Re: Server crash in pg_replication_slot_advance function

On 16/02/18 10:55, amul sul wrote:
> On Fri, Feb 16, 2018 at 3:06 PM, amul sul <sulamul(at)gmail(dot)com> wrote:
>> On Fri, Feb 16, 2018 at 1:44 PM, tushar <tushar(dot)ahuja(at)enterprisedb(dot)com> wrote:
>>> Hi,
>> [....]
>>> postgres=# SELECT end_lsn FROM
>>> pg_replication_slot_advance('regression_slot1', '0/2000071');
>>> server closed the connection unexpectedly
>>> This probably means the server terminated abnormally
>>> before or while processing the request.
>>> !>
>>>
>>
>> I am able to reproduce this on the latest master head, the problem is in the
>> following hunk of pg_replication_slot_advance() where oldest LSN value
>> is accessed after releasing replication slot:
>>
>>
>> 476 if (moveto < startlsn)
>> 477 {
>> 478 ReplicationSlotRelease();
>> 479 ereport(ERROR,
>> 480 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
>> 481 errmsg("cannot move slot to %X/%X, minimum is %X/%X",
>> 482 (uint32) (moveto >> 32), (uint32) moveto,
>> 483 (uint32)
>> (MyReplicationSlot->data.confirmed_flush >> 32),
>> 484 (uint32)
>> (MyReplicationSlot->data.confirmed_flush))));
>> 485 }
>> 486
>>

Yeah, of course we can't use MyReplicationSlot after calling
ReplicationSlotRelease().

>
> Attached patch proposes a required fix.
>

Looks correct to me.

--
Petr Jelinek http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services