[pgpool-general: 2024] Re: Problem with watchdog...

[Tatsuo Ishii]

> You said: "I killed server2 pgpool-II parent process by kill -9"
> Try "killall -9 pgpool" in active server...

Sorry for confusion but in my case server2 is the active.
--
Tatsuo Ishii
SRA OSS, Inc. Japan
English: http://www.sraoss.co.jp/index_en.php
Japanese: http://www.sraoss.co.jp

> The stand by will grow up the delegate IP and the same will still on old
> active server and will conflict.
> So I am using a shell script in crontab to check this and restart pgpool if
> necessary.
> 
> 
> 2013/8/12 Tatsuo Ishii <ishii at postgresql.org>
> 
>> > On 08/05/13 14:07, Fernando Buzon wrote:
>> >>
>> >> FINAL:
>> >> Like I said, all is working nice.
>> >> And now I am with the 2 pgpools up and working again.
>> >> The escaled pgpool is pgpool-01.
>> >> I stop it with "killall -9 pgpool" and now wd_lifecheck worked fine on
>> >> pgpool-02!
>> >> I dont now what was the problem early, but now is working!
>> >>
>> >
>> > Maybe you fix some small issue in the config during your testing.
>> >
>> >> log on pgpool-02:
>> >> 2013-08-05 17:52:42 LOG:   pid 11524: wd_lifecheck: lifecheck failed 3
>> >> times. pgpool 1 (10.0.0.21:5432 <http://10.0.0.21:5432>) seems not to
>> be working
>> >> 2013-08-05 17:52:42 LOG:   pid 11524: wd_escalation: escalated to
>> master pgpool
>> >> 2013-08-05 17:52:42 LOG:   pid 11524: wd_escalation:  escalated to
>> >> delegate_IP holder
>> >> 2013-08-05 17:52:52 LOG:   pid 11524: wd_lifecheck: lifecheck failed 3
>> >> times. pgpool 1 (10.0.0.21:5432 <http://10.0.0.21:5432>) seems not to
>> be working
>> >>
>> >> So rest only one problem, that is how to down delegate_ip from the
>> pgpool-01?
>> >> Because both servers is responding to delegate_ip.
>> >
>> > Well, the reason it doesn't get removed on pgpool-01 is because the
>> killall -9
>> > kills the pgpool processes including the watchdog without any hope of
>> them
>> > running the ifconfig down command.
>> >
>> > That said, you just need to run the ifconfig down command on pgpool-01.
>> >
>> > I'm sure what you're trying to simulate is a crash, but I'm not sure
>> killing
>> > ALL the pgpool processes with -9 is a good simulation, because more
>> likely
>> > only one of the backends would crash.
>> >
>> > Maybe one of the other folks on the list can suggest a better simulation
>> for a
>> > crashing pgpool service.
>>
>> I have tried with pgpool-II 3.3.0 to test the case. Initially
>> "server2" is the watchdog active, and "server1" is the watchdog
>> "standby".
>>
>> I killed server2 pgpool-II parent process by kill -9.
>>
>> - server2 releases the VIP. server2 watchdog goes to "down" status.
>>
>> - server1 becomes active and grab the VIP.
>>
>> So my guess is, 3.2's watchdog is not capable to handle the situation.
>>
>> Pgpool-II 3.3's watchdog is much more enhanced than 3.2's. I recommend
>> to use 3.3 if you want to seriously use watchdog.
>> --
>> Tatsuo Ishii
>> SRA OSS, Inc. Japan
>> English: http://www.sraoss.co.jp/index_en.php
>> Japanese: http://www.sraoss.co.jp
>>