Thanks Michael=2C=0A=
=0A=
After the pg_rewind in the scenario I described=2C =0A=
=0A=
1) on System B (new Primary) I see=0A=
=0A=
Sat Apr 23 14:19:18 EDT 2016=0A=
=0A=
control file indicates =0A=
last check point WAL id : 0000000C00000009000000A3=0A=
=0A=
=A0client_addr |=A0=A0=A0=A0=A0=A0=A0=A0 backend_start=A0=A0=A0=A0=A0=A0=A0=
=A0 |=A0 state=A0 | sent_location | write_location | flush_location | repla=
y_location =0A=
-------------+-------------------------------+---------+---------------+---=
-------------+----------------+-----------------=0A=
=A010.19.0.1=A0=A0 | 2016-04-23 18:19:50.812509+00 | startup | 9/A30000D0=
=A0=A0=A0 | 9/A30000D0=A0=A0=A0=A0 | 9/A30000D0=A0=A0=A0=A0 | 9/A30000D0=0A=
=0A=
=0A=
2) whereas on System A after pg_rewind=A0 I see=0A=
=0A=
Sat Apr 23 14:19:54 EDT 2016=0A=
=0A=
control file indicates =0A=
=0A=
last check point WAL id : 0000000B00000009000000A3=0A=
=0A=
=A0pg_last_xlog_receive_location() =2C pg_last_xlog_replay_location() indic=
ates=0A=
=0A=
=A0pg_last_xlog_receive_location | pg_last_xlog_replay_location =0A=
-------------------------------+------------------------------=0A=
=A09/A3000000=A0=A0=A0=A0=A0=A0=A0=A0=A0=A0=A0=A0=A0=A0=A0=A0=A0=A0=A0 | 9/=
A30000D0=0A=
(1 row)=0A=
=0A=
Note the difference in timeline=0A=
=0A=
and then=2C=A0 as I described=2C=A0=A0 no WAL is replicated from B to A.=0A=
=0A=
Did you try this scenario yourself?=A0=A0=A0=A0 I hope you agree it is a bu=
g?=0A=
I will defer to you on what part of the code is the true cause=2C=0A=
but to me it looks very much as though pg_rewind ought to update the contro=
l file in this scenario.=0A=
That certainly does fix it.=0A=
If not that=2C=A0=A0 then what?=0A=
=0A=
Cheers=2C=A0=A0 John=0A=
=0A=
----------------------------------------=0A=
> Date: Mon=2C 25 Apr 2016 16:23:58 +0900=0A=
> Subject: Re: [BUGS] BUG #14109: pg_rewind fails to update target control =
file in one scenario=0A=
> From: michael.paquier@gmail.com=0A=
> To: johnlumby@hotmail.com=0A=
> CC: pgsql-bugs@postgresql.org=0A=
>=0A=
> On Mon=2C Apr 25=2C 2016 at 4:25 AM=2C <johnlumby@hotmail.com> wrote:=0A=
>> However=2C what I believe *is* needed is to update the target control fi=
le=0A=
>> with the new timeline and other information from the source.=0A=
>=0A=
> No=2C this is incorrect. There is no need to update the control file of=
=0A=
> a node that has not been rewound=2C and pg_rewind should not mess up=0A=
> with that if there is no divergence point between the target and the=0A=
> source nodes or it would update the minimum recovery point of a node=0A=
> without real need to do so. It should be able to join back the cluster=0A=
> depending on its initial shutdown state (when you shut down systemA).=0A=
> What are the logs of your system A telling you regarding its startup=0A=
> state?=0A=
> --=0A=
> Michael=0A=
=