httpd-cvs mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From minf...@apache.org
Subject svn commit: r1705492 - in /httpd/httpd/branches/2.4.x: ./ docs/manual/mod/ include/ server/ server/mpm/event/ server/mpm/netware/ server/mpm/prefork/ server/mpm/winnt/ server/mpm/worker/
Date Sat, 26 Sep 2015 22:20:14 GMT
Author: minfrin
Date: Sat Sep 26 22:20:14 2015
New Revision: 1705492

URL: http://svn.apache.org/viewvc?rev=1705492&view=rev
Log:
MPMs: Support SO_REUSEPORT to create multiple duplicated listener
records for scalability.

Submitted by: Yingqi Lu <yingqi.lu@intel.com>, Jeff Trawick,
              Jim Jagielski, Yann Ylavic

Reviewed by: ylavic, jim, minfrin

Modified:
    httpd/httpd/branches/2.4.x/CHANGES
    httpd/httpd/branches/2.4.x/STATUS
    httpd/httpd/branches/2.4.x/docs/manual/mod/mpm_common.xml
    httpd/httpd/branches/2.4.x/include/ap_listen.h
    httpd/httpd/branches/2.4.x/include/ap_mmn.h
    httpd/httpd/branches/2.4.x/include/http_log.h
    httpd/httpd/branches/2.4.x/include/scoreboard.h
    httpd/httpd/branches/2.4.x/server/listen.c
    httpd/httpd/branches/2.4.x/server/log.c
    httpd/httpd/branches/2.4.x/server/mpm/event/event.c
    httpd/httpd/branches/2.4.x/server/mpm/netware/mpm_netware.c
    httpd/httpd/branches/2.4.x/server/mpm/prefork/prefork.c
    httpd/httpd/branches/2.4.x/server/mpm/winnt/mpm_winnt.c
    httpd/httpd/branches/2.4.x/server/mpm/worker/worker.c

Modified: httpd/httpd/branches/2.4.x/CHANGES
URL: http://svn.apache.org/viewvc/httpd/httpd/branches/2.4.x/CHANGES?rev=1705492&r1=1705491&r2=1705492&view=diff
==============================================================================
--- httpd/httpd/branches/2.4.x/CHANGES [utf-8] (original)
+++ httpd/httpd/branches/2.4.x/CHANGES [utf-8] Sat Sep 26 22:20:14 2015
@@ -2,6 +2,10 @@
 
 Changes with Apache 2.4.17
 
+  *) MPMs: Support SO_REUSEPORT to create multiple duplicated listener
+     records for scalability. [Yingqi Lu <yingqi.lu@intel.com>,
+     Jeff Trawick, Jim Jagielski, Yann Ylavic]
+
   *) mod_proxy: Fix a race condition that caused a failed worker to be retried
      before the retry period is over. [Ruediger Pluem]
 

Modified: httpd/httpd/branches/2.4.x/STATUS
URL: http://svn.apache.org/viewvc/httpd/httpd/branches/2.4.x/STATUS?rev=1705492&r1=1705491&r2=1705492&view=diff
==============================================================================
--- httpd/httpd/branches/2.4.x/STATUS (original)
+++ httpd/httpd/branches/2.4.x/STATUS Sat Sep 26 22:20:14 2015
@@ -109,27 +109,6 @@ RELEASE SHOWSTOPPERS:
 PATCHES ACCEPTED TO BACKPORT FROM TRUNK:
   [ start all new proposals below, under PATCHES PROPOSED. ]
 
-   * MPMs: Support SO_REUSEPORT to create multiple duplicated listener
-           records for scalability (full log in 2.4.x patch).
-     trunk patch: http://svn.apache.org/r1599531
-                  http://svn.apache.org/r1599593
-                  http://svn.apache.org/r1599601
-                  http://svn.apache.org/r1599603
-                  http://svn.apache.org/r1601558
-                  http://svn.apache.org/r1629909
-                  http://svn.apache.org/r1629918
-                  http://svn.apache.org/r1629990
-                  http://svn.apache.org/r1635521
-                  http://svn.apache.org/r1635859
-                  http://svn.apache.org/r1640145
-                  http://svn.apache.org/r1640161
-                  http://svn.apache.org/r1640184
-                  http://svn.apache.org/r1640763
-                  http://svn.apache.org/r1643179
-                  http://svn.apache.org/r1656368
-                  http://svn.apache.org/r1679714
-     2.4.x patch: http://people.apache.org/~ylavic/httpd-2.4.x-ap_listeners_buckets-v3.patch
-     +1: ylavic, jim, minfrin
  
 
 PATCHES PROPOSED TO BACKPORT FROM TRUNK:

Modified: httpd/httpd/branches/2.4.x/docs/manual/mod/mpm_common.xml
URL: http://svn.apache.org/viewvc/httpd/httpd/branches/2.4.x/docs/manual/mod/mpm_common.xml?rev=1705492&r1=1705491&r2=1705492&view=diff
==============================================================================
--- httpd/httpd/branches/2.4.x/docs/manual/mod/mpm_common.xml (original)
+++ httpd/httpd/branches/2.4.x/docs/manual/mod/mpm_common.xml Sat Sep 26 22:20:14 2015
@@ -253,6 +253,44 @@ including other causes.</a></seealso>
 </directivesynopsis>
 
 <directivesynopsis>
+<name>ListenCoresBucketsRatio</name>
+<description>Ratio between the number of CPU cores (online) and the number of
+listeners' buckets</description>
+<syntax>ListenCoresBucketsRatio <var>ratio</var></syntax>
+<default>ListenCoresBucketsRatio 0 (disabled)</default>
+<contextlist><context>server config</context></contextlist>
+<modulelist>
+<module>event</module>
+<module>prefork</module>
+<module>worker</module></modulelist>
+<compatibility>Available in Apache HTTP Server 2.4.13, with a kernel supporting
+the socket option <code>SO_REUSEPORT</code> and distributing new connections
+evenly accross listening processes' (or threads') sockets using it (eg. Linux
+3.9 and later, but not the current implementations of <code>SO_REUSEPORT</code>
+in *BSDs.</compatibility>
+
+<usage>
+    <p>A <var>ratio</var> between the number of (online) CPU cores and the
+    number of listeners' buckets can be used to make Apache HTTP Server create
+    <code>num_cpu_cores / ratio</code> listening buckets, each containing its
+    own <directive>Listen</directive>-ing socket(s) on the same port(s), and
+    then make each child handle a single bucket (with round-robin distribution
+    of the buckets at children creation time).</p>
+
+    <p><directive>ListenCoresBucketsRatio</directive> can improve the
+    scalability when accepting new connections is/becomes the bottleneck.
+    On systems with a large number of CPU cores, enabling this feature has
+    been tested to show significant performances improvement and shorter
+    responses time.</p>
+
+    <p>There must be at least twice the number of CPU cores than the
+    configured <var>ratio</var> for this to be active. The recommended
+    <var>ratio</var> is <code>8</code>, hence at least <code>16</code>
+    cores should be available at runtime when this value is used.</p>
+</usage>
+</directivesynopsis>
+
+<directivesynopsis>
 <name>ListenBackLog</name>
 <description>Maximum length of the queue of pending connections</description>
 <syntax>ListenBacklog <var>backlog</var></syntax>

Modified: httpd/httpd/branches/2.4.x/include/ap_listen.h
URL: http://svn.apache.org/viewvc/httpd/httpd/branches/2.4.x/include/ap_listen.h?rev=1705492&r1=1705491&r2=1705492&view=diff
==============================================================================
--- httpd/httpd/branches/2.4.x/include/ap_listen.h (original)
+++ httpd/httpd/branches/2.4.x/include/ap_listen.h Sat Sep 26 22:20:14 2015
@@ -77,6 +77,8 @@ struct ap_listen_rec {
  * The global list of ap_listen_rec structures
  */
 AP_DECLARE_DATA extern ap_listen_rec *ap_listeners;
+AP_DECLARE_DATA extern int ap_num_listen_buckets;
+AP_DECLARE_DATA extern int ap_have_so_reuseport;
 
 /**
  * Setup all of the defaults for the listener list
@@ -92,11 +94,34 @@ AP_DECLARE(void) ap_listen_pre_config(vo
 AP_DECLARE(int) ap_setup_listeners(server_rec *s);
 
 /**
+ * This function duplicates ap_listeners into multiple buckets when configured
+ * to (see ListenCoresBucketsRatio) and the platform supports it (eg. number of
+ * online CPU cores and SO_REUSEPORT available).
+ * @param p The config pool
+ * @param s The global server_rec
+ * @param buckets The array of listeners buckets.
+ * @param num_buckets The total number of listeners buckets (array size).
+ * @remark If the given *num_buckets is 0 (input), it will be computed
+ *         according to the platform capacities, otherwise (positive) it
+ *         will be preserved. The number of listeners duplicated will
+ *         always match *num_buckets, be it computed or given.
+ */
+AP_DECLARE(apr_status_t) ap_duplicate_listeners(apr_pool_t *p, server_rec *s,
+                                                ap_listen_rec ***buckets,
+                                                int *num_buckets);
+
+/**
  * Loop through the global ap_listen_rec list and close each of the sockets.
  */
 AP_DECLARE_NONSTD(void) ap_close_listeners(void);
 
 /**
+ * Loop through the given ap_listen_rec list and close each of the sockets.
+ * @param listener The listener to close.
+ */
+AP_DECLARE_NONSTD(void) ap_close_listeners_ex(ap_listen_rec *listeners);
+
+/**
  * FIXMEDOC
  */
 AP_DECLARE_NONSTD(int) ap_close_selected_listeners(ap_slave_t *);
@@ -109,6 +134,7 @@ AP_DECLARE_NONSTD(int) ap_close_selected
  * called.
  */
 AP_DECLARE_NONSTD(const char *) ap_set_listenbacklog(cmd_parms *cmd, void *dummy, const char *arg);
+AP_DECLARE_NONSTD(const char *) ap_set_listencbratio(cmd_parms *cmd, void *dummy, const char *arg);
 AP_DECLARE_NONSTD(const char *) ap_set_listener(cmd_parms *cmd, void *dummy,
                                                 int argc, char *const argv[]);
 AP_DECLARE_NONSTD(const char *) ap_set_send_buffer_size(cmd_parms *cmd, void *dummy,
@@ -120,6 +146,8 @@ AP_DECLARE_NONSTD(const char *) ap_set_r
 #define LISTEN_COMMANDS \
 AP_INIT_TAKE1("ListenBacklog", ap_set_listenbacklog, NULL, RSRC_CONF, \
   "Maximum length of the queue of pending connections, as used by listen(2)"), \
+AP_INIT_TAKE1("ListenCoresBucketsRatio", ap_set_listencbratio, NULL, RSRC_CONF, \
+  "Ratio between the number of CPU cores (online) and the number of listeners buckets"), \
 AP_INIT_TAKE_ARGV("Listen", ap_set_listener, NULL, RSRC_CONF, \
   "A port number or a numeric IP address and a port number, and an optional protocol"), \
 AP_INIT_TAKE1("SendBufferSize", ap_set_send_buffer_size, NULL, RSRC_CONF, \

Modified: httpd/httpd/branches/2.4.x/include/ap_mmn.h
URL: http://svn.apache.org/viewvc/httpd/httpd/branches/2.4.x/include/ap_mmn.h?rev=1705492&r1=1705491&r2=1705492&view=diff
==============================================================================
--- httpd/httpd/branches/2.4.x/include/ap_mmn.h (original)
+++ httpd/httpd/branches/2.4.x/include/ap_mmn.h Sat Sep 26 22:20:14 2015
@@ -445,6 +445,11 @@
  * 20120211.46 (2.4.13-dev) Add ap_map_http_request_error()
  * 20120211.47 (2.4.13-dev) Add ap_some_authn_required, ap_force_authn hook.
  *                          Deprecate broken ap_some_auth_required.
+ * 20120211.48 (2.4.13-dev) Added ap_log_mpm_common().
+ * 20120211.49 (2.4.13-dev) Add listener bucket in scoreboard.h's process_score.
+ * 20120211.50 (2.4.13-dev) Add ap_set_listencbratio(), ap_close_listeners_ex(),
+ *                          ap_duplicate_listeners(), ap_num_listen_buckets and
+ *                          ap_have_so_reuseport to ap_listen.h.
  */
 
 #define MODULE_MAGIC_COOKIE 0x41503234UL /* "AP24" */
@@ -452,7 +457,7 @@
 #ifndef MODULE_MAGIC_NUMBER_MAJOR
 #define MODULE_MAGIC_NUMBER_MAJOR 20120211
 #endif
-#define MODULE_MAGIC_NUMBER_MINOR 47                   /* 0...n */
+#define MODULE_MAGIC_NUMBER_MINOR 50                   /* 0...n */
 
 /**
  * Determine if the server's current MODULE_MAGIC_NUMBER is at least a

Modified: httpd/httpd/branches/2.4.x/include/http_log.h
URL: http://svn.apache.org/viewvc/httpd/httpd/branches/2.4.x/include/http_log.h?rev=1705492&r1=1705491&r2=1705492&view=diff
==============================================================================
--- httpd/httpd/branches/2.4.x/include/http_log.h (original)
+++ httpd/httpd/branches/2.4.x/include/http_log.h Sat Sep 26 22:20:14 2015
@@ -741,6 +741,13 @@ AP_DECLARE(void) ap_error_log2stderr(ser
 AP_DECLARE(void) ap_log_command_line(apr_pool_t *p, server_rec *s);
 
 /**
+ * Log common (various) MPM shared data at startup.
+ * @param s The server_rec of the error log we want to log to.
+ * Misc commonly logged data is logged to that server's error log.
+ */
+AP_DECLARE(void) ap_log_mpm_common(server_rec *s);
+
+/**
  * Log the current pid of the parent process
  * @param p The pool to use for processing
  * @param fname The name of the file to log to.  If the filename is not

Modified: httpd/httpd/branches/2.4.x/include/scoreboard.h
URL: http://svn.apache.org/viewvc/httpd/httpd/branches/2.4.x/include/scoreboard.h?rev=1705492&r1=1705491&r2=1705492&view=diff
==============================================================================
--- httpd/httpd/branches/2.4.x/include/scoreboard.h (original)
+++ httpd/httpd/branches/2.4.x/include/scoreboard.h Sat Sep 26 22:20:14 2015
@@ -142,6 +142,7 @@ struct process_score {
     apr_uint32_t lingering_close;   /* async connections in lingering close */
     apr_uint32_t keep_alive;        /* async connections in keep alive */
     apr_uint32_t suspended;         /* connections suspended by some module */
+    int bucket;             /* Listener bucket used by this child */
 };
 
 /* Scoreboard is now in 'local' memory, since it isn't updated once created,

Modified: httpd/httpd/branches/2.4.x/server/listen.c
URL: http://svn.apache.org/viewvc/httpd/httpd/branches/2.4.x/server/listen.c?rev=1705492&r1=1705491&r2=1705492&view=diff
==============================================================================
--- httpd/httpd/branches/2.4.x/server/listen.c (original)
+++ httpd/httpd/branches/2.4.x/server/listen.c Sat Sep 26 22:20:14 2015
@@ -22,20 +22,41 @@
 
 #include "ap_config.h"
 #include "httpd.h"
+#include "http_main.h"
 #include "http_config.h"
 #include "http_core.h"
 #include "ap_listen.h"
 #include "http_log.h"
 #include "mpm_common.h"
 
+#include <stdlib.h>
+#if APR_HAVE_UNISTD_H
+#include <unistd.h>
+#endif
+
 /* we know core's module_index is 0 */
 #undef APLOG_MODULE_INDEX
 #define APLOG_MODULE_INDEX AP_CORE_MODULE_INDEX
 
 AP_DECLARE_DATA ap_listen_rec *ap_listeners = NULL;
 
+/* Let ap_num_listen_buckets be global so that it can
+ * be printed by ap_log_mpm_common(), but keep the listeners
+ * buckets static since it is used only here to close them
+ * all (including duplicated) with ap_close_listeners().
+ */
+AP_DECLARE_DATA int ap_num_listen_buckets;
+static ap_listen_rec **ap_listen_buckets;
+
+/* Determine once, at runtime, whether or not SO_REUSEPORT
+ * is usable on this platform, and hence whether or not
+ * listeners can be duplicated (if configured).
+ */
+AP_DECLARE_DATA int ap_have_so_reuseport = -1;
+
 static ap_listen_rec *old_listeners;
 static int ap_listenbacklog;
+static int ap_listencbratio;
 static int send_buffer_size;
 static int receive_buffer_size;
 
@@ -131,6 +152,23 @@ static apr_status_t make_sock(apr_pool_t
     ap_sock_disable_nagle(s);
 #endif
 
+#if defined(SO_REUSEPORT)
+    if (ap_have_so_reuseport) {
+        int thesock;
+        apr_os_sock_get(&thesock, s);
+        if (setsockopt(thesock, SOL_SOCKET, SO_REUSEPORT,
+                       (void *)&one, sizeof(int)) < 0) {
+            stat = apr_get_netos_error();
+            ap_log_perror(APLOG_MARK, APLOG_CRIT, stat, p, APLOGNO(02638)
+                          "make_sock: for address %pI, apr_socket_opt_set: "
+                          "(SO_REUSEPORT)",
+                          server->bind_addr);
+            apr_socket_close(s);
+            return stat;
+        }
+    }
+#endif
+
     if ((stat = apr_socket_bind(s, server->bind_addr)) != APR_SUCCESS) {
         ap_log_perror(APLOG_MARK, APLOG_STARTUP|APLOG_CRIT, stat, p, APLOGNO(00072)
                       "make_sock: could not bind to address %pI",
@@ -482,11 +520,7 @@ static int open_listeners(apr_pool_t *po
     }
 
     /* close the old listeners */
-    for (lr = old_listeners; lr; lr = next) {
-        apr_socket_close(lr->sd);
-        lr->active = 0;
-        next = lr->next;
-    }
+    ap_close_listeners_ex(old_listeners);
     old_listeners = NULL;
 
 #if AP_NONBLOCK_WHEN_MULTI_LISTEN
@@ -558,7 +592,7 @@ AP_DECLARE(int) ap_setup_listeners(serve
     }
 
     if (open_listeners(s->process->pool)) {
-       return 0;
+        return 0;
     }
 
     for (lr = ap_listeners; lr; lr = lr->next) {
@@ -582,15 +616,124 @@ AP_DECLARE(int) ap_setup_listeners(serve
     return num_listeners;
 }
 
-AP_DECLARE_NONSTD(void) ap_close_listeners(void)
+AP_DECLARE(apr_status_t) ap_duplicate_listeners(apr_pool_t *p, server_rec *s,
+                                                ap_listen_rec ***buckets,
+                                                int *num_buckets)
 {
+    static int warn_once;
+    int i;
+    apr_status_t stat;
+    int use_nonblock = 0;
     ap_listen_rec *lr;
 
-    for (lr = ap_listeners; lr; lr = lr->next) {
+    if (*num_buckets < 1) {
+        *num_buckets = 1;
+        if (ap_listencbratio > 0) {
+#ifdef _SC_NPROCESSORS_ONLN
+            if (ap_have_so_reuseport) {
+                int num_online_cores = sysconf(_SC_NPROCESSORS_ONLN),
+                    val = num_online_cores / ap_listencbratio;
+                if (val > 1) {
+                    *num_buckets = val;
+                }
+                ap_log_perror(APLOG_MARK, APLOG_INFO, 0, p, APLOGNO(02819)
+                              "Using %i listeners bucket(s) based on %i "
+                              "online CPU cores and a ratio of %i",
+                              *num_buckets, num_online_cores,
+                              ap_listencbratio);
+            }
+            else
+#endif
+            if (!warn_once) {
+                ap_log_perror(APLOG_MARK, APLOG_WARNING, 0, p, APLOGNO(02820)
+                              "ListenCoresBucketsRatio ignored without "
+                              "SO_REUSEPORT and _SC_NPROCESSORS_ONLN "
+                              "support: using a single listeners bucket");
+                warn_once = 1;
+            }
+        }
+    }
+
+    *buckets = apr_pcalloc(p, *num_buckets * sizeof(ap_listen_rec *));
+    (*buckets)[0] = ap_listeners;
+
+    for (i = 1; i < *num_buckets; i++) {
+        ap_listen_rec *last = NULL;
+        lr = ap_listeners;
+        while (lr) {
+            ap_listen_rec *duplr;
+            char *hostname;
+            apr_port_t port;
+            apr_sockaddr_t *sa;
+            duplr = apr_palloc(p, sizeof(ap_listen_rec));
+            duplr->slave = NULL;
+            duplr->protocol = apr_pstrdup(p, lr->protocol);
+            hostname = apr_pstrdup(p, lr->bind_addr->hostname);
+            port = lr->bind_addr->port;
+            apr_sockaddr_info_get(&sa, hostname, APR_UNSPEC, port, 0, p);
+            duplr->bind_addr = sa;
+            duplr->next = NULL;
+            stat = apr_socket_create(&duplr->sd, duplr->bind_addr->family,
+                                     SOCK_STREAM, 0, p);
+            if (stat != APR_SUCCESS) {
+                ap_log_perror(APLOG_MARK, APLOG_CRIT, 0, p, APLOGNO(02640)
+                            "ap_duplicate_listeners: for address %pI, "
+                            "cannot duplicate a new socket!",
+                            duplr->bind_addr);
+                return stat;
+            }
+            make_sock(p, duplr);
+#if AP_NONBLOCK_WHEN_MULTI_LISTEN
+            use_nonblock = (ap_listeners && ap_listeners->next);
+            stat = apr_socket_opt_set(duplr->sd, APR_SO_NONBLOCK, use_nonblock);
+            if (stat != APR_SUCCESS) {
+                ap_log_perror(APLOG_MARK, APLOG_CRIT, stat, p, APLOGNO(02641)
+                              "unable to control socket non-blocking status");
+                return stat;
+            }
+#endif
+            ap_apply_accept_filter(p, duplr, s);
+
+            if (last == NULL) {
+                (*buckets)[i] = last = duplr;
+            }
+            else {
+                last->next = duplr;
+                last = duplr;
+            }
+            lr = lr->next;
+        }
+    }
+
+    ap_listen_buckets = *buckets;
+    ap_num_listen_buckets = *num_buckets;
+    return APR_SUCCESS;
+}
+
+AP_DECLARE_NONSTD(void) ap_close_listeners(void)
+{
+    int i;
+
+    ap_close_listeners_ex(ap_listeners);
+
+    /* Start from index 1 since either ap_duplicate_listeners()
+     * was called and ap_listen_buckets[0] == ap_listeners, or
+     * it wasn't and ap_num_listen_buckets == 0.
+     */
+    for (i = 1; i < ap_num_listen_buckets; i++) {
+        ap_close_listeners_ex(ap_listen_buckets[i]);
+    }
+}
+
+AP_DECLARE_NONSTD(void) ap_close_listeners_ex(ap_listen_rec *listeners)
+{
+    ap_listen_rec *lr;
+    for (lr = listeners; lr; lr = lr->next) {
         apr_socket_close(lr->sd);
         lr->active = 0;
     }
 }
+
 AP_DECLARE_NONSTD(int) ap_close_selected_listeners(ap_slave_t *slave)
 {
     ap_listen_rec *lr;
@@ -612,7 +755,43 @@ AP_DECLARE(void) ap_listen_pre_config(vo
 {
     old_listeners = ap_listeners;
     ap_listeners = NULL;
+    ap_listen_buckets = NULL;
+    ap_num_listen_buckets = 0;
     ap_listenbacklog = DEFAULT_LISTENBACKLOG;
+    ap_listencbratio = 0;
+
+    /* Check once whether or not SO_REUSEPORT is supported. */
+    if (ap_have_so_reuseport < 0) {
+        /* This is limited to Linux with defined SO_REUSEPORT (ie. 3.9+) for
+         * now since the implementation evenly distributes connections accross
+         * all the listening threads/processes.
+         *
+         * *BSDs have SO_REUSEPORT too but with a different semantic: the first
+         * wildcard address bound socket or the last non-wildcard address bound
+         * socket will receive connections (no evenness garantee); the rest of
+         * the sockets bound to the same port will not.
+         * This can't (always) work for httpd.
+         *
+         * TODO: latests DragonFlyBSD's SO_REUSEPORT (seems to?) have the same
+         * semantic as Linux, so we may need HAVE_SO_REUSEPORT available from
+         * configure.in some day.
+         */
+#if defined(SO_REUSEPORT) && defined(__linux__)
+        apr_socket_t *sock;
+        if (apr_socket_create(&sock, APR_UNSPEC, SOCK_STREAM, 0,
+                              ap_pglobal) == APR_SUCCESS) {
+            int thesock, on = 1;
+            apr_os_sock_get(&thesock, sock);
+            ap_have_so_reuseport = (setsockopt(thesock, SOL_SOCKET,
+                                               SO_REUSEPORT, (void *)&on,
+                                               sizeof(int)) == 0);
+            apr_socket_close(sock);
+        }
+        else
+#endif
+        ap_have_so_reuseport = 0;
+
+    }
 }
 
 AP_DECLARE_NONSTD(const char *) ap_set_listener(cmd_parms *cmd, void *dummy,
@@ -684,6 +863,26 @@ AP_DECLARE_NONSTD(const char *) ap_set_l
     return NULL;
 }
 
+AP_DECLARE_NONSTD(const char *) ap_set_listencbratio(cmd_parms *cmd,
+                                                     void *dummy,
+                                                     const char *arg)
+{
+    int b;
+    const char *err = ap_check_cmd_context(cmd, GLOBAL_ONLY);
+
+    if (err != NULL) {
+        return err;
+    }
+
+    b = atoi(arg);
+    if (b < 1) {
+        return "ListenCoresBucketsRatio must be > 0";
+    }
+
+    ap_listencbratio = b;
+    return NULL;
+}
+
 AP_DECLARE_NONSTD(const char *) ap_set_send_buffer_size(cmd_parms *cmd,
                                                         void *dummy,
                                                         const char *arg)

Modified: httpd/httpd/branches/2.4.x/server/log.c
URL: http://svn.apache.org/viewvc/httpd/httpd/branches/2.4.x/server/log.c?rev=1705492&r1=1705491&r2=1705492&view=diff
==============================================================================
--- httpd/httpd/branches/2.4.x/server/log.c (original)
+++ httpd/httpd/branches/2.4.x/server/log.c Sat Sep 26 22:20:14 2015
@@ -53,6 +53,7 @@
 #include "http_main.h"
 #include "util_time.h"
 #include "ap_mpm.h"
+#include "ap_listen.h"
 
 #if HAVE_GETTID
 #include <sys/syscall.h>
@@ -1536,6 +1537,15 @@ AP_DECLARE(void) ap_log_command_line(apr
                  "Command line: '%s'", result);
 }
 
+/* grab bag function to log commonly logged and shared info */
+AP_DECLARE(void) ap_log_mpm_common(server_rec *s)
+{
+    ap_log_error(APLOG_MARK, APLOG_DEBUG , 0, s, APLOGNO(02639)
+                 "Using SO_REUSEPORT: %s (%d)",
+                 ap_have_so_reuseport ? "yes" : "no",
+                 ap_num_listen_buckets);
+}
+
 AP_DECLARE(void) ap_remove_pid(apr_pool_t *p, const char *rel_fname)
 {
     apr_status_t rv;

Modified: httpd/httpd/branches/2.4.x/server/mpm/event/event.c
URL: http://svn.apache.org/viewvc/httpd/httpd/branches/2.4.x/server/mpm/event/event.c?rev=1705492&r1=1705491&r2=1705492&view=diff
==============================================================================
--- httpd/httpd/branches/2.4.x/server/mpm/event/event.c (original)
+++ httpd/httpd/branches/2.4.x/server/mpm/event/event.c Sat Sep 26 22:20:14 2015
@@ -59,6 +59,8 @@
 #include "apr_want.h"
 #include "apr_version.h"
 
+#include <stdlib.h>
+
 #if APR_HAVE_UNISTD_H
 #include <unistd.h>
 #endif
@@ -336,16 +338,29 @@ typedef struct event_retained_data {
     /*
      * 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.
+     * 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;
+    int *idle_spawn_rate;
 #ifndef MAX_SPAWN_RATE
 #define MAX_SPAWN_RATE        (32)
 #endif
     int hold_off_on_exponential_spawning;
+    /*
+     * Current number of listeners buckets and maximum reached accross
+     * restarts (to size retained data according to dynamic num_buckets,
+     * eg. idle_spawn_rate).
+     */
+    int num_buckets, max_buckets;
 } 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,
@@ -354,8 +369,6 @@ struct event_srv_cfg_s {
 
 #define ID_FROM_CHILD_THREAD(c, t)    ((c * thread_limit) + t)
 
-static ap_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
@@ -1190,11 +1203,12 @@ static void check_infinite_requests(void
     }
 }
 
-static void close_listeners(int process_slot, int *closed) {
+static void close_listeners(int process_slot, int *closed)
+{
     if (!*closed) {
         int i;
         disable_listensocks(process_slot);
-        ap_close_listeners();
+        ap_close_listeners_ex(my_bucket->listeners);
         *closed = 1;
         dying = 1;
         ap_scoreboard_image->parent[process_slot].quiescing = 1;
@@ -1235,7 +1249,7 @@ static apr_status_t init_pollset(apr_poo
     int i = 0;
 
     listener_pollfd = apr_palloc(p, sizeof(apr_pollfd_t) * num_listensocks);
-    for (lr = ap_listeners; lr != NULL; lr = lr->next, i++) {
+    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];
@@ -2216,13 +2230,14 @@ static void join_start_thread(apr_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;
 
     mpm_state = AP_MPMQ_STARTING;       /* for benefit of any hooks that run as this
                                          * child initializes
@@ -2231,6 +2246,14 @@ static void child_main(int child_num_arg
     ap_fatal_signal_child_setup(ap_server_conf);
     apr_pool_create(&pchild, pconf);
 
+    /* close unused listeners and pods */
+    for (i = 0; i < retained->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);
 
@@ -2340,7 +2363,7 @@ static void child_main(int child_num_arg
         apr_signal(SIGTERM, dummy_signal_handler);
         /* Watch for any messages from the parent over the POD */
         while (1) {
-            rv = ap_mpm_podx_check(pod);
+            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) {
@@ -2378,7 +2401,7 @@ static void child_main(int child_num_arg
     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;
 
@@ -2387,10 +2410,14 @@ static int make_child(server_rec * s, in
     }
 
     if (one_process) {
+        my_bucket = &all_buckets[0];
+
         set_signals();
         event_note_child_started(slot, getpid());
-        child_main(slot);
+        child_main(slot, 0);
         /* NOTREACHED */
+        ap_assert(0);
+        return -1;
     }
 
     if ((pid = fork()) == -1) {
@@ -2413,6 +2440,8 @@ static int make_child(server_rec * s, in
     }
 
     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.
@@ -2427,10 +2456,12 @@ static int make_child(server_rec * s, in
         RAISE_SIGSTOP(MAKE_CHILD);
 
         apr_signal(SIGTERM, just_die);
-        child_main(slot);
+        child_main(slot, bucket);
         /* NOTREACHED */
+        ap_assert(0);
+        return -1;
     }
-    /* 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
@@ -2440,6 +2471,7 @@ static int make_child(server_rec * s, in
     }
     ap_scoreboard_image->parent[slot].quiescing = 0;
     ap_scoreboard_image->parent[slot].not_accepting = 0;
+    ap_scoreboard_image->parent[slot].bucket = bucket;
     event_note_child_started(slot, pid);
     return 0;
 }
@@ -2453,14 +2485,14 @@ static void startup_children(int number_
         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->num_buckets) < 0) {
             break;
         }
         --number_to_start;
     }
 }
 
-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;
@@ -2490,7 +2522,7 @@ static void perform_idle_server_maintena
         int child_threads_active = 0;
 
         if (i >= retained->max_daemons_limit
-            && totally_free_length == retained->idle_spawn_rate)
+            && totally_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
              */
@@ -2517,7 +2549,8 @@ static void perform_idle_server_maintena
             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->not_accepting
-                    && ps->generation == retained->my_generation)
+                    && ps->generation == retained->my_generation
+                    && ps->bucket == child_bucket)
                 {
                     ++idle_thread_count;
                 }
@@ -2528,8 +2561,8 @@ static void perform_idle_server_maintena
         }
         active_thread_count += child_threads_active;
         if (any_dead_threads
-            && totally_free_length < retained->idle_spawn_rate
-            && free_length < MAX_SPAWN_RATE
+            && totally_free_length < retained->idle_spawn_rate[child_bucket]
+            && free_length < MAX_SPAWN_RATE / num_buckets
             && (!ps->pid      /* no process in the slot */
                   || ps->quiescing)) {  /* or at least one is going away */
             if (all_dead_threads) {
@@ -2585,12 +2618,13 @@ static void perform_idle_server_maintena
 
     retained->max_daemons_limit = last_non_dead + 1;
 
-    if (idle_thread_count > max_spare_threads) {
+    if (idle_thread_count > max_spare_threads / num_buckets) {
         /* Kill off one child */
-        ap_mpm_podx_signal(pod, AP_MPM_PODX_GRACEFUL);
-        retained->idle_spawn_rate = 1;
+        ap_mpm_podx_signal(all_buckets[child_bucket].pod,
+                           AP_MPM_PODX_GRACEFUL);
+        retained->idle_spawn_rate[child_bucket] = 1;
     }
-    else if (idle_thread_count < min_spare_threads) {
+    else if (idle_thread_count < min_spare_threads / num_buckets) {
         /* terminate the free list */
         if (free_length == 0) { /* scoreboard is full, can't fork */
 
@@ -2608,13 +2642,13 @@ static void perform_idle_server_maintena
                 ap_log_error(APLOG_MARK, APLOG_ERR, 0, ap_server_conf, APLOGNO(00485)
                              "scoreboard is full, not at MaxRequestWorkers");
             }
-            retained->idle_spawn_rate = 1;
+            retained->idle_spawn_rate[child_bucket] = 1;
         }
         else {
-            if (free_length > retained->idle_spawn_rate) {
-                free_length = retained->idle_spawn_rate;
+            if (free_length > retained->idle_spawn_rate[child_bucket]) {
+                free_length = retained->idle_spawn_rate[child_bucket];
             }
-            if (retained->idle_spawn_rate >= 8) {
+            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 "
@@ -2624,7 +2658,7 @@ static void perform_idle_server_maintena
                              idle_thread_count, total_non_dead);
             }
             for (i = 0; i < free_length; ++i) {
-                make_child(ap_server_conf, free_slots[i]);
+                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
@@ -2632,17 +2666,18 @@ static void perform_idle_server_maintena
             if (retained->hold_off_on_exponential_spawning) {
                 --retained->hold_off_on_exponential_spawning;
             }
-            else if (retained->idle_spawn_rate < MAX_SPAWN_RATE) {
-                retained->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 {
-        retained->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)
 {
     ap_generation_t old_gen;
     int child_slot;
@@ -2686,23 +2721,26 @@ static void server_main_loop(int remaini
             }
             /* non-fatal death... note that it's gone in the scoreboard. */
             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);
 
                 event_note_child_killed(child_slot, 0, 0);
-                ap_scoreboard_image->parent[child_slot].quiescing = 0;
+                ps = &ap_scoreboard_image->parent[child_slot];
+                ps->quiescing = 0;
                 if (processed_status == APEXIT_CHILDSICK) {
                     /* resource shortage, minimize the fork rate */
-                    retained->idle_spawn_rate = 1;
+                    retained->idle_spawn_rate[ps->bucket] = 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);
+                    make_child(ap_server_conf, child_slot, ps->bucket);
                     --remaining_children_to_start;
                 }
             }
@@ -2713,7 +2751,9 @@ static void server_main_loop(int remaini
                 if (processed_status == APEXIT_CHILDSICK
                     && old_gen == retained->my_generation) {
                     /* resource shortage, minimize the fork rate */
-                    retained->idle_spawn_rate = 1;
+                    for (i = 0; i < num_buckets; i++) {
+                        retained->idle_spawn_rate[i] = 1;
+                    }
                 }
 #if APR_HAS_OTHER_CHILD
             }
@@ -2752,13 +2792,17 @@ static void server_main_loop(int remaini
             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->num_buckets;
     int remaining_children_to_start;
+    int i;
 
     ap_log_pid(pconf, ap_pid_fname);
 
@@ -2775,9 +2819,18 @@ static int event_run(apr_pool_t * _pconf
 
     restart_pending = shutdown_pending = 0;
     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 (ap_daemons_limit < num_buckets)
+        ap_daemons_limit = num_buckets;
+    if (ap_daemons_to_start < num_buckets)
+        ap_daemons_to_start = num_buckets;
+    if (min_spare_threads < threads_per_child * num_buckets)
+        min_spare_threads = threads_per_child * num_buckets;
+    if (max_spare_threads < min_spare_threads + threads_per_child * num_buckets)
+        max_spare_threads = min_spare_threads + threads_per_child * 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
@@ -2808,17 +2861,21 @@ static int event_run(apr_pool_t * _pconf
     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);
 
     mpm_state = AP_MPMQ_RUNNING;
 
-    server_main_loop(remaining_children_to_start);
+    server_main_loop(remaining_children_to_start, num_buckets);
     mpm_state = AP_MPMQ_STOPPING;
 
     if (shutdown_pending && !retained->is_graceful) {
         /* Time to shut down:
          * Kill child processes, tell them to call child_exit, etc...
          */
-        ap_mpm_podx_killpg(pod, ap_daemons_limit, AP_MPM_PODX_RESTART);
+        for (i = 0; i < num_buckets; i++) {
+            ap_mpm_podx_killpg(all_buckets[i].pod, ap_daemons_limit,
+                               AP_MPM_PODX_RESTART);
+        }
         ap_reclaim_child_processes(1, /* Start with SIGTERM */
                                    event_note_child_killed);
 
@@ -2839,7 +2896,10 @@ static int event_run(apr_pool_t * _pconf
 
         /* Close our listeners, and then ask our children to do same */
         ap_close_listeners();
-        ap_mpm_podx_killpg(pod, ap_daemons_limit, AP_MPM_PODX_GRACEFUL);
+        for (i = 0; i < num_buckets; i++) {
+            ap_mpm_podx_killpg(all_buckets[i].pod, ap_daemons_limit,
+                               AP_MPM_PODX_GRACEFUL);
+        }
         ap_relieve_child_processes(event_note_child_killed);
 
         if (!child_fatal) {
@@ -2879,7 +2939,10 @@ static int event_run(apr_pool_t * _pconf
          * way, try and make sure that all of our processes are
          * really dead.
          */
-        ap_mpm_podx_killpg(pod, ap_daemons_limit, AP_MPM_PODX_RESTART);
+        for (i = 0; i < num_buckets; i++) {
+            ap_mpm_podx_killpg(all_buckets[i].pod, ap_daemons_limit,
+                               AP_MPM_PODX_RESTART);
+        }
         ap_reclaim_child_processes(1, event_note_child_killed);
 
         return DONE;
@@ -2905,8 +2968,10 @@ static int event_run(apr_pool_t * _pconf
                      AP_SIG_GRACEFUL_STRING
                      " received.  Doing graceful restart");
         /* wake up the children...time to die.  But we'll have more soon */
-        ap_mpm_podx_killpg(pod, ap_daemons_limit, AP_MPM_PODX_GRACEFUL);
-
+        for (i = 0; i < num_buckets; i++) {
+            ap_mpm_podx_killpg(all_buckets[i].pod, ap_daemons_limit,
+                               AP_MPM_PODX_GRACEFUL);
+        }
 
         /* This is mostly for debugging... so that we know what is still
          * gracefully dealing with existing request.
@@ -2918,7 +2983,10 @@ static int event_run(apr_pool_t * _pconf
          * and a SIGHUP, we may as well use the same signal, because some user
          * pthreads are stealing signals from us left and right.
          */
-        ap_mpm_podx_killpg(pod, ap_daemons_limit, AP_MPM_PODX_RESTART);
+        for (i = 0; i < num_buckets; i++) {
+            ap_mpm_podx_killpg(all_buckets[i].pod, ap_daemons_limit,
+                               AP_MPM_PODX_RESTART);
+        }
 
         ap_reclaim_child_processes(1,  /* Start with SIGTERM */
                                    event_note_child_killed);
@@ -2937,7 +3005,10 @@ static int event_open_logs(apr_pool_t *
 {
     int startup = 0;
     int level_flags = 0;
+    int num_buckets = 0;
+    ap_listen_rec **listen_buckets;
     apr_status_t rv;
+    int i;
 
     pconf = p;
 
@@ -2954,14 +3025,62 @@ static int event_open_logs(apr_pool_t *
         return DONE;
     }
 
-    if (!one_process) {
-        if ((rv = ap_mpm_podx_open(pconf, &pod))) {
+    if (one_process) {
+        num_buckets = 1;
+    }
+    else if (retained->is_graceful) {
+        /* Preserve the number of buckets on graceful restarts. */
+        num_buckets = retained->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 DONE;
+    }
+
+    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;
         }
+        all_buckets[i].listeners = listen_buckets[i];
+    }
+
+    if (retained->max_buckets < num_buckets) {
+        int new_max, *new_ptr;
+        new_max = retained->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->num_buckets * sizeof(int));
+        retained->idle_spawn_rate = new_ptr;
+        retained->max_buckets = new_max;
     }
+    if (retained->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->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->num_buckets = num_buckets;
+
     /* for skiplist */
     srand((unsigned int)apr_time_now());
     return OK;
@@ -2993,7 +3112,6 @@ static int event_pre_config(apr_pool_t *
     if (!retained) {
         retained = ap_retained_data_create(userdata_key, sizeof(*retained));
         retained->max_daemons_limit = -1;
-        retained->idle_spawn_rate = 1;
     }
     ++retained->module_loads;
     if (retained->module_loads == 2) {
@@ -3007,6 +3125,7 @@ static int event_pre_config(apr_pool_t *
                          "atomics not working as expected - add32 of negative number");
             return HTTP_INTERNAL_SERVER_ERROR;
         }
+
         rv = apr_pollset_create(&event_pollset, 1, plog,
                                 APR_POLLSET_THREADSAFE | APR_POLLSET_NOCOPY);
         if (rv != APR_SUCCESS) {
@@ -3309,7 +3428,7 @@ static int event_check_config(apr_pool_t
     }
 
     /* ap_daemons_to_start > ap_daemons_limit checked in ap_mpm_run() */
-    if (ap_daemons_to_start < 0) {
+    if (ap_daemons_to_start < 1) {
         if (startup) {
             ap_log_error(APLOG_MARK, APLOG_WARNING | APLOG_STARTUP, 0, NULL, APLOGNO(00517)
                          "WARNING: StartServers of %d not allowed, "

Modified: httpd/httpd/branches/2.4.x/server/mpm/netware/mpm_netware.c
URL: http://svn.apache.org/viewvc/httpd/httpd/branches/2.4.x/server/mpm/netware/mpm_netware.c?rev=1705492&r1=1705491&r2=1705492&view=diff
==============================================================================
--- httpd/httpd/branches/2.4.x/server/mpm/netware/mpm_netware.c (original)
+++ httpd/httpd/branches/2.4.x/server/mpm/netware/mpm_netware.c Sat Sep 26 22:20:14 2015
@@ -916,6 +916,7 @@ static int netware_run(apr_pool_t *_pcon
     ap_log_error(APLOG_MARK, APLOG_INFO, 0, ap_server_conf, APLOGNO(00225)
             "Server built: %s", ap_get_server_built());
     ap_log_command_line(plog, s);
+    ap_log_mpm_common(s);
     show_server_data();
 
     mpm_state = AP_MPMQ_RUNNING;

Modified: httpd/httpd/branches/2.4.x/server/mpm/prefork/prefork.c
URL: http://svn.apache.org/viewvc/httpd/httpd/branches/2.4.x/server/mpm/prefork/prefork.c?rev=1705492&r1=1705491&r2=1705492&view=diff
==============================================================================
--- httpd/httpd/branches/2.4.x/server/mpm/prefork/prefork.c (original)
+++ httpd/httpd/branches/2.4.x/server/mpm/prefork/prefork.c Sat Sep 26 22:20:14 2015
@@ -48,6 +48,8 @@
 #include "ap_mmn.h"
 #include "apr_poll.h"
 
+#include <stdlib.h>
+
 #ifdef HAVE_TIME_H
 #include <time.h>
 #endif
@@ -86,14 +88,12 @@
 
 /* config globals */
 
-static apr_proc_mutex_t *accept_mutex;
 static int ap_daemons_to_start=0;
 static int ap_daemons_min_free=0;
 static int ap_daemons_max_free=0;
 static int ap_daemons_limit=0;      /* MaxRequestWorkers */
 static int server_limit = 0;
 static int mpm_state = AP_MPMQ_STARTING;
-static ap_pod_t *pod;
 
 /* data retained by prefork across load/unload of the module
  * allocated on first call to pre-config hook; located on
@@ -125,6 +125,15 @@ typedef struct prefork_retained_data {
 } prefork_retained_data;
 static prefork_retained_data *retained;
 
+typedef struct prefork_child_bucket {
+    ap_pod_t *pod;
+    ap_listen_rec *listeners;
+    apr_proc_mutex_t *mutex;
+} prefork_child_bucket;
+static int                   num_buckets; /* Number of listeners buckets */
+static prefork_child_bucket *all_buckets, /* All listeners buckets */
+                            *my_bucket;   /* Current child bucket */
+
 #define MPM_CHILD_PID(i) (ap_scoreboard_image->parent[i].pid)
 
 /* one_process --- debugging mode variable; can be set from the command line
@@ -222,14 +231,14 @@ static void clean_child_exit(int code)
         prefork_note_child_killed(/* slot */ 0, 0, 0);
     }
 
-    ap_mpm_pod_close(pod);
+    ap_mpm_pod_close(my_bucket->pod);
     chdir_for_gprof();
     exit(code);
 }
 
-static void accept_mutex_on(void)
+static apr_status_t accept_mutex_on(void)
 {
-    apr_status_t rv = apr_proc_mutex_lock(accept_mutex);
+    apr_status_t rv = apr_proc_mutex_lock(my_bucket->mutex);
     if (rv != APR_SUCCESS) {
         const char *msg = "couldn't grab the accept mutex";
 
@@ -243,11 +252,12 @@ static void accept_mutex_on(void)
             exit(APEXIT_CHILDFATAL);
         }
     }
+    return APR_SUCCESS;
 }
 
-static void accept_mutex_off(void)
+static apr_status_t accept_mutex_off(void)
 {
-    apr_status_t rv = apr_proc_mutex_unlock(accept_mutex);
+    apr_status_t rv = apr_proc_mutex_unlock(my_bucket->mutex);
     if (rv != APR_SUCCESS) {
         const char *msg = "couldn't release the accept mutex";
 
@@ -264,6 +274,7 @@ static void accept_mutex_off(void)
             exit(APEXIT_CHILDFATAL);
         }
     }
+    return APR_SUCCESS;
 }
 
 /* On some architectures it's safe to do unserialized accept()s in the single
@@ -272,9 +283,9 @@ static void accept_mutex_off(void)
  * when it's safe in the single Listen case.
  */
 #ifdef SINGLE_LISTEN_UNSERIALIZED_ACCEPT
-#define SAFE_ACCEPT(stmt) do {if (ap_listeners->next) {stmt;}} while(0)
+#define SAFE_ACCEPT(stmt) (ap_listeners->next ? (stmt) : APR_SUCCESS)
 #else
-#define SAFE_ACCEPT(stmt) do {stmt;} while(0)
+#define SAFE_ACCEPT(stmt) (stmt)
 #endif
 
 static int prefork_query(int query_code, int *result, apr_status_t *rv)
@@ -352,7 +363,7 @@ static int volatile die_now = 0;
 static void stop_listening(int sig)
 {
     mpm_state = AP_MPMQ_STOPPING;
-    ap_close_listeners();
+    ap_close_listeners_ex(my_bucket->listeners);
 
     /* For a graceful stop, we want the child to exit when done */
     die_now = 1;
@@ -477,7 +488,7 @@ static void set_signals(void)
 static int requests_this_child;
 static int num_listensocks = 0;
 
-static void child_main(int child_num_arg)
+static void child_main(int child_num_arg, int child_bucket)
 {
 #if APR_HAS_THREADS
     apr_thread_t *thd = NULL;
@@ -521,18 +532,26 @@ static void child_main(int child_num_arg
     apr_pool_create(&ptrans, pchild);
     apr_pool_tag(ptrans, "transaction");
 
+    /* close unused listeners and pods */
+    for (i = 0; i < num_buckets; i++) {
+        if (i != child_bucket) {
+            ap_close_listeners_ex(all_buckets[i].listeners);
+            ap_mpm_pod_close(all_buckets[i].pod);
+        }
+    }
+
     /* needs to be done before we switch UIDs so we have permissions */
     ap_reopen_scoreboard(pchild, NULL, 0);
-    lockfile = apr_proc_mutex_lockfile(accept_mutex);
-    status = apr_proc_mutex_child_init(&accept_mutex,
-                                       lockfile,
-                                       pchild);
+    status = SAFE_ACCEPT(apr_proc_mutex_child_init(&my_bucket->mutex,
+                                    apr_proc_mutex_lockfile(my_bucket->mutex),
+                                    pchild));
     if (status != APR_SUCCESS) {
+        lockfile = apr_proc_mutex_lockfile(my_bucket->mutex);
         ap_log_error(APLOG_MARK, APLOG_EMERG, status, ap_server_conf, APLOGNO(00155)
                      "Couldn't initialize cross-process lock in child "
                      "(%s) (%s)",
                      lockfile ? lockfile : "none",
-                     apr_proc_mutex_name(accept_mutex));
+                     apr_proc_mutex_name(my_bucket->mutex));
         clean_child_exit(APEXIT_CHILDFATAL);
     }
 
@@ -554,7 +573,7 @@ static void child_main(int child_num_arg
         clean_child_exit(APEXIT_CHILDSICK); /* assume temporary resource issue */
     }
 
-    for (lr = ap_listeners, i = num_listensocks; i--; lr = lr->next) {
+    for (lr = my_bucket->listeners, i = num_listensocks; i--; lr = lr->next) {
         apr_pollfd_t pfd = { 0 };
 
         pfd.desc_type = APR_POLL_SOCKET;
@@ -612,7 +631,7 @@ static void child_main(int child_num_arg
 
         if (num_listensocks == 1) {
             /* There is only one listener record, so refer to that one. */
-            lr = ap_listeners;
+            lr = my_bucket->listeners;
         }
         else {
             /* multiple listening sockets - need to poll */
@@ -710,7 +729,7 @@ static void child_main(int child_num_arg
          * while we were processing the connection or we are the lucky
          * idle server process that gets to die.
          */
-        if (ap_mpm_pod_check(pod) == APR_SUCCESS) { /* selected as idle? */
+        if (ap_mpm_pod_check(my_bucket->pod) == APR_SUCCESS) { /* selected as idle? */
             die_now = 1;
         }
         else if (retained->my_generation !=
@@ -726,7 +745,7 @@ static void child_main(int child_num_arg
 }
 
 
-static int make_child(server_rec *s, int slot)
+static int make_child(server_rec *s, int slot, int bucket)
 {
     int pid;
 
@@ -735,6 +754,8 @@ static int make_child(server_rec *s, int
     }
 
     if (one_process) {
+        my_bucket = &all_buckets[0];
+
         apr_signal(SIGHUP, sig_term);
         /* Don't catch AP_SIG_GRACEFUL in ONE_PROCESS mode :) */
         apr_signal(SIGINT, sig_term);
@@ -743,14 +764,15 @@ static int make_child(server_rec *s, int
 #endif
         apr_signal(SIGTERM, sig_term);
         prefork_note_child_started(slot, getpid());
-        child_main(slot);
+        child_main(slot, 0);
         /* NOTREACHED */
+        ap_assert(0);
+        return -1;
     }
 
     (void) ap_update_child_status_from_indexes(slot, 0, SERVER_STARTING,
                                                (request_rec *) NULL);
 
-
 #ifdef _OSD_POSIX
     /* BS2000 requires a "special" version of fork() before a setuid() call */
     if ((pid = os_fork(ap_unixd_config.user_name)) == -1) {
@@ -775,6 +797,8 @@ static int make_child(server_rec *s, int
     }
 
     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
@@ -797,9 +821,10 @@ static int make_child(server_rec *s, int
          * The pod is used for signalling the graceful restart.
          */
         apr_signal(AP_SIG_GRACEFUL, stop_listening);
-        child_main(slot);
+        child_main(slot, bucket);
     }
 
+    ap_scoreboard_image->parent[slot].bucket = bucket;
     prefork_note_child_started(slot, pid);
 
     return 0;
@@ -815,7 +840,7 @@ static void startup_children(int number_
         if (ap_scoreboard_image->servers[i][0].status != SERVER_DEAD) {
             continue;
         }
-        if (make_child(ap_server_conf, i) < 0) {
+        if (make_child(ap_server_conf, i, i % num_buckets) < 0) {
             break;
         }
         --number_to_start;
@@ -824,6 +849,8 @@ static void startup_children(int number_
 
 static void perform_idle_server_maintenance(apr_pool_t *p)
 {
+    static int bucket_make_child_record = -1;
+    static int bucket_kill_child_record = -1;
     int i;
     int idle_count;
     worker_score *ws;
@@ -874,7 +901,8 @@ static void perform_idle_server_maintena
          * shut down gracefully, in case it happened to pick up a request
          * while we were counting
          */
-        ap_mpm_pod_signal(pod);
+        bucket_kill_child_record = (bucket_kill_child_record + 1) % num_buckets;
+        ap_mpm_pod_signal(all_buckets[bucket_kill_child_record].pod);
         retained->idle_spawn_rate = 1;
     }
     else if (idle_count < ap_daemons_min_free) {
@@ -899,7 +927,10 @@ static void perform_idle_server_maintena
                     idle_count, total_non_dead);
             }
             for (i = 0; i < free_length; ++i) {
-                make_child(ap_server_conf, free_slots[i]);
+                bucket_make_child_record++;
+                bucket_make_child_record %= num_buckets;
+                make_child(ap_server_conf, free_slots[i],
+                           bucket_make_child_record);
             }
             /* 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
@@ -925,18 +956,10 @@ static int prefork_run(apr_pool_t *_pcon
 {
     int index;
     int remaining_children_to_start;
-    apr_status_t rv;
+    int i;
 
     ap_log_pid(pconf, ap_pid_fname);
 
-    /* Initialize cross-process accept lock */
-    rv = ap_proc_mutex_create(&accept_mutex, NULL, AP_ACCEPT_MUTEX_TYPE, NULL,
-                              s, _pconf, 0);
-    if (rv != APR_SUCCESS) {
-        mpm_state = AP_MPMQ_STOPPING;
-        return DONE;
-    }
-
     if (!retained->is_graceful) {
         if (ap_run_pre_mpm(s->process->pool, SB_SHARED) != OK) {
             mpm_state = AP_MPMQ_STOPPING;
@@ -953,12 +976,23 @@ static int prefork_run(apr_pool_t *_pcon
 
     if (one_process) {
         AP_MONCONTROL(1);
-        make_child(ap_server_conf, 0);
+        make_child(ap_server_conf, 0, 0);
         /* NOTREACHED */
+        ap_assert(0);
+        return DONE;
     }
-    else {
-    if (ap_daemons_max_free < ap_daemons_min_free + 1)  /* Don't thrash... */
-        ap_daemons_max_free = ap_daemons_min_free + 1;
+
+    /* Don't thrash since num_buckets depends on the
+     * system and the number of online CPU cores...
+     */
+    if (ap_daemons_limit < num_buckets)
+        ap_daemons_limit = num_buckets;
+    if (ap_daemons_to_start < num_buckets)
+        ap_daemons_to_start = num_buckets;
+    if (ap_daemons_min_free < num_buckets)
+        ap_daemons_min_free = num_buckets;
+    if (ap_daemons_max_free < ap_daemons_min_free + num_buckets)
+        ap_daemons_max_free = ap_daemons_min_free + 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
@@ -989,9 +1023,12 @@ static int prefork_run(apr_pool_t *_pcon
     ap_log_error(APLOG_MARK, APLOG_INFO, 0, ap_server_conf, APLOGNO(00164)
                 "Server built: %s", ap_get_server_built());
     ap_log_command_line(plog, s);
+    ap_log_mpm_common(s);
     ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, ap_server_conf, APLOGNO(00165)
                 "Accept mutex: %s (default: %s)",
-                apr_proc_mutex_name(accept_mutex),
+                (all_buckets[0].mutex)
+                    ? apr_proc_mutex_name(all_buckets[0].mutex)
+                    : "none",
                 apr_proc_mutex_defname());
 
     mpm_state = AP_MPMQ_RUNNING;
@@ -1048,7 +1085,8 @@ static int prefork_run(apr_pool_t *_pcon
                     /* 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,
+                               ap_get_scoreboard_process(child_slot)->bucket);
                     --remaining_children_to_start;
                 }
 #if APR_HAS_OTHER_CHILD
@@ -1089,7 +1127,6 @@ static int prefork_run(apr_pool_t *_pcon
 
         perform_idle_server_maintenance(pconf);
     }
-    } /* one_process */
 
     mpm_state = AP_MPMQ_STOPPING;
 
@@ -1122,7 +1159,9 @@ static int prefork_run(apr_pool_t *_pcon
         ap_close_listeners();
 
         /* kill off the idle ones */
-        ap_mpm_pod_killpg(pod, retained->max_daemons_limit);
+        for (i = 0; i < num_buckets; i++) {
+            ap_mpm_pod_killpg(all_buckets[i].pod, retained->max_daemons_limit);
+        }
 
         /* Send SIGUSR1 to the active children */
         active_children = 0;
@@ -1196,7 +1235,9 @@ static int prefork_run(apr_pool_t *_pcon
                     "Graceful restart requested, doing restart");
 
         /* kill off the idle ones */
-        ap_mpm_pod_killpg(pod, retained->max_daemons_limit);
+        for (i = 0; i < num_buckets; i++) {
+            ap_mpm_pod_killpg(all_buckets[i].pod, retained->max_daemons_limit);
+        }
 
         /* This is mostly for debugging... so that we know what is still
          * gracefully dealing with existing request.  This will break
@@ -1238,7 +1279,10 @@ static int prefork_open_logs(apr_pool_t
 {
     int startup = 0;
     int level_flags = 0;
+    ap_listen_rec **listen_buckets;
     apr_status_t rv;
+    char id[16];
+    int i;
 
     pconf = p;
 
@@ -1255,12 +1299,43 @@ static int prefork_open_logs(apr_pool_t
         return DONE;
     }
 
-    if ((rv = ap_mpm_pod_open(pconf, &pod))) {
+    if (one_process) {
+        num_buckets = 1;
+    }
+    else if (!retained->is_graceful) { /* Preserve the number of buckets
+                                          on graceful restarts. */
+        num_buckets = 0;
+    }
+    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 open pipe-of-death");
+                     "could not duplicate listeners");
         return DONE;
     }
+    all_buckets = apr_pcalloc(pconf, num_buckets *
+                                     sizeof(prefork_child_bucket));
+    for (i = 0; i < num_buckets; i++) {
+        if (!one_process && /* no POD in one_process mode */
+                (rv = ap_mpm_pod_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;
+        }
+        /* Initialize cross-process accept lock (safe accept needed only) */
+        if ((rv = SAFE_ACCEPT((apr_snprintf(id, sizeof id, "%i", i),
+                               ap_proc_mutex_create(&all_buckets[i].mutex,
+                                                    NULL, AP_ACCEPT_MUTEX_TYPE,
+                                                    id, s, pconf, 0))))) {
+            ap_log_error(APLOG_MARK, APLOG_CRIT | level_flags, rv,
+                         (startup ? NULL : s),
+                         "could not create accept mutex");
+            return DONE;
+        }
+        all_buckets[i].listeners = listen_buckets[i];
+    }
+
     return OK;
 }
 

Modified: httpd/httpd/branches/2.4.x/server/mpm/winnt/mpm_winnt.c
URL: http://svn.apache.org/viewvc/httpd/httpd/branches/2.4.x/server/mpm/winnt/mpm_winnt.c?rev=1705492&r1=1705491&r2=1705492&view=diff
==============================================================================
--- httpd/httpd/branches/2.4.x/server/mpm/winnt/mpm_winnt.c (original)
+++ httpd/httpd/branches/2.4.x/server/mpm/winnt/mpm_winnt.c Sat Sep 26 22:20:14 2015
@@ -1723,6 +1723,7 @@ static int winnt_run(apr_pool_t *_pconf,
         ap_log_error(APLOG_MARK, APLOG_NOTICE, 0, ap_server_conf, APLOGNO(00456)
                      "Server built: %s", ap_get_server_built());
         ap_log_command_line(plog, s);
+        ap_log_mpm_common(s);
 
         restart = master_main(ap_server_conf, shutdown_event, restart_event);
 

Modified: httpd/httpd/branches/2.4.x/server/mpm/worker/worker.c
URL: http://svn.apache.org/viewvc/httpd/httpd/branches/2.4.x/server/mpm/worker/worker.c?rev=1705492&r1=1705491&r2=1705492&view=diff
==============================================================================
--- httpd/httpd/branches/2.4.x/server/mpm/worker/worker.c (original)
+++ httpd/httpd/branches/2.4.x/server/mpm/worker/worker.c Sat Sep 26 22:20:14 2015
@@ -30,6 +30,9 @@
 #include "apr_thread_mutex.h"
 #include "apr_proc_mutex.h"
 #include "apr_poll.h"
+
+#include <stdlib.h>
+
 #define APR_WANT_STRFUNC
 #include "apr_want.h"
 
@@ -156,17 +159,31 @@ typedef struct worker_retained_data {
     /*
      * 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.
+     * 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;
+    int *idle_spawn_rate;
 #ifndef MAX_SPAWN_RATE
 #define MAX_SPAWN_RATE        (32)
 #endif
     int hold_off_on_exponential_spawning;
+    /*
+     * Current number of listeners buckets and maximum reached accross
+     * restarts (to size retained data according to dynamic num_buckets,
+     * eg. idle_spawn_rate).
+     */
+    int num_buckets, max_buckets;
 } worker_retained_data;
 static worker_retained_data *retained;
 
+typedef struct worker_child_bucket {
+    ap_pod_t *pod;
+    ap_listen_rec *listeners;
+    apr_proc_mutex_t *mutex;
+} worker_child_bucket;
+static worker_child_bucket *all_buckets, /* All listeners buckets */
+                           *my_bucket;   /* Current child bucket */
+
 #define MPM_CHILD_PID(i) (ap_scoreboard_image->parent[i].pid)
 
 /* The structure used to pass unique initialization info to each thread */
@@ -188,8 +205,6 @@ typedef struct {
 
 #define ID_FROM_CHILD_THREAD(c, t)    ((c * thread_limit) + t)
 
-static ap_pod_t *pod;
-
 /* The worker 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
@@ -217,9 +232,6 @@ static pid_t ap_my_pid; /* Linux getpid(
 static pid_t parent_pid;
 static apr_os_thread_t *listener_os_thread;
 
-/* Locks for accept serialization */
-static apr_proc_mutex_t *accept_mutex;
-
 #ifdef SINGLE_LISTEN_UNSERIALIZED_ACCEPT
 #define SAFE_ACCEPT(stmt) (ap_listeners->next ? (stmt) : APR_SUCCESS)
 #else
@@ -701,7 +713,7 @@ static void * APR_THREAD_FUNC listener_t
         clean_child_exit(APEXIT_CHILDSICK);
     }
 
-    for (lr = ap_listeners; lr != NULL; lr = lr->next) {
+    for (lr = my_bucket->listeners; lr != NULL; lr = lr->next) {
         apr_pollfd_t pfd = { 0 };
 
         pfd.desc_type = APR_POLL_SOCKET;
@@ -758,7 +770,7 @@ static void * APR_THREAD_FUNC listener_t
         /* We've already decremented the idle worker count inside
          * ap_queue_info_wait_for_idler. */
 
-        if ((rv = SAFE_ACCEPT(apr_proc_mutex_lock(accept_mutex)))
+        if ((rv = SAFE_ACCEPT(apr_proc_mutex_lock(my_bucket->mutex)))
             != APR_SUCCESS) {
 
             if (!listener_may_exit) {
@@ -767,9 +779,9 @@ static void * APR_THREAD_FUNC listener_t
             break;                    /* skip the lock release */
         }
 
-        if (!ap_listeners->next) {
+        if (!my_bucket->listeners->next) {
             /* Only one listener, so skip the poll */
-            lr = ap_listeners;
+            lr = my_bucket->listeners;
         }
         else {
             while (!listener_may_exit) {
@@ -839,7 +851,7 @@ static void * APR_THREAD_FUNC listener_t
                 resource_shortage = 1;
                 signal_threads(ST_GRACEFUL);
             }
-            if ((rv = SAFE_ACCEPT(apr_proc_mutex_unlock(accept_mutex)))
+            if ((rv = SAFE_ACCEPT(apr_proc_mutex_unlock(my_bucket->mutex)))
                 != APR_SUCCESS) {
 
                 if (listener_may_exit) {
@@ -863,7 +875,7 @@ static void * APR_THREAD_FUNC listener_t
             }
         }
         else {
-            if ((rv = SAFE_ACCEPT(apr_proc_mutex_unlock(accept_mutex)))
+            if ((rv = SAFE_ACCEPT(apr_proc_mutex_unlock(my_bucket->mutex)))
                 != APR_SUCCESS) {
                 int level = APLOG_EMERG;
 
@@ -880,7 +892,7 @@ static void * APR_THREAD_FUNC listener_t
         }
     }
 
-    ap_close_listeners();
+    ap_close_listeners_ex(my_bucket->listeners);
     ap_queue_term(worker_queue);
     dying = 1;
     ap_scoreboard_image->parent[process_slot].quiescing = 1;
@@ -1210,13 +1222,14 @@ static void join_start_thread(apr_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;
 
     mpm_state = AP_MPMQ_STARTING; /* for benefit of any hooks that run as this
                                    * child initializes
@@ -1225,12 +1238,20 @@ static void child_main(int child_num_arg
     ap_fatal_signal_child_setup(ap_server_conf);
     apr_pool_create(&pchild, pconf);
 
+    /* close unused listeners and pods */
+    for (i = 0; i < retained->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);
 
-    rv = SAFE_ACCEPT(apr_proc_mutex_child_init(&accept_mutex,
-                                               apr_proc_mutex_lockfile(accept_mutex),
-                                               pchild));
+    rv = SAFE_ACCEPT(apr_proc_mutex_child_init(&my_bucket->mutex,
+                                    apr_proc_mutex_lockfile(my_bucket->mutex),
+                                    pchild));
     if (rv != APR_SUCCESS) {
         ap_log_error(APLOG_MARK, APLOG_EMERG, rv, ap_server_conf, APLOGNO(00280)
                      "Couldn't initialize cross-process lock in child");
@@ -1338,7 +1359,7 @@ static void child_main(int child_num_arg
         apr_signal(SIGTERM, dummy_signal_handler);
         /* Watch for any messages from the parent over the POD */
         while (1) {
-            rv = ap_mpm_podx_check(pod);
+            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) {
@@ -1376,7 +1397,7 @@ static void child_main(int child_num_arg
     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;
 
@@ -1385,10 +1406,14 @@ static int make_child(server_rec *s, int
     }
 
     if (one_process) {
+        my_bucket = &all_buckets[0];
+
         set_signals();
         worker_note_child_started(slot, getpid());
-        child_main(slot);
+        child_main(slot, 0);
         /* NOTREACHED */
+        ap_assert(0);
+        return -1;
     }
 
     if ((pid = fork()) == -1) {
@@ -1410,6 +1435,8 @@ static int make_child(server_rec *s, int
     }
 
     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.
@@ -1424,10 +1451,12 @@ static int make_child(server_rec *s, int
         RAISE_SIGSTOP(MAKE_CHILD);
 
         apr_signal(SIGTERM, just_die);
-        child_main(slot);
+        child_main(slot, bucket);
         /* NOTREACHED */
+        ap_assert(0);
+        return -1;
     }
-    /* 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
@@ -1436,6 +1465,7 @@ static int make_child(server_rec *s, int
         worker_note_child_lost_slot(slot, pid);
     }
     ap_scoreboard_image->parent[slot].quiescing = 0;
+    ap_scoreboard_image->parent[slot].bucket = bucket;
     worker_note_child_started(slot, pid);
     return 0;
 }
@@ -1449,14 +1479,14 @@ static void startup_children(int number_
         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->num_buckets) < 0) {
             break;
         }
         --number_to_start;
     }
 }
 
-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;
@@ -1485,7 +1515,7 @@ static void perform_idle_server_maintena
         int all_dead_threads = 1;
         int child_threads_active = 0;
 
-        if (i >= retained->max_daemons_limit && totally_free_length == retained->idle_spawn_rate)
+        if (i >= retained->max_daemons_limit && totally_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
              */
@@ -1513,7 +1543,8 @@ static void perform_idle_server_maintena
                                    loop if no pid?  not much else matters */
                 if (status <= SERVER_READY &&
                         !ps->quiescing &&
-                        ps->generation == retained->my_generation) {
+                        ps->generation == retained->my_generation &&
+                        ps->bucket == child_bucket) {
                     ++idle_thread_count;
                 }
                 if (status >= SERVER_READY && status < SERVER_GRACEFUL) {
@@ -1522,8 +1553,8 @@ static void perform_idle_server_maintena
             }
         }
         active_thread_count += child_threads_active;
-        if (any_dead_threads && totally_free_length < retained->idle_spawn_rate
-                && free_length < MAX_SPAWN_RATE
+        if (any_dead_threads && totally_free_length < retained->idle_spawn_rate[child_bucket]
+                && free_length < MAX_SPAWN_RATE / num_buckets
                 && (!ps->pid               /* no process in the slot */
                     || ps->quiescing)) {   /* or at least one is going away */
             if (all_dead_threads) {
@@ -1579,12 +1610,13 @@ static void perform_idle_server_maintena
 
     retained->max_daemons_limit = last_non_dead + 1;
 
-    if (idle_thread_count > max_spare_threads) {
+    if (idle_thread_count > max_spare_threads / num_buckets) {
         /* Kill off one child */
-        ap_mpm_podx_signal(pod, AP_MPM_PODX_GRACEFUL);
-        retained->idle_spawn_rate = 1;
+        ap_mpm_podx_signal(all_buckets[child_bucket].pod,
+                           AP_MPM_PODX_GRACEFUL);
+        retained->idle_spawn_rate[child_bucket] = 1;
     }
-    else if (idle_thread_count < min_spare_threads) {
+    else if (idle_thread_count < min_spare_threads / num_buckets) {
         /* terminate the free list */
         if (free_length == 0) { /* scoreboard is full, can't fork */
 
@@ -1615,13 +1647,13 @@ static void perform_idle_server_maintena
                              ap_server_conf, APLOGNO(00288)
                              "scoreboard is full, not at MaxRequestWorkers");
             }
-            retained->idle_spawn_rate = 1;
+            retained->idle_spawn_rate[child_bucket] = 1;
         }
         else {
-            if (free_length > retained->idle_spawn_rate) {
-                free_length = retained->idle_spawn_rate;
+            if (free_length > retained->idle_spawn_rate[child_bucket]) {
+                free_length = retained->idle_spawn_rate[child_bucket];
             }
-            if (retained->idle_spawn_rate >= 8) {
+            if (retained->idle_spawn_rate[child_bucket] >= 8) {
                 ap_log_error(APLOG_MARK, APLOG_INFO, 0,
                              ap_server_conf, APLOGNO(00289)
                              "server seems busy, (you may need "
@@ -1632,7 +1664,7 @@ static void perform_idle_server_maintena
                              idle_thread_count, total_non_dead);
             }
             for (i = 0; i < free_length; ++i) {
-                make_child(ap_server_conf, free_slots[i]);
+                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
@@ -1640,17 +1672,18 @@ static void perform_idle_server_maintena
             if (retained->hold_off_on_exponential_spawning) {
                 --retained->hold_off_on_exponential_spawning;
             }
-            else if (retained->idle_spawn_rate < MAX_SPAWN_RATE) {
-                retained->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 {
-      retained->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)
 {
     ap_generation_t old_gen;
     int child_slot;
@@ -1694,22 +1727,25 @@ static void server_main_loop(int remaini
             }
             /* non-fatal death... note that it's gone in the scoreboard. */
             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);
 
                 worker_note_child_killed(child_slot, 0, 0);
-                ap_scoreboard_image->parent[child_slot].quiescing = 0;
+                ps = &ap_scoreboard_image->parent[child_slot];
+                ps->quiescing = 0;
                 if (processed_status == APEXIT_CHILDSICK) {
                     /* resource shortage, minimize the fork rate */
-                    retained->idle_spawn_rate = 1;
+                    retained->idle_spawn_rate[ps->bucket] = 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);
+                    make_child(ap_server_conf, child_slot, ps->bucket);
                     --remaining_children_to_start;
                 }
             }
@@ -1719,7 +1755,9 @@ static void server_main_loop(int remaini
                 if (processed_status == APEXIT_CHILDSICK
                     && old_gen == retained->my_generation) {
                     /* resource shortage, minimize the fork rate */
-                    retained->idle_spawn_rate = 1;
+                    for (i = 0; i < num_buckets; i++) {
+                        retained->idle_spawn_rate[i] = 1;
+                    }
                 }
 #if APR_HAS_OTHER_CHILD
             }
@@ -1758,25 +1796,20 @@ static void server_main_loop(int remaini
             continue;
         }
 
-        perform_idle_server_maintenance();
+        for (i = 0; i < num_buckets; i++) {
+            perform_idle_server_maintenance(i, num_buckets);
+        }
     }
 }
 
 static int worker_run(apr_pool_t *_pconf, apr_pool_t *plog, server_rec *s)
 {
+    int num_buckets = retained->num_buckets;
     int remaining_children_to_start;
-    apr_status_t rv;
+    int i;
 
     ap_log_pid(pconf, ap_pid_fname);
 
-    /* Initialize cross-process accept lock */
-    rv = ap_proc_mutex_create(&accept_mutex, NULL, AP_ACCEPT_MUTEX_TYPE, NULL,
-                              s, _pconf, 0);
-    if (rv != APR_SUCCESS) {
-        mpm_state = AP_MPMQ_STOPPING;
-        return DONE;
-    }
-
     if (!retained->is_graceful) {
         if (ap_run_pre_mpm(s->process->pool, SB_SHARED) != OK) {
             mpm_state = AP_MPMQ_STOPPING;
@@ -1790,9 +1823,18 @@ static int worker_run(apr_pool_t *_pconf
 
     restart_pending = shutdown_pending = 0;
     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 (ap_daemons_limit < num_buckets)
+        ap_daemons_limit = num_buckets;
+    if (ap_daemons_to_start < num_buckets)
+        ap_daemons_to_start = num_buckets;
+    if (min_spare_threads < threads_per_child * num_buckets)
+        min_spare_threads = threads_per_child * num_buckets;
+    if (max_spare_threads < min_spare_threads + threads_per_child * num_buckets)
+        max_spare_threads = min_spare_threads + threads_per_child * 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
@@ -1823,20 +1865,26 @@ static int worker_run(apr_pool_t *_pconf
     ap_log_error(APLOG_MARK, APLOG_INFO, 0, ap_server_conf, APLOGNO(00293)
                 "Server built: %s", ap_get_server_built());
     ap_log_command_line(plog, s);
+    ap_log_mpm_common(s);
     ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, ap_server_conf, APLOGNO(00294)
                 "Accept mutex: %s (default: %s)",
-                apr_proc_mutex_name(accept_mutex),
+                (all_buckets[0].mutex)
+                    ? apr_proc_mutex_name(all_buckets[0].mutex)
+                    : "none",
                 apr_proc_mutex_defname());
     mpm_state = AP_MPMQ_RUNNING;
 
-    server_main_loop(remaining_children_to_start);
+    server_main_loop(remaining_children_to_start, num_buckets);
     mpm_state = AP_MPMQ_STOPPING;
 
     if (shutdown_pending && !retained->is_graceful) {
         /* Time to shut down:
          * Kill child processes, tell them to call child_exit, etc...
          */
-        ap_mpm_podx_killpg(pod, ap_daemons_limit, AP_MPM_PODX_RESTART);
+        for (i = 0; i < num_buckets; i++) {
+            ap_mpm_podx_killpg(all_buckets[i].pod, ap_daemons_limit,
+                               AP_MPM_PODX_RESTART);
+        }
         ap_reclaim_child_processes(1, /* Start with SIGTERM */
                                    worker_note_child_killed);
 
@@ -1857,7 +1905,11 @@ static int worker_run(apr_pool_t *_pconf
 
         /* Close our listeners, and then ask our children to do same */
         ap_close_listeners();
-        ap_mpm_podx_killpg(pod, ap_daemons_limit, AP_MPM_PODX_GRACEFUL);
+
+        for (i = 0; i < num_buckets; i++) {
+            ap_mpm_podx_killpg(all_buckets[i].pod, ap_daemons_limit,
+                               AP_MPM_PODX_GRACEFUL);
+        }
         ap_relieve_child_processes(worker_note_child_killed);
 
         if (!child_fatal) {
@@ -1897,7 +1949,10 @@ static int worker_run(apr_pool_t *_pconf
          * way, try and make sure that all of our processes are
          * really dead.
          */
-        ap_mpm_podx_killpg(pod, ap_daemons_limit, AP_MPM_PODX_RESTART);
+        for (i = 0; i < num_buckets; i++) {
+            ap_mpm_podx_killpg(all_buckets[i].pod, ap_daemons_limit,
+                               AP_MPM_PODX_RESTART);
+        }
         ap_reclaim_child_processes(1, worker_note_child_killed);
 
         return DONE;
@@ -1922,8 +1977,10 @@ static int worker_run(apr_pool_t *_pconf
         ap_log_error(APLOG_MARK, APLOG_NOTICE, 0, ap_server_conf, APLOGNO(00297)
                      AP_SIG_GRACEFUL_STRING " received.  Doing graceful restart");
         /* wake up the children...time to die.  But we'll have more soon */
-        ap_mpm_podx_killpg(pod, ap_daemons_limit, AP_MPM_PODX_GRACEFUL);
-
+        for (i = 0; i < num_buckets; i++) {
+            ap_mpm_podx_killpg(all_buckets[i].pod, ap_daemons_limit,
+                               AP_MPM_PODX_GRACEFUL);
+        }
 
         /* This is mostly for debugging... so that we know what is still
          * gracefully dealing with existing request.
@@ -1935,7 +1992,10 @@ static int worker_run(apr_pool_t *_pconf
          * and a SIGHUP, we may as well use the same signal, because some user
          * pthreads are stealing signals from us left and right.
          */
-        ap_mpm_podx_killpg(pod, ap_daemons_limit, AP_MPM_PODX_RESTART);
+        for (i = 0; i < num_buckets; i++) {
+            ap_mpm_podx_killpg(all_buckets[i].pod, ap_daemons_limit,
+                               AP_MPM_PODX_RESTART);
+        }
 
         ap_reclaim_child_processes(1, /* Start with SIGTERM */
                                    worker_note_child_killed);
@@ -1953,7 +2013,11 @@ static int worker_open_logs(apr_pool_t *
 {
     int startup = 0;
     int level_flags = 0;
+    int num_buckets = 0;
+    ap_listen_rec **listen_buckets;
     apr_status_t rv;
+    char id[16];
+    int i;
 
     pconf = p;
 
@@ -1970,14 +2034,72 @@ static int worker_open_logs(apr_pool_t *
         return DONE;
     }
 
-    if (!one_process) {
-        if ((rv = ap_mpm_podx_open(pconf, &pod))) {
+    if (one_process) {
+        num_buckets = 1;
+    }
+    else if (retained->is_graceful) {
+        /* Preserve the number of buckets on graceful restarts. */
+        num_buckets = retained->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 DONE;
+    }
+
+    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;
         }
+        /* Initialize cross-process accept lock (safe accept needed only) */
+        if ((rv = SAFE_ACCEPT((apr_snprintf(id, sizeof id, "%i", i),
+                               ap_proc_mutex_create(&all_buckets[i].mutex,
+                                                    NULL, AP_ACCEPT_MUTEX_TYPE,
+                                                    id, s, pconf, 0))))) {
+            ap_log_error(APLOG_MARK, APLOG_CRIT | level_flags, rv,
+                         (startup ? NULL : s),
+                         "could not create accept mutex");
+            return DONE;
+        }
+        all_buckets[i].listeners = listen_buckets[i];
     }
+
+    if (retained->max_buckets < num_buckets) {
+        int new_max, *new_ptr;
+        new_max = retained->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->num_buckets * sizeof(int));
+        retained->idle_spawn_rate = new_ptr;
+        retained->max_buckets = new_max;
+    }
+    if (retained->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->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->num_buckets = num_buckets;
+
     return OK;
 }
 
@@ -2009,7 +2131,6 @@ static int worker_pre_config(apr_pool_t
     if (!retained) {
         retained = ap_retained_data_create(userdata_key, sizeof(*retained));
         retained->max_daemons_limit = -1;
-        retained->idle_spawn_rate = 1;
     }
     ++retained->module_loads;
     if (retained->module_loads == 2) {
@@ -2241,7 +2362,7 @@ static int worker_check_config(apr_pool_
     }
 
     /* ap_daemons_to_start > ap_daemons_limit checked in worker_run() */
-    if (ap_daemons_to_start < 0) {
+    if (ap_daemons_to_start < 1) {
         if (startup) {
             ap_log_error(APLOG_MARK, APLOG_WARNING | APLOG_STARTUP, 0, NULL, APLOGNO(00320)
                          "WARNING: StartServers of %d not allowed, "




Mime
View raw message