Author: mhabersack
Date: 2007-08-09 04:20:02 -0400 (Thu, 09 Aug 2007)
New Revision: 83716
Modified:
trunk/mod_mono/ChangeLog
trunk/mod_mono/src/mod_mono.c
Log:
2007-08-08 Joshua Tauberer <[EMAIL PROTECTED]>
* src/mod_mono.c: some fixes for the cross-process
locking and refactored process limits.
- Added a flag in the dashboard to ensure only one child process
issues an automated restart request to xsp.
- Don't assume the mutex is actually ever created in case of problems.
- Don't call apr_global_mutex_child_init if MOD_MONO_LOCKING_MECHANISM
is set, since for me this causes all locks to block forever.
- Improved some error messages and added one about command stream
corruption.
- When the auto-restart check cannot get a lock, don't return a
HTTP_SERVICE_UNAVAILABLE status code, since we've already successfully
sent page output.
- Don't reset the dashboard in start_xsp if no fork was necessary.
- Included dashboard info in the control panel.
- Refactored process limits in order to aid debugging on why it
doesn't work at all.
Modified: trunk/mod_mono/ChangeLog
===================================================================
--- trunk/mod_mono/ChangeLog 2007-08-09 08:05:54 UTC (rev 83715)
+++ trunk/mod_mono/ChangeLog 2007-08-09 08:20:02 UTC (rev 83716)
@@ -1,3 +1,22 @@
+2007-08-08 Joshua Tauberer <[EMAIL PROTECTED]>
+
+ * src/mod_mono.c: some fixes for the cross-process
+ locking and refactored process limits.
+ - Added a flag in the dashboard to ensure only one child process
+ issues an automated restart request to xsp.
+ - Don't assume the mutex is actually ever created in case of problems.
+ - Don't call apr_global_mutex_child_init if MOD_MONO_LOCKING_MECHANISM
+ is set, since for me this causes all locks to block forever.
+ - Improved some error messages and added one about command stream
+ corruption.
+ - When the auto-restart check cannot get a lock, don't return a
+ HTTP_SERVICE_UNAVAILABLE status code, since we've already successfully
+ sent page output.
+ - Don't reset the dashboard in start_xsp if no fork was necessary.
+ - Included dashboard info in the control panel.
+ - Refactored process limits in order to aid debugging on why it
+ doesn't work at all.
+
2007-08-06 Marek Habersack <[EMAIL PROTECTED]>
* configure.in: added --enable-gcov option to enable gcov(1)
Modified: trunk/mod_mono/src/mod_mono.c
===================================================================
--- trunk/mod_mono/src/mod_mono.c 2007-08-09 08:05:54 UTC (rev 83715)
+++ trunk/mod_mono/src/mod_mono.c 2007-08-09 08:20:02 UTC (rev 83716)
@@ -62,6 +62,7 @@
typedef struct {
uint32_t handled_requests;
time_t start_time;
+ char restart_issued;
} dashboard_data;
typedef struct xsp_data {
@@ -392,6 +393,7 @@
xsp->dashboard = apr_shm_baseaddr_get
(xsp->dashboard_shm);
xsp->dashboard->start_time = time (NULL);
xsp->dashboard->handled_requests = 0;
+ xsp->dashboard->restart_issued = 0;
goto restore_creds;
}
}
@@ -472,7 +474,7 @@
server->restart_mode = AUTORESTART_MODE_INVALID;
server->restart_requests = 0;
server->restart_time = 0;
-
+
ensure_dashboard_initialized (config, server, pool);
nservers = config->nservers + 1;
@@ -1362,27 +1364,39 @@
}
static void
-set_process_limits (int max_cpu_time, int max_memory)
-{
-#ifdef HAVE_SETRLIMIT
+set_process_limits2 (int resource, int max, char *name) {
struct rlimit limit;
- if (max_cpu_time > 0) {
+ if (max > 0) {
+#ifdef HAVE_SETRLIMIT
/* We don't want SIGXCPU */
- limit.rlim_cur = max_cpu_time;
- limit.rlim_max = max_cpu_time;
- DEBUG_PRINT (1, "Setting CPU time limit to %d", max_cpu_time);
- (void) setrlimit (RLIMIT_CPU, &limit);
+ limit.rlim_cur = max;
+ limit.rlim_max = max;
+ DEBUG_PRINT (1, "Setting %s limit to %d", name, max);
+ if (setrlimit (resource, &limit) == -1) {
+ if (errno == EPERM)
+ ap_log_error (APLOG_MARK, APLOG_ERR,
STATUS_AND_SERVER,
+ "Failed to set %s process limit
on mod-mono-server to %d: The value is greater than an existing hard limit",
name, max);
+ else
+ ap_log_error (APLOG_MARK, APLOG_ERR,
STATUS_AND_SERVER,
+ "Failed to set %s process limit
on mod-mono-server to %d.", name, max);
+ }
+#else
+ ap_log_error (APLOG_MARK, APLOG_ERR, STATUS_AND_SERVER,
+ "Setting %s process limit is not supported on
this platform.", name);
+#endif
}
+}
- if (max_memory > 0) {
- /* We don't want ENOMEM */
- limit.rlim_cur = max_memory;
- limit.rlim_max = max_memory;
- DEBUG_PRINT (1, "Setting memory limit to %d", max_memory);
- (void) setrlimit (RLIMIT_DATA, &limit);
- }
+static void
+set_process_limits (int max_cpu_time, int max_memory)
+{
+#ifndef HAVE_SETRLIMIT
+#define RLIMIT_CPU 0
+#define RLIMIT_DATA 0
#endif
+ set_process_limits2 (RLIMIT_CPU, max_cpu_time, "CPU time");
+ set_process_limits2 (RLIMIT_DATA, max_memory, "memory (data segment)");
}
static void
@@ -1813,7 +1827,7 @@
{
apr_socket_t *sock;
apr_status_t rv;
- int command;
+ int command = -1;
int result = FALSE;
apr_status_t input;
int status = 0;
@@ -1866,8 +1880,20 @@
if (start_wait_time < 2)
start_wait_time = 2;
- if (!conf->dashboard_mutex_initialized_in_child) {
- rv = apr_global_mutex_child_init (&conf->dashboard_mutex,
conf->dashboard_lock_file, pconf);
+ ensure_dashboard_initialized (config, conf, pconf);
+ if (conf->dashboard_mutex &&
!conf->dashboard_mutex_initialized_in_child) {
+ /* Avoiding to call apr_global_mutex_child_init is a hack since
in certain
+ * conditions it may lead to apache deadlock. Since we don't
know the exact cause
+ * and usually it is not necessary to use the environment
variable to work around
+ * the apr's default locking mechanism, we skip the call in
case the envvar is
+ * used.
+ */
+ if (!getenv ("MOD_MONO_LOCKING_MECHANISM")) {
+ rv = apr_global_mutex_child_init
(&conf->dashboard_mutex, conf->dashboard_lock_file, pconf);
+ } else {
+ DEBUG_PRINT (0, "Skipping apr_global_mutex_child_init
on '%s'", conf->dashboard_lock_file);
+ rv = APR_SUCCESS;
+ }
if (rv != APR_SUCCESS) {
ap_log_error (APLOG_MARK, APLOG_CRIT,
STATCODE_AND_SERVER (rv),
"Failed to initialize the dashboard mutex
'%s' in child process",
@@ -1885,8 +1911,11 @@
return HTTP_SERVICE_UNAVAILABLE;
DEBUG_PRINT (2, "No backend found, will start a new
copy.");
- rv = apr_global_mutex_lock (conf->dashboard_mutex);
- DEBUG_PRINT (1, "Acquired the %s lock for backend
start", conf->dashboard_lock_file);
+ if (conf->dashboard_mutex)
+ rv = apr_global_mutex_lock
(conf->dashboard_mutex);
+ else
+ rv = APR_SUCCESS;
+ DEBUG_PRINT (1, "Acquiring the %s lock for backend
start", conf->dashboard_lock_file);
if (rv != APR_SUCCESS) {
ap_log_error (APLOG_MARK, APLOG_CRIT,
STATCODE_AND_SERVER (rv),
@@ -1903,11 +1932,13 @@
/* give some time for warm-up */
DEBUG_PRINT (2, "Started new backend, sleeping %us to
let it configure", (unsigned)start_wait_time);
apr_sleep (apr_time_from_sec (start_wait_time));
- rv = apr_global_mutex_unlock (conf->dashboard_mutex);
- if (rv != APR_SUCCESS)
- ap_log_error (APLOG_MARK, APLOG_ALERT,
STATCODE_AND_SERVER (rv),
- "Failed to release %s lock, the
process may deadlock!",
- conf->dashboard_lock_file);
+ if (conf->dashboard_mutex) {
+ rv = apr_global_mutex_unlock
(conf->dashboard_mutex);
+ if (rv != APR_SUCCESS)
+ ap_log_error (APLOG_MARK, APLOG_ALERT,
STATCODE_AND_SERVER (rv),
+ "Failed to release %s
lock, the process may deadlock!",
+
conf->dashboard_lock_file);
+ }
} else
break; /* connected */
}
@@ -1930,23 +1961,29 @@
} while (input == sizeof (int32_t) && result == TRUE);
apr_socket_close (sock);
- if (input != sizeof (int32_t))
+ if (input != sizeof (int32_t)) {
+ ap_log_error (APLOG_MARK, APLOG_ERR, STATUS_AND_SERVER,
+ "Command stream corrupted, last command was %d",
command);
status = HTTP_INTERNAL_SERVER_ERROR;
+ }
if (conf->restart_mode > AUTORESTART_MODE_NONE) {
int do_restart = 0;
DEBUG_PRINT (2, "Auto-restart enabled for '%s', checking if
restart required", conf->alias);
+ ensure_dashboard_initialized (config, conf, pconf);
+ if (!conf->dashboard_mutex)
+ return status;
+
rv = apr_global_mutex_lock (conf->dashboard_mutex);
DEBUG_PRINT (1, "Acquired the %s lock for backend auto-restart
check", conf->dashboard_lock_file);
- ensure_dashboard_initialized (config, conf, pconf);
if (rv != APR_SUCCESS) {
ap_log_error (APLOG_MARK, APLOG_CRIT,
STATCODE_AND_SERVER (rv),
- "Failed to acquire %s lock, cannot
continue starting new process",
+ "Failed to acquire %s lock, cannot
continue auto-restart check",
conf->dashboard_lock_file);
- return HTTP_SERVICE_UNAVAILABLE;
+ return status;
}
if (conf->restart_mode == AUTORESTART_MODE_REQUESTS) {
@@ -1970,17 +2007,21 @@
}
}
- if (do_restart) {
+ if (do_restart && !conf->dashboard->restart_issued) {
/* we just need to stop it, it will be started at the
next request */
DEBUG_PRINT (0, "Stopping the backend '%s', it will be
started at the next request",
conf->alias);
+ ap_log_error (APLOG_MARK, APLOG_ALERT,
STATCODE_AND_SERVER (rv),
+ "Requesting termination of %s
mod-mono-server for restart...",
+ conf->alias);
+ conf->dashboard->restart_issued = 1;
terminate_xsp2 (r->server, conf->alias, 1);
}
rv = apr_global_mutex_unlock (conf->dashboard_mutex);
if (rv != APR_SUCCESS)
ap_log_error (APLOG_MARK, APLOG_ALERT,
STATCODE_AND_SERVER (rv),
- "Failed to release %s lock, the process
may deadlock!",
+ "Failed to release %s lock after
auto-restart check, the process may deadlock!",
conf->dashboard_lock_file);
}
@@ -2100,7 +2141,7 @@
if (!strcmp ("XXGLOBAL", xsp->alias) && config->auto_app ==
FALSE)
continue;
-
+
#ifdef APACHE13
sock = apr_pcalloc (pconf, sizeof (apr_socket_t));
#endif
@@ -2116,7 +2157,6 @@
i--;
continue; /* Wait for the previous to die */
}
-
apr_socket_close (sock);
xsp->status = FORK_SUCCEEDED;
} else {
@@ -2124,14 +2164,14 @@
/* need fork */
xsp->status = FORK_INPROCESS;
DEBUG_PRINT (0, "forking %s", xsp->alias);
- /* Give some time for clean up */
fork_mod_mono_server (pconf, xsp);
xsp->status = FORK_SUCCEEDED;
+ if (xsp->dashboard) {
+ xsp->dashboard->start_time = time (NULL);
+ xsp->dashboard->handled_requests = 0;
+ xsp->dashboard->restart_issued = 0;
+ }
}
- if (xsp->dashboard) {
- xsp->dashboard->start_time = time (NULL);
- xsp->dashboard->handled_requests = 0;
- }
}
}
@@ -2216,6 +2256,7 @@
xsp_data *xsp;
int i;
char *buffer;
+ apr_status_t rv;
if (strcmp (r->handler, "mono-ctrl"))
return DECLINED;
@@ -2241,8 +2282,32 @@
continue;
buffer = apr_psprintf (r->pool, "<li>%s: <a
href=\"?restart=%s\">"
- "Restart Server</a></li>\n",
xsp->alias, xsp->alias);
+ "Restart Server</a>",
xsp->alias, xsp->alias);
request_send_response_string(r, buffer);
+
+ ensure_dashboard_initialized (config, xsp, pconf);
+ if (xsp->dashboard_mutex) {
+ rv = apr_global_mutex_lock
(xsp->dashboard_mutex);
+ if (rv == APR_SUCCESS) {
+ if (xsp->restart_mode ==
AUTORESTART_MODE_REQUESTS) {
+ buffer = apr_psprintf (r->pool,
" [%d requests served; limit: %d]",
+
xsp->dashboard->handled_requests, xsp->restart_requests);
+ request_send_response_string(r,
buffer);
+ } else if (xsp->restart_mode ==
AUTORESTART_MODE_TIME) {
+ buffer = apr_psprintf (r->pool,
" [%ds time running; limit: %ds]",
+
(int)(time(NULL) - xsp->dashboard->start_time), (int)xsp->restart_time);
+ request_send_response_string(r,
buffer);
+ }
+
+ rv = apr_global_mutex_unlock
(xsp->dashboard_mutex);
+ if (rv != APR_SUCCESS)
+ ap_log_error (APLOG_MARK,
APLOG_ALERT, STATCODE_AND_SERVER (rv),
+ "Failed to
release %s lock after mono-ctrl output, the process may deadlock!",
+
xsp->dashboard_lock_file);
+ }
+ }
+
+ request_send_response_string(r, "</li>\n");
}
request_send_response_string (r, "</ul>\n");
_______________________________________________
Mono-patches maillist - [email protected]
http://lists.ximian.com/mailman/listinfo/mono-patches
