cluster4 dlm dlm_stonith – should it really fence by turning node off?

Hello,

The dlm_stonith fencing helper is really convenient when Pacemaker is in use. Though, it doesn't quite work as I would expect ? when fencing is needed it requests a node to be turned off instead of rebooting. And it doesn't handle unfencing ? so automatic recovery is not possible (rebooted node could join the cluster cleanly later, provided quorum handling is properly configured in the cluster stack).

Preferably this behaviour should be configurable. I have hacked a work-around by (ab)using argv[0] ? when 'dlm_stonith' is called as 'dlm_stonith_reboot' the node would be rebooted instead of halting ? this works for me well-enough, but I don't think this is the right solution.

Any ideas how to solve that properly? An argument for the helper to be included in the config file? Or, maybe, just change the default behaviour?

Greets,
Jacek

Could you send the patch? Do you think the patch is not right or reboot is not right? If the later, what do you think is wrong with reboot?

I don't like the patch as probably it would be better to use extra argument to the dlm_stonith invokation in the dlm_controld file. I did not fix it that way, as I was not sure how arguments are supposed to be passed from dlm_controld configuration file to the fencing handlers. I'll sent the patch anyway.

If reboot is wrong ? I have assumed the 'off' was chosen for purpose and that my attempt to make it reboot instead could be a mistake. Though, I found no reason why reboot would be worse.

Greets,
Jacek

My workaround for the missing functionality (or wrong default):

[PATCH] dlm_stonith_{off,reboot} aliases for fence helper

Greets,
Jacek

Not so pretty hack to allow rebooting instead of halting the node fenced by dlm_stonith.

'dlm_stonith_reboot' can be now used instead of 'dlm_stonith' in the dlm.conf file to request node reboot.

'dlm_stonith_off' alias is also provided to explicitly request power-off.

---
 fence/Makefile         |   10 +++++++++-
 fence/stonith_helper.c |    8 +++++++-
 2 files changed, 16 insertions(+), 2 deletions(-)

index b4c59dd..2f24677 100644
@@ -6,6 +6,8 @@ BINDIR=$(PREFIX)/sbin
 BIN_TARGET = dlm_stonith
 #MAN_TARGET = dlm_stonith.8
 
+SYMLINKS = dlm_stonith_off dlm_stonith_reboot
+
 BIN_SOURCE = stonith_helper.c
 
 BIN_CFLAGS += -D_GNU_SOURCE -O2 -ggdb \

@@ -37,11 +39,14 @@ BIN_LDFLAGS += -Wl,-z,now -Wl,-z,relro -pie
 BIN_LDFLAGS += `xml2-config --libs`
 BIN_LDFLAGS += -ldl
 
-all: $(BIN_TARGET)
+all: $(BIN_TARGET) $(SYMLINKS)
 
 $(BIN_TARGET): $(BIN_SOURCE)
 	$(CC) $(BIN_SOURCE) $(BIN_CFLAGS) $(BIN_LDFLAGS) -o $@ -L.
 
+$(SYMLINKS): $(BIN_TARGET)
+	for link in $(SYMLINKS) ; do ln -sf $(BIN_TARGET) $$link ; done
+
 clean:
 	rm -f *.o *.so *.so.* $(BIN_TARGET)
 

@@ -53,5 +58,8 @@ install: all
 	$(INSTALL) -d $(DESTDIR)/$(BINDIR)
 	$(INSTALL) -d $(DESTDIR)/$(MANDIR)/man8
 	$(INSTALL) -c -m 755 $(BIN_TARGET) $(DESTDIR)/$(BINDIR)
+	for link in $(SYMLINKS) ; do \
+		ln -sf $(BIN_TARGET) $(DESTDIR)/$(BINDIR)/$$link ; \
+	done
 #	$(INSTALL) -m 644 $(MAN_TARGET) $(DESTDIR)/$(MANDIR)/man8/
 

index 5b384c1..73a2245 100644
@@ -16,6 +16,7 @@
 
 int nodeid;
 uint64_t fail_time;
+int turn_off = 1;
 
 #define MAX_ARG_LEN 1024
 

@@ -26,6 +27,11 @@ static int get_options(int argc, char *argv[])
 	char val[MAX_ARG_LEN];
 	char c;
 	int rv;
+	int arg0_l;
+
+	arg0_l = strlen(argv[0]);
+	if (arg0_l>7 && !strcmp(argv[0] + arg0_l - 7, "_reboot")) turn_off = 0;
+	else if (arg0_l>4 && !strcmp(argv[0] + arg0_l - 4, "_off")) turn_off = 1;
 
 	if (argc > 1) {
 		while ((c = getopt(argc, argv, "n:t:")) != -1) {

@@ -77,7 +83,7 @@ int main(int argc, char *argv[])
 	if (t >= fail_time)
 		return 0;
 
-	rv = stonith_api_kick_helper(nodeid, 300, 1);
+	rv = stonith_api_kick_helper(nodeid, 300, turn_off);
 	if (rv) {
 		fprintf(stderr, "kick_helper error %d nodeid %d\n", rv, nodeid);
 		openlog("stonith_helper", LOG_CONS | LOG_PID, LOG_DAEMON);

1.7.7.4

If it is 'off' by mistake and not for any better reason, then I think this will be the best ? to make this '0' by default. I have verified the Pacemaker sources twice ? 'off=0' means 'reboot'.

Effort to make this configurable may be postponed until anybody is interested in fencing by power-off.

Greets,
Jacek