Re: pg_receivewal fail to streams when the partial file to write is not fully initialized present in the wal receiver directory

From: Kyotaro Horiguchi <horikyota(dot)ntt(at)gmail(dot)com>
To: bharath(dot)rupireddyforpostgres(at)gmail(dot)com
Cc: satyanarlapuram(at)gmail(dot)com, michael(at)paquier(dot)xyz, pgsql-hackers(at)lists(dot)postgresql(dot)org
Subject: Re: pg_receivewal fail to streams when the partial file to write is not fully initialized present in the wal receiver directory
Date: 2022-04-11 06:16:25
Message-ID: 20220411.151625.1056047583986161593.horikyota.ntt@gmail.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

Sorry for the terrible typos..

At Sat, 9 Apr 2022 18:03:01 +0530, Bharath Rupireddy <bharath(dot)rupireddyforpostgres(at)gmail(dot)com> wrote in
> On Tue, Jan 4, 2022 at 1:40 AM SATYANARAYANA NARLAPURAM
> <satyanarlapuram(at)gmail(dot)com> wrote:
> >
> > On Sun, Jan 2, 2022 at 11:56 PM Michael Paquier <michael(at)paquier(dot)xyz> wrote:
> >> Are you referring to the pre-padding when creating a new partial
> >> segment, aka when we write chunks of XLOG_BLCKSZ full of zeros until
> >> the file is fully created? What kind of error did you see? I guess
> >> that a write() with ENOSPC would be more likely, but you got a
> >> different problem?
> >
> > I see two cases, 1/ when no space is left on the device and 2/ when the process is taken down forcibly (a VM/container crash)
>
> Yeah, these cases can occur leaving uninitialized .partial files which
> can be a problem for both pg_receivewal and pg_basebackup that uses
> dir_open_for_write (CreateWalDirectoryMethod).
>
> >> I don't disagree with improving such cases, but we
> >> should not do things so as there is a risk of leaving behind an
> >> infinite set of segments in case of repeated errors
> >
> > Do you see a problem with the proposed patch that leaves the files behind, at least in my testing I don't see any files left behind?

I guess that Michael took this patch as creating a temp file with a
name such like "tmp.n" every time finding an incomplete file.

> With the proposed patch, it doesn't leave the unpadded .partial files.
> Also, the v2 patch always removes a leftover .partial.temp file before
> it creates a new one.
>
> >> , and partial
> >> segments are already a kind of temporary file.

I'm not sure this is true for pg_receivewal case. The .partial file
is not a temporary file but the current working file for the tool.

> > if the .partial file exists with not zero-padded up to the wal segment size (WalSegSz), then open_walfile fails with the below error. I have two options here, 1/ to continue padding the existing partial file and let it zero up to WalSegSz , 2/create a temp file as I did in the patch. I thought the latter is safe because it can handle corrupt cases as described below. Thoughts?

I think this patch shouldn't involve pg_basebackup. I agree to Cary
that deleting the erroring file should be fine.

We already "skipping" (complete = non-.partial) WAL files with a wrong
size in FindStreamingStart so we can error-out with suggesting a hint.

$ pg_receivewal -D xlog -p 5432 -h /tmp
pg_receivewal: error: segment file "0000000100000022000000F5.partial" has incorrect size 8404992
hint: You can continue after removing the file.

regards.

--
Kyotaro Horiguchi
NTT Open Source Software Center

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Filip Janus 2022-04-11 06:36:17 Re: Practical Timing Side Channel Attacks on Memory Compression
Previous Message Kyotaro Horiguchi 2022-04-11 06:13:14 Re: pg_receivewal fail to streams when the partial file to write is not fully initialized present in the wal receiver directory