MXS-1405: Take script_timeout into use

The script_timeout parameter is now used by all monitors.

include/maxscale/monitor.h

@@ -203,6 +203,7 @@ struct mxs_monitor
        * If yes, the next monitor loop starts early.  */
     bool active; /**< True if monitor is active */
     time_t journal_max_age; /**< Maximum age of journal file */
+    uint32_t script_timeout; /**< Timeout in seconds for the monitor scripts */
     struct mxs_monitor *next;     /**< Next monitor in the linked list */
 };
 
@@ -248,6 +249,7 @@ extern const char CN_BACKEND_WRITE_TIMEOUT[];
 extern const char CN_BACKEND_CONNECT_TIMEOUT[];
 extern const char CN_MONITOR_INTERVAL[];
 extern const char CN_JOURNAL_MAX_AGE[];
+extern const char CN_SCRIPT_TIMEOUT[];
 extern const char CN_SCRIPT[];
 extern const char CN_EVENTS[];
 

server/core/config.cc

@@ -242,6 +242,7 @@ const char *config_monitor_params[] =
     CN_EVENTS,
     CN_MONITOR_INTERVAL,
     CN_JOURNAL_MAX_AGE,
+    CN_SCRIPT_TIMEOUT,
     CN_BACKEND_CONNECT_TIMEOUT,
     CN_BACKEND_READ_TIMEOUT,
     CN_BACKEND_WRITE_TIMEOUT,
@@ -3141,8 +3142,7 @@ int create_new_monitor(CONFIG_CONTEXT *context, CONFIG_CONTEXT *obj, HASHTABLE*
         {
             char *endptr;
             long interval = strtol(journal_age, &endptr, 0);
-            /* The interval must be >0 because it is used as a divisor.
+
-                Perhaps a greater minimum value should be added? */
             if (*endptr == '\0' && interval > 0)
             {
                 monitorSetJournalMaxAge(monitor, (time_t)interval);
@@ -3161,6 +3161,30 @@ int create_new_monitor(CONFIG_CONTEXT *context, CONFIG_CONTEXT *obj, HASHTABLE*
                        obj->object, CN_JOURNAL_MAX_AGE, DEFAULT_JOURNAL_MAX_AGE);
         }
 
+        char *script_timeout = config_get_value(obj->parameters, CN_SCRIPT_TIMEOUT);
+        if (script_timeout)
+        {
+            char *endptr;
+            long interval = strtol(script_timeout, &endptr, 0);
+
+            if (*endptr == '\0' && interval > 0)
+            {
+                monitorSetScriptTimeout(monitor, (uint32_t)interval);
+            }
+            else
+            {
+                error_count++;
+                MXS_NOTICE("Invalid '%s' parameter for monitor '%s'",
+                           CN_SCRIPT_TIMEOUT, obj->object);
+            }
+        }
+        else
+        {
+            MXS_NOTICE("Monitor '%s' is missing the '%s' parameter, "
+                       "using default value of %d seconds.",
+                       obj->object, CN_SCRIPT_TIMEOUT, DEFAULT_SCRIPT_TIMEOUT);
+        }
+
         char *connect_timeout = config_get_value(obj->parameters, CN_BACKEND_CONNECT_TIMEOUT);
         if (connect_timeout)
         {

server/core/maxscale/monitor.h

@@ -32,6 +32,9 @@ MXS_BEGIN_DECLS
 /** Default maximum journal age in seconds */
 #define DEFAULT_JOURNAL_MAX_AGE 28800
 
+/** Default script execution timeout in seconds */
+#define DEFAULT_SCRIPT_TIMEOUT 90
+
 /**
  * Monitor network timeout types
  */
@@ -70,6 +73,7 @@ bool monitorRemoveParameter(MXS_MONITOR *monitor, const char *key);
 void monitorSetInterval (MXS_MONITOR *, unsigned long);
 bool monitorSetNetworkTimeout(MXS_MONITOR *, int, int);
 void monitorSetJournalMaxAge(MXS_MONITOR *mon, time_t value);
+void monitorSetScriptTimeout(MXS_MONITOR *mon, uint32_t value);
 
 /**
  * @brief Serialize a monitor to a file

server/core/monitor.cc

@@ -68,6 +68,7 @@ const char CN_BACKEND_WRITE_TIMEOUT[]    = "backend_write_timeout";
 const char CN_BACKEND_CONNECT_TIMEOUT[]  = "backend_connect_timeout";
 const char CN_MONITOR_INTERVAL[]         = "monitor_interval";
 const char CN_JOURNAL_MAX_AGE[]          = "journal_max_age";
+const char CN_SCRIPT_TIMEOUT[]           = "script_timeout";
 const char CN_SCRIPT[]                   = "script";
 const char CN_EVENTS[]                   = "events";
 
@@ -130,6 +131,7 @@ MXS_MONITOR* monitor_alloc(const char *name, const char *module)
     mon->connect_attempts = DEFAULT_CONNECTION_ATTEMPTS;
     mon->interval = DEFAULT_MONITOR_INTERVAL;
     mon->journal_max_age = DEFAULT_JOURNAL_MAX_AGE;
+    mon->script_timeout = DEFAULT_SCRIPT_TIMEOUT;
     mon->parameters = NULL;
     mon->server_pending_changes = false;
     spinlock_init(&mon->lock);
@@ -648,6 +650,17 @@ void monitorSetJournalMaxAge(MXS_MONITOR *mon, time_t value)
     mon->journal_max_age = value;
 }
 
+/**
+ * Set the maximum age of the monitor journal
+ *
+ * @param mon           The monitor instance
+ * @param interval      The journal age in seconds
+ */
+void monitorSetScriptTimeout(MXS_MONITOR *mon, uint32_t value)
+{
+    mon->script_timeout = value;
+}
+
 /**
  * Set Monitor timeouts for connect/read/write
  *
@@ -1143,7 +1156,7 @@ monitor_launch_script(MXS_MONITOR* mon, MXS_MONITOR_SERVERS* ptr, const char* sc
     char arg[strlen(script) + 1];
     strcpy(arg, script);
 
-    EXTERNCMD* cmd = externcmd_allocate(arg);
+    EXTERNCMD* cmd = externcmd_allocate(arg, mon->script_timeout);
 
     if (cmd == NULL)
     {
@@ -1196,12 +1209,24 @@ monitor_launch_script(MXS_MONITOR* mon, MXS_MONITOR_SERVERS* ptr, const char* sc
         externcmd_substitute_arg(cmd, "[$]SYNCEDLIST", nodelist);
     }
 
-    if (externcmd_execute(cmd))
+    int rv = externcmd_execute(cmd);
+
+    if (rv)
     {
-        MXS_ERROR("Failed to execute script '%s' on server state change event '%s'.",
+        if (rv == -1)
+        {
+            // Internal error
+            MXS_ERROR("Failed to execute script '%s' on server state change event '%s'",
                       script, mon_get_event_name(ptr));
         }
         else
+        {
+            // Script returned a non-zero value
+            MXS_ERROR("Script '%s' returned %d on event '%s'",
+                      script, rv, mon_get_event_name(ptr));
+        }
+    }
+    else
     {
         ss_dassert(cmd->argv != NULL && cmd->argv[0] != NULL);
         // Construct a string with the script + arguments