From: Fujii Masao <masao.fujii@oss.nttdata.com>
To: Kyotaro Horiguchi <horikyota.ntt@gmail.com>
Cc: pgsql-hackers@lists.postgresql.org
Subject: Re: Crash by targetted recovery
Date: Thu, 5 Mar 2020 19:51:11 +0900
Message-ID: <0db8f83f-5b66-75e9-6d99-93f19fc33165@oss.nttdata.com> (raw)
In-Reply-To: <20200305.120841.2172224092865554026.horikyota.ntt@gmail.com>
References: <a74bfa5d-3a86-90b8-286e-9ff8822bd8e1@oss.nttdata.com>
<20200228.121318.27427858650486112.horikyota.ntt@gmail.com>
<f7c69cf0-fcac-91b4-6fa2-f2560b6c49b5@oss.nttdata.com>
<20200305.120841.2172224092865554026.horikyota.ntt@gmail.com>
On 2020/03/05 12:08, Kyotaro Horiguchi wrote:
> At Mon, 2 Mar 2020 20:54:04 +0900, Fujii Masao <masao.fujii@oss.nttdata.com> wrote in
>>> And random access during StandbyMode ususally (always?) lets RecPtr go
>>> back. I'm not sure WaitForWALToBecomeAvailable works correctly if we
>>> don't have a file in pg_wal and the REDO point is far back by more
>>> than a segment from the initial checkpoint record.
>>
>> It works correctly. This is why WaitForWALToBecomeAvailable() uses
>> fetching_ckpt argument.
>
> Hmm. You're right. We start streaming from RedoStartLSN when
> fetching_ckpt. So that doesn't happen.
>
>>> If we go back to XLOG_FROM_ARCHIVE by random access, it correctly
>>> re-connects to the primary for the past segment.
>>
>> But this can lead to unnecessary restart of walreceiver. Since
>> fetching_ckpt ensures that the WAL file containing the REDO
>> starting record exists in pg_wal, there is no need to reconnect
>> to the primary when reading the REDO starting record.
>>
>> Is there other case where we need to go back to XLOG_FROM_ARCHIVE
>> by random access?
>
> I understand that the reconnection for REDO record is useless. Ok I
> take the !StandbyMode way.
>
> The attached is the test script that is changed to count the added
> test, and the slight revised main patch.
Thanks for the patch!
+ /* Wal receiver should not active when entring XLOG_FROM_ARCHIVE */
+ Assert(!WalRcvStreaming());
+1 to add this assertion check.
Isn't it better to always check this while trying to read WAL from
archive or pg_wal? So, what about the following change?
{
case XLOG_FROM_ARCHIVE:
case XLOG_FROM_PG_WAL:
+ /*
+ * WAL receiver should not be running while trying to
+ * read WAL from archive or pg_wal.
+ */
+ Assert(!WalRcvStreaming());
+
/* Close any old file we might have open. */
if (readFile >= 0)
+ lastSourceFailed = false; /* We haven't failed on the new source */
Is this really necessary? Since ReadRecord() always reset
lastSourceFailed to false, it seems not necessary.
- else if (currentSource == 0)
+ else if (currentSource == 0 ||
Though this is a *separate topic*, 0 should be XLOG_FROM_ANY?
There are some places where 0 is used as the value of currentSource.
IMO they should be updated so that XLOG_FROM_ANY is used instead of 0.
Regards,
--
Fujii Masao
NTT DATA CORPORATION
Advanced Platform Technology Group
Research and Development Headquarters
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Reply to all the recipients using the --to and --cc options:
reply via email
To: pgsql-hackers@postgresql.org
Cc: masao.fujii@oss.nttdata.com, horikyota.ntt@gmail.com, pgsql-hackers@lists.postgresql.org
Subject: Re: Crash by targetted recovery
In-Reply-To: <0db8f83f-5b66-75e9-6d99-93f19fc33165@oss.nttdata.com>
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
This inbox is served by DDX for PostgreSQL; see mirroring instructions
for how to clone and mirror all data and code used for this inbox