Merge lp:~jeff-apple/openvista-gtm-integration/zombie into lp:openvista-gtm-integration

[jeff.apple]

[Jon Tai (jontai)]

I ran four simulated CPRS clients against the RPC broker simultaneously. All they do is log in, log out, and repeat. Before this fix, a zombie would appear within a few minutes -- sometimes one or two minutes, but no more than five minutes. After the fix, it's been 12 hours without a single zombie.

I'll merge this as soon as #683878 merges.

review: Approve

=== modified file 'mumps/ZISTCPS.m'
--- mumps/ZISTCPS.m	2009-06-05 20:09:12 +0000
+++ mumps/ZISTCPS.m	2010-12-02 00:27:56 +0000
@@ -1,4 +1,4 @@
-%ZISTCPS ;ISF/RWF MSC/JDA - DEVICE HANDLER TCP/IP SERVER CALLS ;22APR2009
+%ZISTCPS ;ISF/RWF MSC/JDA - DEVICE HANDLER TCP/IP SERVER CALLS ;01DEC2010
  ;;8.0;KERNEL;**78,118,127,225,275,388,MSC**;Jul 10, 1995
  Q
  ;
@@ -69,8 +69,11 @@
  U NIO S NIO("ZISTCP",0)=$KEY D LOG("Have port.")
  ;Start Listening
  W /LISTEN(1) S NIO("ZISTCP",1)=$KEY D LOG("Start Listening. "_NIO("ZISTCP",1))
- N ZC,ZR,IDX,DESC
+ N ZC,ZR,IDX,DESC,LDESC
  S ZC="ZSHOW ""D"":ZR"
+ X ZC
+ ; Find the listener file descriptor. The child must close it.
+ S LDESC="" F IDX=1:1:$O(ZR("D",""),-1) S:(ZR("D",IDX)["listener")&(ZR("D",IDX)["PORT="_SOCK) LDESC=$P($P(ZR("D",IDX),"DESC=",2)," ",1) Q:LDESC'=""
  ;Wait for connection
 LG2 S %A=0,EXIT=0 F  D  Q:%A!EXIT
  . U NIO:(SOCKET="listener")
@@ -91,7 +94,7 @@
  S DESC="" F IDX=1:1:$O(ZR("D",""),-1) S:ZR("D",IDX)[NIO("SOCK") DESC=$P($P(ZR("D",IDX),"DESC=",2)," ",1) Q:DESC'=""
  I DESC="" D LOG("Can not find file descriptor!") G LG2
  ;spawn child process
- S SPAWNID=$&openvista.gtmserver(DESC,"GTMLNCH^%ZISTCPS")
+ S SPAWNID=$&openvista.gtmserver(DESC,LDESC,"GTMLNCH^%ZISTCPS")
  L +^TMP("ZISTCPS",SPAWNID)
  S ^TMP("ZISTCPS",SPAWNID)=RTN
  L -^TMP("ZISTCPS",SPAWNID)

=== modified file 'src/libopenvista/gtmserver.c'
--- src/libopenvista/gtmserver.c	2009-04-23 04:45:12 +0000
+++ src/libopenvista/gtmserver.c	2010-12-02 00:27:56 +0000
@@ -26,7 +26,7 @@
 #define RETRY_EINTR(x) for(;;) if ((x)!=EINTR) break
 #define SEND_PID (write(fds[1], &grandchild_id, sizeof(int)))
 
-xc_long_t gtmserver(int count, int connfd, const char *cmd) {
+xc_long_t gtmserver(int count, int connfd, int listen_fd, const char *cmd) {
   int fds[2];
   int child_id;
   int grandchild_id;
@@ -49,6 +49,7 @@
   }
   if (child_id==0) {
     /* In child */
+    RETRY_EINTR(close(listen_fd));
     RETRY_EINTR(result=close(fds[0]) );
     if (result != 0) {
       perror("child close()");
@@ -102,7 +103,11 @@
       perror("parent close(fds[0])");
     }
     kill(child_id, SIGKILL);
-    waitpid(child_id, &status, 0);
+    while(child_id!=waitpid(child_id, &status, 0)) {
+      // Just keep trying this until it works. We could get interrupted in
+      // the syscall, which leaves the zombie around.
+    }
+
     return ret_val;
   }
 }

=== modified file 'src/libopenvista/openvista.xc.in'
--- src/libopenvista/openvista.xc.in	2009-07-08 04:08:14 +0000
+++ src/libopenvista/openvista.xc.in	2010-12-02 00:27:56 +0000
@@ -1,2 +1,2 @@
 @GTM_DIST@/libopenvista.so
-gtmserver: xc_long_t gtmserver(I:xc_long_t, I:xc_char_t*)
+gtmserver: xc_long_t gtmserver(I:xc_long_t, I:xc_long_t, I:xc_char_t*)