[tabled patch 1/1] Stagger the start-daemon

[tabled patch 1/1] Stagger the start-daemon

[Pete Zaitcev]

My rule of thumb is that magic delays are evil or stupid, so I worked on
eliminating them from our scripts. However, in this case it's just not
worth it, because the result is that we have to wait way more than 100s
for several cycles of CLD timeouts to complete, not just one, before we
declare a failure. With this patch, all builds completed that I submitted
to Fedora build system.

Signed-off-by: Pete Zaitcev <zaitcev@redhat.com>

---
 test/start-daemon      |    4 ++++
 test/wait-for-listen.c |    7 ++-----
 2 files changed, 6 insertions(+), 5 deletions(-)

--- tabled-0.5git/test/start-daemon	2010-03-07 05:54:37.000000000 -0700
+++ tabled-tip/test/start-daemon	2010-05-25 15:02:26.001651210 -0600
@@ -18,6 +18,10 @@ fi
 
 # May be different on Solaris... like /usr/libexec or such.
 cld -d data/cld -P cld.pid -p auto --port-file=cld.port -E
+
+# With great sadness we have to use a delay, or else "100 s" happens.
+sleep 3
+
 chunkd -C $top_srcdir/test/chunkd-test.conf -E
 ../server/tabled -C $top_srcdir/test/tabled-test.conf -E
 
--- tabled-0.5git/test/wait-for-listen.c	2010-04-14 13:49:33.000000000 -0600
+++ tabled-tip/test/wait-for-listen.c	2010-06-17 21:19:18.245883298 -0600
@@ -133,12 +133,9 @@ int main(int argc, char **argv)
  		 * Vote in DB4 replication takes about 12-13s.
 		 * In addition we may have retries when tabled polls for
 		 * Chunk daemons to come up. On busy boxes we may miss 20s.
-		 * So, 25s should be plenty, and we used that for a while,
-		 * but sometimes a daemon can fail establishing a session
-		 * with CLD and a retry takes a minute.
 		 */
-		if (time(NULL) >= start_time + 100) {
-			fprintf(stderr, "server is not up after 100 s\n");
+		if (time(NULL) >= start_time + 25) {
+			fprintf(stderr, "server is not up after 25 s\n");
 			exit(1);
 		}
 

Re: [tabled patch 1/1] Stagger the start-daemon

[Jeff Garzik]

On 06/30/2010 10:49 AM, Pete Zaitcev wrote:
> My rule of thumb is that magic delays are evil or stupid, so I worked on
> eliminating them from our scripts. However, in this case it's just not
> worth it, because the result is that we have to wait way more than 100s
> for several cycles of CLD timeouts to complete, not just one, before we
> declare a failure. With this patch, all builds completed that I submitted
> to Fedora build system.
>
> Signed-off-by: Pete Zaitcev<zaitcev@redhat.com>
>
> ---
>   test/start-daemon      |    4 ++++
>   test/wait-for-listen.c |    7 ++-----
>   2 files changed, 6 insertions(+), 5 deletions(-)

applied