[pgpool-hackers: 2497] wd_authkey bug (bug tracker #333)

[Yugo Nagata]

Hi Usama, 

There is the recent bug report about wd_authkey.

 0000333: watchdog fails to add node to master when wd_authkey is not an empty string; pgpool member shuts down
 http://www.pgpool.net/mantisbt/view.php?id=333

We have the same issue report recently from our client. In my analysis, this is a bug 
due to the commit [1]. This changed the definition of tv_sec that is used to check wd_authkey
so that this was affected by the clock of OS. So, if there is a lag between two nodes' clocks,
the wd_authkey check fails.

A simple solution is not to use tv_sec in the wd_authkey check as the attached patch.
However, one concern is that this is a specification change and that this also will break
back-compatibility. Of course, we can diallow watchdog cluster to have Pgpool-II of different
minor-versions. Although this is already implicit restriction of watchdog, we can make this
explicit restriction by checking other Pgpool-II node's version when receiving watchdog
packet.

What do you think about it?

[1] http://www.pgpool.net/pipermail/pgpool-committers/2017-April/003945.html

-- 
Yugo Nagata <nagata at sraoss.co.jp>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: fix_wd_authkey_bug.patch
Type: text/x-diff
Size: 1982 bytes
Desc: not available
URL: <http://www.sraoss.jp/pipermail/pgpool-hackers/attachments/20170824/784be1a9/attachment.bin>