cpgx: improve killing and restarting

of corosync or aisexec, and optionally by starting exec directly
instead of via cman_tool

Signed-off-by: David Teigland <teigland@redhat.com>

1 files changed, 44 insertions, 9 deletions

diff --git a/cpgx/cpgx.c b/cpgx/cpgx.c
index 683a431..23748b7 100644
--- a/cpgx/cpgx.c
+++ b/cpgx/cpgx.c
@@ -32,6 +32,7 @@
 #include <time.h>
 #include <errno.h>
 #include <signal.h>
+#include <syslog.h>
 #include <sys/poll.h>
 #include <sys/time.h>
 #include <sys/wait.h>
@@ -45,6 +46,12 @@
 
 #include "list.h"
 
+#ifdef WHITETANK
+static char *exec_name = "aisexec";
+#else
+static char *exec_name = "corosync";
+#endif
+
 #define CLIENT_NALLOC		2
 #define MAX_NODES		8 /* not easily changed */
 #define DUMP_SIZE		(1024 * 1024)
@@ -131,6 +138,7 @@ static cpg_handle_t dct_cpg_handle;
 static int dct_cpg_client;
 static int dct_cpg_fd;
 
+static int exec_join = 0;
 static int prog_quit;
 static int cluster_down;
 static int opt_leave = 1;
@@ -1487,6 +1495,34 @@ int iterations_done(void)
 	return 0;
 }
 
+void restart_cluster(void)
+{
+	syslog(LOG_WARNING, "%ld killing %s", time(NULL), exec_name);
+	log_debug("killing %s", exec_name);
+
+	if (exec_name[0] == 'a')
+		system("killall -9 aisexec");
+	else
+		system("killall -9 corosync");
+
+	/* others should see us fail before we rejoin, not sure 10s will
+	   be enough for some people */
+
+	sleep(10);
+
+	syslog(LOG_WARNING, "%ld starting %s", time(NULL), exec_name);
+	log_debug("starting %s", exec_name);
+
+	if (!exec_join) {
+		system("cman_tool join -w");
+	} else {
+		if (exec_name[0] == 'a')
+			system("aisexec");
+		else
+			system("corosync");
+	}
+}
+
 void loop(void)
 {
 	void (*workfn) (int ci);
@@ -1592,13 +1628,7 @@ void loop(void)
 		if (we_should_die()) {
 			fflush(stdout);
 			fflush(stderr);
-			log_debug("killall -9 corosync/aisexec");
-			system("killall -9 corosync");
-			system("killall -9 aisexec");
-			/* others should see us fail before we rejoin */
-			sleep(10);
-			log_debug("cman_tool join -w");
-			system("cman_tool join -w");
+			restart_cluster();
 			exit(2);
 		}
 	}
@@ -1614,11 +1644,12 @@ void print_usage(void)
 	printf("  -e [0|1]  exit included in test [off|on], default 1\n");
 	printf("            (program exits without leaving cpg then rejoins)\n");
 	printf("  -d [0|1]  die included in test [off|on], default 0\n");
-	printf("            (kills corosync/aisexec, restarts with cman_tool join)\n");
+	printf("            (program kills and restarts %s)\n", exec_name);
 	printf("  -s <num>  sync up to num events, default %d\n",
 			    DEFAULT_SYNC_MAX);
 	printf("  -t <sec>  timeout after no dispatch for this many seconds, default 0 (never)\n");
 	printf("  -i <sec>  run for this many seconds, default 0 (forever)\n");
+	printf("  -p        run \"%s\" to join with -d rather than cman_tool\n", exec_name);
 	printf("  -c        continue after error\n");
 	printf("  -V        print version\n");
 	printf("\n");
@@ -1642,7 +1673,7 @@ int main(int argc, char **argv)
 	int optchar;
 
 	while (cont) {
-		optchar = getopt(argc, argv, "H:D:l:e:d:s:t:i:chV");
+		optchar = getopt(argc, argv, "H:D:l:e:d:s:t:i:pchV");
 
 		switch (optchar) {
 		case 'H':
@@ -1677,6 +1708,10 @@ int main(int argc, char **argv)
 			iterations_sec = atoi(optarg);
 			break;
 
+		case 'p':
+			exec_join = 1;
+			break;
+
 		case 'c':
 			continue_after_error = 1;
 			break;