common: Debug ctdb_addr_to_str() using new function ctdb_external_trace()

We've seen this function report "Unknown family, 0" and then CTDB
disappeared without a trace.  If we can reproduce it then this might
help us to debug it.

The idea is that you do something like the following in /etc/sysconfig/ctdb:

  export CTDB_EXTERNAL_TRACE="/etc/ctdb/config/gcore_trace.sh"

When we hit this error than we call out to gcore to get a core file so
we can do forensics.  This might block CTDB for a few seconds.

Signed-off-by: Martin Schwenke <martin@meltin.net>

(This used to be ctdb commit 7895bc003f087ab2f3181df3c464386f59bfcc39)

5 files changed, 31 insertions, 0 deletions

diff --git a/ctdb/Makefile.in b/ctdb/Makefile.in
index 48cb57c5d6c..6c822601735 100755
--- a/ctdb/Makefile.in
+++ b/ctdb/Makefile.in
@@ -359,6 +359,7 @@ install: all $(PMDA_INSTALL)
 	if [ ! -f $(DESTDIR)$(etcdir)/ctdb/notify.sh ];then ${INSTALLCMD} -m 755 config/notify.sh $(DESTDIR)$(etcdir)/ctdb; fi
 	${INSTALLCMD} -m 755 config/debug-hung-script.sh $(DESTDIR)$(etcdir)/ctdb
 	if [ ! -f $(DESTDIR)$(etcdir)/ctdb/ctdb-crash-cleanup.sh ];then ${INSTALLCMD} -m 755 config/ctdb-crash-cleanup.sh $(DESTDIR)$(etcdir)/ctdb; fi
+	if [ ! -f $(DESTDIR)$(etcdir)/ctdb/gcore_trace.sh ];then ${INSTALLCMD} -m 755 config/gcore_trace.sh $(DESTDIR)$(etcdir)/ctdb; fi
 
 install_pmda:
 	$(INSTALLCMD) -m 755 -d $(PMDA_DEST_DIR)
diff --git a/ctdb/common/ctdb_util.c b/ctdb/common/ctdb_util.c
index ed322ac8b7f..71dee2b9544 100644
--- a/ctdb/common/ctdb_util.c
+++ b/ctdb/common/ctdb_util.c
@@ -59,6 +59,30 @@ void ctdb_fatal(struct ctdb_context *ctdb, const char *msg)
 	abort();
 }
 
+/* Invoke an external program to do some sort of tracing on the CTDB
+ * process.  This might block for a little while.  The external
+ * program is specified by the environment variable
+ * CTDB_EXTERNAL_TRACE.  This program should take one argument: the
+ * pid of the process to trace.  Commonly, the program would be a
+ * wrapper script around gcore.
+ */
+void ctdb_external_trace(void)
+{
+
+	const char * t = getenv("CTDB_EXTERNAL_TRACE");
+	char * cmd;
+
+	if (t == NULL) {
+		return;
+	}
+
+	cmd = talloc_asprintf(NULL, "%s %lu", t, (unsigned long) getpid());
+	DEBUG(DEBUG_WARNING,("begin external trace: %s\n", cmd));
+	system(cmd);
+	DEBUG(DEBUG_WARNING,("end external trace: %s\n", cmd));
+	talloc_free(cmd);
+}
+
 /*
   parse a IP:port pair
 */
@@ -555,6 +579,7 @@ char *ctdb_addr_to_str(ctdb_sock_addr *addr)
 		break;
 	default:
 		DEBUG(DEBUG_ERR, (__location__ " ERROR, unknown family %u\n", addr->sa.sa_family));
+		ctdb_external_trace();
 	}
 
 	return cip;
diff --git a/ctdb/config/gcore_trace.sh b/ctdb/config/gcore_trace.sh
new file mode 100755
index 00000000000..4d3e1d1d584
--- /dev/null
+++ b/ctdb/config/gcore_trace.sh
@@ -0,0 +1,3 @@
+#!/bin/sh
+
+gcore -o "/var/log/core" "$1" 2>&1 | logger -t "ctdb:gcore_trace"
diff --git a/ctdb/include/ctdb_private.h b/ctdb/include/ctdb_private.h
index 94b45c0d850..6f6d898167d 100644
--- a/ctdb/include/ctdb_private.h
+++ b/ctdb/include/ctdb_private.h
@@ -666,6 +666,7 @@ struct ctdb_fetch_handle {
 /* internal prototypes */
 void ctdb_set_error(struct ctdb_context *ctdb, const char *fmt, ...) PRINTF_ATTRIBUTE(2,3);
 void ctdb_fatal(struct ctdb_context *ctdb, const char *msg);
+void ctdb_external_trace(void);
 bool ctdb_same_address(struct ctdb_address *a1, struct ctdb_address *a2);
 int ctdb_parse_address(struct ctdb_context *ctdb,
 		       TALLOC_CTX *mem_ctx, const char *str,
diff --git a/ctdb/packaging/RPM/ctdb.spec.in b/ctdb/packaging/RPM/ctdb.spec.in
index 6a6398bdc7e..48cd0ab6b5e 100644
--- a/ctdb/packaging/RPM/ctdb.spec.in
+++ b/ctdb/packaging/RPM/ctdb.spec.in
@@ -123,6 +123,7 @@ rm -rf $RPM_BUILD_ROOT
 %config(noreplace) %{_sysconfdir}/ctdb/notify.sh
 %config(noreplace) %{_sysconfdir}/ctdb/debug-hung-script.sh
 %config(noreplace) %{_sysconfdir}/ctdb/ctdb-crash-cleanup.sh
+%config(noreplace) %{_sysconfdir}/ctdb/gcore_trace.sh
 %config(noreplace) %{_sysconfdir}/ctdb/functions
 %attr(755,root,root) %{initdir}/ctdb