_flush_create(c->bucket_alloc);
 
         /* We just send directly to the connection based filters.  At
@@ -218,7 +226,6 @@ static void check_pipeline_flush(request_rec *r)
          */
         APR_BRIGADE_INSERT_HEAD(bb, e);
         ap_pass_brigade(r->connection->output_filters, bb);
-    }
 }
 
 void ap_process_request(request_rec *r)

server/mpm/config.m4

@@ -1,7 +1,7 @@
 AC_MSG_CHECKING(which MPM to use)
 AC_ARG_WITH(mpm,
 APACHE_HELP_STRING(--with-mpm=MPM,Choose the process model for Apache to use.
-                          MPM={beos|worker|prefork|mpmt_os2|perchild|leader|threadpool}),[
+                          MPM={beos|event|worker|prefork|mpmt_os2|perchild|leader|threadpool}),[
   APACHE_MPM=$withval
 ],[
   if test "x$APACHE_MPM" = "x"; then
@@ -12,7 +12,7 @@ AC_MSG_RESULT($APACHE_MPM)
 
 apache_cv_mpm=$APACHE_MPM
 	
-if test "$apache_cv_mpm" = "worker" -o "$apache_cv_mpm" = "perchild" -o "$apache_cv_mpm" = "leader" -o "$apache_cv_mpm" = "threadpool" ; then
+if test "$apache_cv_mpm" = "worker" -o "$apache_cv_mpm" = "event" -o "$apache_cv_mpm" = "perchild" -o "$apache_cv_mpm" = "leader" -o "$apache_cv_mpm" = "threadpool" ; then
   APR_CHECK_APR_DEFINE(APR_HAS_THREADS)
 
   if test "x$ac_cv_define_APR_HAS_THREADS" = "xno"; then
@@ -26,7 +26,7 @@ fi
 APACHE_FAST_OUTPUT(server/mpm/Makefile)
 
 MPM_NAME=$apache_cv_mpm
-if test "$MPM_NAME" = "leader" -o "$MPM_NAME" = "threadpool" -o "$MPM_NAME" = "perchild"; then
+if test "$MPM_NAME" = "event" -o "$MPM_NAME" = "leader" -o "$MPM_NAME" = "threadpool" -o "$MPM_NAME" = "perchild"; then
   AC_MSG_WARN(You have selected an EXPERIMENTAL MPM.  Be warned!)
   MPM_SUBDIR_NAME=experimental/$MPM_NAME
 else

server/mpm/experimental/event/Makefile.in

@@ -0,0 +1,5 @@
+
+LTLIBRARY_NAME    = libevent.la
+LTLIBRARY_SOURCES = event.c fdqueue.c pod.c
+
+include $(top_srcdir)/build/ltlib.mk

server/mpm/experimental/event/config5.m4

@@ -0,0 +1,6 @@
+dnl ## XXX - Need a more thorough check of the proper flags to use
+
+if test "$MPM_NAME" = "event" ; then
+    AC_CHECK_FUNCS(pthread_kill)
+    APACHE_FAST_OUTPUT(server/mpm/$MPM_SUBDIR_NAME/Makefile)
+fi

server/mpm/experimental/event/fdqueue.c

@@ -0,0 +1,418 @@
+/* Copyright 2001-2004 The Apache Software Foundation
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "fdqueue.h"
+#include "apr_atomic.h"
+
+typedef struct recycled_pool
+{
+    apr_pool_t *pool;
+    struct recycled_pool *next;
+} recycled_pool;
+
+struct fd_queue_info_t
+{
+    apr_int32_t idlers;      /**
+	                          * 0 or positive: number of idle worker threads
+                              * negative: number of threads blocked waiting
+                              *           for an idle worker
+                              */
+    apr_thread_mutex_t *idlers_mutex;
+    apr_thread_cond_t *wait_for_idler;
+    int terminated;
+    int max_idlers;
+    recycled_pool *recycled_pools;
+};
+
+static apr_status_t queue_info_cleanup(void *data_)
+{
+    fd_queue_info_t *qi = data_;
+    apr_thread_cond_destroy(qi->wait_for_idler);
+    apr_thread_mutex_destroy(qi->idlers_mutex);
+
+    /* Clean up any pools in the recycled list */
+    for (;;) {
+        struct recycled_pool *first_pool = qi->recycled_pools;
+        if (first_pool == NULL) {
+            break;
+        }
+        if (apr_atomic_casptr
+            ((volatile void **) &(qi->recycled_pools), first_pool->next,
+             first_pool) == first_pool) {
+            apr_pool_destroy(first_pool->pool);
+        }
+    }
+
+    return APR_SUCCESS;
+}
+
+apr_status_t ap_queue_info_create(fd_queue_info_t ** queue_info,
+                                  apr_pool_t * pool, int max_idlers)
+{
+    apr_status_t rv;
+    fd_queue_info_t *qi;
+
+    qi = apr_palloc(pool, sizeof(*qi));
+    memset(qi, 0, sizeof(*qi));
+
+    rv = apr_thread_mutex_create(&qi->idlers_mutex, APR_THREAD_MUTEX_DEFAULT,
+                                 pool);
+    if (rv != APR_SUCCESS) {
+        return rv;
+    }
+    rv = apr_thread_cond_create(&qi->wait_for_idler, pool);
+    if (rv != APR_SUCCESS) {
+        return rv;
+    }
+    qi->recycled_pools = NULL;
+    qi->max_idlers = max_idlers;
+    apr_pool_cleanup_register(pool, qi, queue_info_cleanup,
+                              apr_pool_cleanup_null);
+
+    *queue_info = qi;
+
+    return APR_SUCCESS;
+}
+
+apr_status_t ap_queue_info_set_idle(fd_queue_info_t * queue_info,
+                                    apr_pool_t * pool_to_recycle)
+{
+    apr_status_t rv;
+    int prev_idlers;
+
+    ap_push_pool(queue_info, pool_to_recycle);
+
+    /* Atomically increment the count of idle workers */
+    prev_idlers = apr_atomic_inc32(&(queue_info->idlers));
+
+    /* If other threads are waiting on a worker, wake one up */
+    if (prev_idlers < 0) {
+        rv = apr_thread_mutex_lock(queue_info->idlers_mutex);
+        if (rv != APR_SUCCESS) {
+            AP_DEBUG_ASSERT(0);
+            return rv;
+        }
+        rv = apr_thread_cond_signal(queue_info->wait_for_idler);
+        if (rv != APR_SUCCESS) {
+            apr_thread_mutex_unlock(queue_info->idlers_mutex);
+            return rv;
+        }
+        rv = apr_thread_mutex_unlock(queue_info->idlers_mutex);
+        if (rv != APR_SUCCESS) {
+            return rv;
+        }
+    }
+
+    return APR_SUCCESS;
+}
+
+apr_status_t ap_queue_info_wait_for_idler(fd_queue_info_t * queue_info)
+{
+    apr_status_t rv;
+    int prev_idlers;
+
+    /* Atomically decrement the idle worker count, saving the old value */
+    prev_idlers = apr_atomic_add32(&(queue_info->idlers), -1);
+
+    /* Block if there weren't any idle workers */
+    if (prev_idlers <= 0) {
+        rv = apr_thread_mutex_lock(queue_info->idlers_mutex);
+        if (rv != APR_SUCCESS) {
+            AP_DEBUG_ASSERT(0);
+            apr_atomic_inc32(&(queue_info->idlers));    /* back out dec */
+            return rv;
+        }
+        /* Re-check the idle worker count to guard against a
+         * race condition.  Now that we're in the mutex-protected
+         * region, one of two things may have happened:
+         *   - If the idle worker count is still negative, the
+         *     workers are all still busy, so it's safe to
+         *     block on a condition variable.
+         *   - If the idle worker count is non-negative, then a
+         *     worker has become idle since the first check
+         *     of queue_info->idlers above.  It's possible
+         *     that the worker has also signaled the condition
+         *     variable--and if so, the listener missed it
+         *     because it wasn't yet blocked on the condition
+         *     variable.  But if the idle worker count is
+         *     now non-negative, it's safe for this function to
+         *     return immediately.
+         *
+         *     A negative value in queue_info->idlers tells how many
+         *     threads are waiting on an idle worker.
+         */
+        if (queue_info->idlers < 0) {
+            rv = apr_thread_cond_wait(queue_info->wait_for_idler,
+                                      queue_info->idlers_mutex);
+            if (rv != APR_SUCCESS) {
+                apr_status_t rv2;
+                AP_DEBUG_ASSERT(0);
+                rv2 = apr_thread_mutex_unlock(queue_info->idlers_mutex);
+                if (rv2 != APR_SUCCESS) {
+                    return rv2;
+                }
+                return rv;
+            }
+        }
+        rv = apr_thread_mutex_unlock(queue_info->idlers_mutex);
+        if (rv != APR_SUCCESS) {
+            return rv;
+        }
+    }
+
+    if (queue_info->terminated) {
+        return APR_EOF;
+    }
+    else {
+        return APR_SUCCESS;
+    }
+}
+
+
+void ap_push_pool(fd_queue_info_t * queue_info,
+                                    apr_pool_t * pool_to_recycle)
+{
+    /* If we have been given a pool to recycle, atomically link
+     * it into the queue_info's list of recycled pools
+     */
+    if (pool_to_recycle) {
+        struct recycled_pool *new_recycle;
+        new_recycle = (struct recycled_pool *) apr_palloc(pool_to_recycle,
+                                                          sizeof
+                                                          (*new_recycle));
+        new_recycle->pool = pool_to_recycle;
+        for (;;) {
+            new_recycle->next = queue_info->recycled_pools;
+            if (apr_atomic_casptr
+                ((volatile void **) &(queue_info->recycled_pools),
+                 new_recycle, new_recycle->next) == new_recycle->next) {
+                break;
+            }
+        }
+    }
+}
+
+void ap_pop_pool(apr_pool_t ** recycled_pool, fd_queue_info_t * queue_info)
+{
+    /* Atomically pop a pool from the recycled list */
+
+    /* This function is safe only as long as it is single threaded because
+     * it reaches into the queue and accesses "next" which can change.
+     * We are OK today because it is only called from the listener thread.
+     * cas-based pushes do not have the same limitation - any number can
+     * happen concurrently with a single cas-based pop.
+     */
+
+    *recycled_pool = NULL;
+
+
+    /* Atomically pop a pool from the recycled list */
+    for (;;) {
+        struct recycled_pool *first_pool = queue_info->recycled_pools;
+        if (first_pool == NULL) {
+            break;
+        }
+        if (apr_atomic_casptr
+            ((volatile void **) &(queue_info->recycled_pools),
+             first_pool->next, first_pool) == first_pool) {
+            *recycled_pool = first_pool->pool;
+            break;
+        }
+    }
+}
+
+apr_status_t ap_queue_info_term(fd_queue_info_t * queue_info)
+{
+    apr_status_t rv;
+    rv = apr_thread_mutex_lock(queue_info->idlers_mutex);
+    if (rv != APR_SUCCESS) {
+        return rv;
+    }
+    queue_info->terminated = 1;
+    apr_thread_cond_broadcast(queue_info->wait_for_idler);
+    return apr_thread_mutex_unlock(queue_info->idlers_mutex);
+}
+
+/**
+ * Detects when the fd_queue_t is full. This utility function is expected
+ * to be called from within critical sections, and is not threadsafe.
+ */
+#define ap_queue_full(queue) ((queue)->nelts == (queue)->bounds)
+
+/**
+ * Detects when the fd_queue_t is empty. This utility function is expected
+ * to be called from within critical sections, and is not threadsafe.
+ */
+#define ap_queue_empty(queue) ((queue)->nelts == 0)
+
+/**
+ * Callback routine that is called to destroy this
+ * fd_queue_t when its pool is destroyed.
+ */
+static apr_status_t ap_queue_destroy(void *data)
+{
+    fd_queue_t *queue = data;
+
+    /* Ignore errors here, we can't do anything about them anyway.
+     * XXX: We should at least try to signal an error here, it is
+     * indicative of a programmer error. -aaron */
+    apr_thread_cond_destroy(queue->not_empty);
+    apr_thread_mutex_destroy(queue->one_big_mutex);
+
+    return APR_SUCCESS;
+}
+
+/**
+ * Initialize the fd_queue_t.
+ */
+apr_status_t ap_queue_init(fd_queue_t * queue, int queue_capacity,
+                           apr_pool_t * a)
+{
+    int i;
+    apr_status_t rv;
+
+    if ((rv = apr_thread_mutex_create(&queue->one_big_mutex,
+                                      APR_THREAD_MUTEX_DEFAULT,
+                                      a)) != APR_SUCCESS) {
+        return rv;
+    }
+    if ((rv = apr_thread_cond_create(&queue->not_empty, a)) != APR_SUCCESS) {
+        return rv;
+    }
+
+    queue->data = apr_palloc(a, queue_capacity * sizeof(fd_queue_elem_t));
+    queue->bounds = queue_capacity;
+    queue->nelts = 0;
+
+    /* Set all the sockets in the queue to NULL */
+    for (i = 0; i < queue_capacity; ++i)
+        queue->data[i].sd = NULL;
+
+    apr_pool_cleanup_register(a, queue, ap_queue_destroy,
+                              apr_pool_cleanup_null);
+
+    return APR_SUCCESS;
+}
+
+/**
+ * Push a new socket onto the queue. Blocks if the queue is full. Once
+ * the push operation has completed, it signals other threads waiting
+ * in ap_queue_pop() that they may continue consuming sockets.
+ */
+apr_status_t ap_queue_push(fd_queue_t * queue, apr_socket_t * sd,
+                           conn_state_t * cs, apr_pool_t * p)
+{
+    fd_queue_elem_t *elem;
+    apr_status_t rv;
+
+    if ((rv = apr_thread_mutex_lock(queue->one_big_mutex)) != APR_SUCCESS) {
+        return rv;
+    }
+
+    AP_DEBUG_ASSERT(!queue->terminated);
+    AP_DEBUG_ASSERT(!ap_queue_full(queue));
+
+    elem = &queue->data[queue->nelts];
+    elem->sd = sd;
+    elem->cs = cs;
+    elem->p = p;
+    queue->nelts++;
+
+    apr_thread_cond_signal(queue->not_empty);
+
+    if ((rv = apr_thread_mutex_unlock(queue->one_big_mutex)) != APR_SUCCESS) {
+        return rv;
+    }
+
+    return APR_SUCCESS;
+}
+
+/**
+ * Retrieves the next available socket from the queue. If there are no
+ * sockets available, it will block until one becomes available.
+ * Once retrieved, the socket is placed into the address specified by
+ * 'sd'.
+ */
+apr_status_t ap_queue_pop(fd_queue_t * queue, apr_socket_t ** sd,
+                          conn_state_t ** cs, apr_pool_t ** p)
+{
+    fd_queue_elem_t *elem;
+    apr_status_t rv;
+
+    if ((rv = apr_thread_mutex_lock(queue->one_big_mutex)) != APR_SUCCESS) {
+        return rv;
+    }
+
+    /* Keep waiting until we wake up and find that the queue is not empty. */
+    if (ap_queue_empty(queue)) {
+        if (!queue->terminated) {
+            apr_thread_cond_wait(queue->not_empty, queue->one_big_mutex);
+        }
+        /* If we wake up and it's still empty, then we were interrupted */
+        if (ap_queue_empty(queue)) {
+            rv = apr_thread_mutex_unlock(queue->one_big_mutex);
+            if (rv != APR_SUCCESS) {
+                return rv;
+            }
+            if (queue->terminated) {
+                return APR_EOF; /* no more elements ever again */
+            }
+            else {
+                return APR_EINTR;
+            }
+        }
+    }
+
+    elem = &queue->data[--queue->nelts];
+    *sd = elem->sd;
+    *cs = elem->cs;
+    *p = elem->p;
+#ifdef AP_DEBUG
+    elem->sd = NULL;
+    elem->p = NULL;
+#endif /* AP_DEBUG */
+
+    rv = apr_thread_mutex_unlock(queue->one_big_mutex);
+    return rv;
+}
+
+apr_status_t ap_queue_interrupt_all(fd_queue_t * queue)
+{
+    apr_status_t rv;
+
+    if ((rv = apr_thread_mutex_lock(queue->one_big_mutex)) != APR_SUCCESS) {
+        return rv;
+    }
+    apr_thread_cond_broadcast(queue->not_empty);
+    return apr_thread_mutex_unlock(queue->one_big_mutex);
+}
+
+apr_status_t ap_queue_term(fd_queue_t * queue)
+{
+    apr_status_t rv;
+
+    if ((rv = apr_thread_mutex_lock(queue->one_big_mutex)) != APR_SUCCESS) {
+        return rv;
+    }
+    /* we must hold one_big_mutex when setting this... otherwise,
+     * we could end up setting it and waking everybody up just after a 
+     * would-be popper checks it but right before they block
+     */
+    queue->terminated = 1;
+    if ((rv = apr_thread_mutex_unlock(queue->one_big_mutex)) != APR_SUCCESS) {
+        return rv;
+    }
+    return ap_queue_interrupt_all(queue);
+}

server/mpm/experimental/event/fdqueue.h

@@ -0,0 +1,72 @@
+/* Copyright 2001-2004 The Apache Software Foundation
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef FDQUEUE_H
+#define FDQUEUE_H
+#include "httpd.h"
+#include <stdlib.h>
+#if APR_HAVE_UNISTD_H
+#include <unistd.h>
+#endif
+#include <apr_thread_mutex.h>
+#include <apr_thread_cond.h>
+#include <sys/types.h>
+#if APR_HAVE_SYS_SOCKET_H
+#include <sys/socket.h>
+#endif
+#include <apr_errno.h>
+
+typedef struct fd_queue_info_t fd_queue_info_t;
+
+apr_status_t ap_queue_info_create(fd_queue_info_t ** queue_info,
+                                  apr_pool_t * pool, int max_idlers);
+apr_status_t ap_queue_info_set_idle(fd_queue_info_t * queue_info,
+                                    apr_pool_t * pool_to_recycle);
+apr_status_t ap_queue_info_wait_for_idler(fd_queue_info_t * queue_info);
+apr_status_t ap_queue_info_term(fd_queue_info_t * queue_info);
+
+struct fd_queue_elem_t
+{
+    apr_socket_t *sd;
+    apr_pool_t *p;
+    conn_state_t *cs;
+};
+typedef struct fd_queue_elem_t fd_queue_elem_t;
+
+struct fd_queue_t
+{
+    fd_queue_elem_t *data;
+    int nelts;
+    int bounds;
+    apr_thread_mutex_t *one_big_mutex;
+    apr_thread_cond_t *not_empty;
+    int terminated;
+};
+typedef struct fd_queue_t fd_queue_t;
+
+void ap_pop_pool(apr_pool_t ** recycled_pool, fd_queue_info_t * queue_info);
+void ap_push_pool(fd_queue_info_t * queue_info,          
+                                    apr_pool_t * pool_to_recycle);
+
+apr_status_t ap_queue_init(fd_queue_t * queue, int queue_capacity,
+                           apr_pool_t * a);
+apr_status_t ap_queue_push(fd_queue_t * queue, apr_socket_t * sd,
+                           conn_state_t * cs, apr_pool_t * p);
+apr_status_t ap_queue_pop(fd_queue_t * queue, apr_socket_t ** sd,
+                          conn_state_t ** cs, apr_pool_t ** p);
+apr_status_t ap_queue_interrupt_all(fd_queue_t * queue);
+apr_status_t ap_queue_term(fd_queue_t * queue);
+
+#endif /* FDQUEUE_H */

server/mpm/experimental/event/mpm.h

@@ -0,0 +1,50 @@
+/* Copyright 2001-2004 The Apache Software Foundation
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "scoreboard.h"
+#include "unixd.h"
+
+#ifndef APACHE_MPM_EVENT_H
+#define APACHE_MPM_EVENT_H
+
+#define EVENT_MPM
+
+#define MPM_NAME "Event"
+
+#define AP_MPM_WANT_RECLAIM_CHILD_PROCESSES
+#define AP_MPM_WANT_WAIT_OR_TIMEOUT
+#define AP_MPM_WANT_PROCESS_CHILD_STATUS
+#define AP_MPM_WANT_SET_PIDFILE
+#define AP_MPM_WANT_SET_SCOREBOARD
+#define AP_MPM_WANT_SET_LOCKFILE
+#define AP_MPM_WANT_SET_MAX_REQUESTS
+#define AP_MPM_WANT_SET_COREDUMPDIR
+#define AP_MPM_WANT_SET_ACCEPT_LOCK_MECH
+#define AP_MPM_WANT_SIGNAL_SERVER
+#define AP_MPM_WANT_SET_MAX_MEM_FREE
+#define AP_MPM_WANT_SET_STACKSIZE
+#define AP_MPM_WANT_FATAL_SIGNAL_HANDLER
+#define AP_MPM_DISABLE_NAGLE_ACCEPTED_SOCK
+
+#define MPM_CHILD_PID(i) (ap_scoreboard_image->parent[i].pid)
+#define MPM_NOTE_CHILD_KILLED(i) (MPM_CHILD_PID(i) = 0)
+#define MPM_ACCEPT_FUNC unixd_accept
+
+extern int ap_threads_per_child;
+extern int ap_max_daemons_limit;
+extern server_rec *ap_server_conf;
+extern char ap_coredump_dir[MAX_STRING_LEN];
+
+#endif /* APACHE_MPM_EVENT_H */

server/mpm/experimental/event/mpm_default.h

@@ -0,0 +1,68 @@
+/* Copyright 2001-2004 The Apache Software Foundation
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef APACHE_MPM_DEFAULT_H
+#define APACHE_MPM_DEFAULT_H
+
+/* Number of servers to spawn off by default --- also, if fewer than
+ * this free when the caretaker checks, it will spawn more.
+ */
+#ifndef DEFAULT_START_DAEMON
+#define DEFAULT_START_DAEMON 3
+#endif
+
+/* Maximum number of *free* server processes --- more than this, and
+ * they will die off.
+ */
+
+#ifndef DEFAULT_MAX_FREE_DAEMON
+#define DEFAULT_MAX_FREE_DAEMON 10
+#endif
+
+/* Minimum --- fewer than this, and more will be created */
+
+#ifndef DEFAULT_MIN_FREE_DAEMON
+#define DEFAULT_MIN_FREE_DAEMON 3
+#endif
+
+#ifndef DEFAULT_THREADS_PER_CHILD
+#define DEFAULT_THREADS_PER_CHILD 25
+#endif
+
+/* File used for accept locking, when we use a file */
+#ifndef DEFAULT_LOCKFILE
+#define DEFAULT_LOCKFILE DEFAULT_REL_RUNTIMEDIR "/accept.lock"
+#endif
+
+/* Where the main/parent process's pid is logged */
+#ifndef DEFAULT_PIDLOG
+#define DEFAULT_PIDLOG DEFAULT_REL_RUNTIMEDIR "/httpd.pid"
+#endif
+
+/*
+ * Interval, in microseconds, between scoreboard maintenance.
+ */
+#ifndef SCOREBOARD_MAINTENANCE_INTERVAL
+#define SCOREBOARD_MAINTENANCE_INTERVAL 1000000
+#endif
+
+/* Number of requests to try to handle in a single process.  If <= 0,
+ * the children don't die off.
+ */
+#ifndef DEFAULT_MAX_REQUESTS_PER_CHILD
+#define DEFAULT_MAX_REQUESTS_PER_CHILD 10000
+#endif
+
+#endif /* AP_MPM_DEFAULT_H */

server/mpm/experimental/event/pod.c

@@ -0,0 +1,108 @@
+/* Copyright 2002-2004 The Apache Software Foundation
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "pod.h"
+
+#if APR_HAVE_UNISTD_H
+#include <unistd.h>
+#endif
+
+AP_DECLARE(apr_status_t) ap_mpm_pod_open(apr_pool_t * p, ap_pod_t ** pod)
+{
+    apr_status_t rv;
+
+    *pod = apr_palloc(p, sizeof(**pod));
+    rv = apr_file_pipe_create(&((*pod)->pod_in), &((*pod)->pod_out), p);
+    if (rv != APR_SUCCESS) {
+        return rv;
+    }
+/*
+    apr_file_pipe_timeout_set((*pod)->pod_in, 0);
+*/
+    (*pod)->p = p;
+
+    /* close these before exec. */
+    apr_file_inherit_unset((*pod)->pod_in);
+    apr_file_inherit_unset((*pod)->pod_out);
+
+    return APR_SUCCESS;
+}
+
+AP_DECLARE(int) ap_mpm_pod_check(ap_pod_t * pod)
+{
+    char c;
+    apr_os_file_t fd;
+    int rc;
+
+    /* we need to surface EINTR so we'll have to grab the
+     * native file descriptor and do the OS read() ourselves
+     */
+    apr_os_file_get(&fd, pod->pod_in);
+    rc = read(fd, &c, 1);
+    if (rc == 1) {
+        switch (c) {
+        case RESTART_CHAR:
+            return AP_RESTART;
+        case GRACEFUL_CHAR:
+            return AP_GRACEFUL;
+        }
+    }
+    return AP_NORESTART;
+}
+
+AP_DECLARE(apr_status_t) ap_mpm_pod_close(ap_pod_t * pod)
+{
+    apr_status_t rv;
+
+    rv = apr_file_close(pod->pod_out);
+    if (rv != APR_SUCCESS) {
+        return rv;
+    }
+
+    rv = apr_file_close(pod->pod_in);
+    if (rv != APR_SUCCESS) {
+        return rv;
+    }
+    return rv;
+}
+
+static apr_status_t pod_signal_internal(ap_pod_t * pod, int graceful)
+{
+    apr_status_t rv;
+    char char_of_death = graceful ? GRACEFUL_CHAR : RESTART_CHAR;
+    apr_size_t one = 1;
+
+    rv = apr_file_write(pod->pod_out, &char_of_death, &one);
+    if (rv != APR_SUCCESS) {
+        ap_log_error(APLOG_MARK, APLOG_WARNING, rv, ap_server_conf,
+                     "write pipe_of_death");
+    }
+    return rv;
+}
+
+AP_DECLARE(apr_status_t) ap_mpm_pod_signal(ap_pod_t * pod, int graceful)
+{
+    return pod_signal_internal(pod, graceful);
+}
+
+AP_DECLARE(void) ap_mpm_pod_killpg(ap_pod_t * pod, int num, int graceful)
+{
+    int i;
+    apr_status_t rv = APR_SUCCESS;
+
+    for (i = 0; i < num && rv == APR_SUCCESS; i++) {
+        rv = pod_signal_internal(pod, graceful);
+    }
+}

server/mpm/experimental/event/pod.h

@@ -0,0 +1,50 @@
+/* Copyright 2002-2004 The Apache Software Foundation
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "apr.h"
+#include "apr_strings.h"
+#define APR_WANT_STRFUNC
+#include "apr_want.h"
+
+#include "httpd.h"
+#include "http_config.h"
+#include "http_log.h"
+#include "http_main.h"
+#include "mpm.h"
+#include "mpm_common.h"
+#include "ap_mpm.h"
+#include "ap_listen.h"
+#include "mpm_default.h"
+
+#define RESTART_CHAR '$'
+#define GRACEFUL_CHAR '!'
+
+#define AP_RESTART  0
+#define AP_GRACEFUL 1
+
+typedef struct ap_pod_t ap_pod_t;
+
+struct ap_pod_t
+{
+    apr_file_t *pod_in;
+    apr_file_t *pod_out;
+    apr_pool_t *p;
+};
+
+AP_DECLARE(apr_status_t) ap_mpm_pod_open(apr_pool_t * p, ap_pod_t ** pod);
+AP_DECLARE(int) ap_mpm_pod_check(ap_pod_t * pod);
+AP_DECLARE(apr_status_t) ap_mpm_pod_close(ap_pod_t * pod);
+AP_DECLARE(apr_status_t) ap_mpm_pod_signal(ap_pod_t * pod, int graceful);
+AP_DECLARE(void) ap_mpm_pod_killpg(ap_pod_t * pod, int num, int graceful);