[naviserver-commits] naviserver/nsd nsmain.c,1.19,1.20

[Zoran V. <vas...@us...>]

Update of /cvsroot/naviserver/naviserver/nsd
In directory sc8-pr-cvs1.sourceforge.net:/tmp/cvs-serv25813/nsd

Modified Files:
	nsmain.c 
Log Message:
Added alarm of 300 seconds for the watchdog on Darwin. 
When the alarm triggers, the watchdog re-checks the 
existence of the server process (by sending kill 0) and
clears nsconf-pid value if the process is gone. This is the
signal to re-start the server again. All this mess is needed
because under some strange circumstances, the WaitForServer()
never returns although the server process is already gone.
This smells of a broken signal delivery to me... 


Index: nsmain.c
===================================================================
RCS file: /cvsroot/naviserver/naviserver/nsd/nsmain.c,v
retrieving revision 1.19
retrieving revision 1.20
diff -C2 -d -r1.19 -r1.20
*** nsmain.c	8 Oct 2005 12:06:07 -0000	1.19
--- nsmain.c	8 Oct 2005 16:26:25 -0000	1.20
***************
*** 59,62 ****
--- 59,68 ----
  #define MAX_NUM_RESTARTS   256 /* Quit after somany unsuccessful restarts */
  
+ #ifdef __APPLE__
+ # define WAKEUP_IN_SECONDS 600 /* Wakeup watchdog after somuch seconds */
+ #else
+ # define WAKEUP_IN_SECONDS   0 /* Wakeup watchdog after somuch seconds */
+ #endif
+ 
  /*
   * Local functions defined in this file.
***************
*** 65,69 ****
  static int  StartWatchedServer(void);
  static void SysLog(int priority, char *fmt, ...);
! static void WatchdogSigtermHandler(int sig);
  static int  WaitForServer();
  
--- 71,76 ----
  static int  StartWatchedServer(void);
  static void SysLog(int priority, char *fmt, ...);
! static void WatchdogSIGTERMHandler(int sig);
! static void WatchdogSIGALRMHandler(int sig);
  static int  WaitForServer();
  
***************
*** 1128,1132 ****
   *----------------------------------------------------------------------
   *
!  * WatchdogSigtermHandler --
   *
   *      Handle SIGTERM and pass to server process.
--- 1135,1139 ----
   *----------------------------------------------------------------------
   *
!  * WatchdogSIGTERMHandler --
   *
   *      Handle SIGTERM and pass to server process.
***************
*** 1142,1150 ****
  
  static void 
! WatchdogSigtermHandler(int sig)
  {
      kill((pid_t) nsconf.pid, sig);
      watchdogExit = 1;
  }
  
  /*
--- 1149,1186 ----
  
  static void 
! WatchdogSIGTERMHandler(int sig)
  {
      kill((pid_t) nsconf.pid, sig);
      watchdogExit = 1;
  }
+ 
+ 
+ /*
+  *----------------------------------------------------------------------
+  *
+  * WatchdogSIGALRMHandler --
+  *
+  *      Handle SIGALRM to check existence of the nsconf.pid server
+  *      process.
+  *
+  * Results:
+  *      None.
+  *
+  * Side effects:
+  *      Zero-out the nsconf.pid element indicating absence of the
+  *      server process.
+  *
+  *----------------------------------------------------------------------
+  */
+ 
+ static void 
+ WatchdogSIGALRMHandler(int sig)
+ {
+     if (kill((pid_t) nsconf.pid, 0)) {
+         SysLog(LOG_WARNING, "watchdog: server %d terminated?", nsconf.pid);
+         nsconf.pid = 0;
+     }
+ }
+ 
  
  /*
***************
*** 1168,1180 ****
  WaitForServer()
  {
!     int ret, status;
!     pid_t pid;
!     char *msg;
  
      do {
          pid = waitpid(nsconf.pid, &status, 0);
!     } while (pid == -1 && errno == EINTR);
  
!     if (WIFEXITED(status)) {
          ret = WEXITSTATUS(status);
          msg = "exited";
--- 1204,1219 ----
  WaitForServer()
  {
!     int    ret, status;
!     pid_t  pid;
!     char  *msg;
  
      do {
          pid = waitpid(nsconf.pid, &status, 0);
!     } while (pid == -1 && errno == EINTR && nsconf.pid);
  
!     if (nsconf.pid == 0) {
!         msg = "terminated";
!         ret = -1; /* Alarm handler found no server present? */
!     } else if (WIFEXITED(status)) {
          ret = WEXITSTATUS(status);
          msg = "exited";
***************
*** 1213,1217 ****
  StartWatchedServer(void)
  {
!     unsigned int setSigterm = 0, startTime, numRestarts = 0, restartWait = 0;
  
      SysLog(LOG_NOTICE, "watchdog: started.");
--- 1252,1257 ----
  StartWatchedServer(void)
  {
!     unsigned int setSigHandlers=0, startTime, numRestarts=0, restartWait=0;
!     struct itimerval timer;
  
      SysLog(LOG_NOTICE, "watchdog: started.");
***************
*** 1240,1250 ****
          /*
           * Register SIGTERM handler so we can gracefully stop the server. 
!          * The watchdog will exit w/o stopping the server if got signalled 
!          * with any other signal, though.
           */
  
!         if (setSigterm == 0) {
!             setSigterm = 1;
!             ns_signal(SIGTERM, WatchdogSigtermHandler);
          }
  
--- 1280,1301 ----
          /*
           * Register SIGTERM handler so we can gracefully stop the server. 
!          * The watchdog passes the signal to the server, if possible.
!          *
!          * Register SIGALRM handler to wake up the watchdog to check if
!          * the server is still present. This tries to solve issues with 
!          * signal delivery on some systems where waitpid() fails to report
!          * process exitus (i.e. it is just stuck).
           */
  
!         if (setSigHandlers == 0) {
!             setSigHandlers = 1;
!             timer.it_interval.tv_sec = WAKEUP_IN_SECONDS;
!             timer.it_value.tv_sec  = timer.it_interval.tv_sec;
!             timer.it_value.tv_usec = timer.it_interval.tv_usec = 0;
!             if (timer.it_value.tv_sec || timer.it_value.tv_usec) {
!                 setitimer(ITIMER_REAL, &timer, NULL);
!                 ns_signal(SIGALRM, WatchdogSIGALRMHandler);
!             }
!             ns_signal(SIGTERM, WatchdogSIGTERMHandler);
          }