Re: [patch] reorder tablespaces in basebackup tar stream for backup_label

From: Robert Haas <robertmhaas(at)gmail(dot)com>
To: Michael Banck <michael(dot)banck(at)credativ(dot)de>
Cc: PostgreSQL Development <pgsql-hackers(at)postgresql(dot)org>
Subject: Re: [patch] reorder tablespaces in basebackup tar stream for backup_label
Date: 2017-02-26 16:55:14
Message-ID: CA+TgmobavqABMHDiOd1CC_aL21xAqqNHZKi-pmet3DG+aNizhA@mail.gmail.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

On Tue, Feb 21, 2017 at 3:47 PM, Michael Banck
<michael(dot)banck(at)credativ(dot)de> wrote:
> currently, the backup_label and (I think) the tablespace_map files are
> (by design) conveniently located at the beginning of the main tablespace
> tarball when making a basebackup. However, (by accident or also by
> design?) the main tablespace is the last tarball[1] to be streamed via
> the BASE_BACKUP replication protocol command.
>
> For pg_basebackup, this is not a real problem, as either each tablespace
> is its own tarfile (in tar format mode), or the files are extracted
> anyway (in plain mode).
>
> However, third party tools using the BASE_BACKUP command might want to
> extract the backup_label, e.g. in order to figure out the START WAL
> LOCATION. If they make a big tarball for the whole cluster potentially
> including all external tablespaces, then the backup_label file is
> somewhere in the middle of it and it takes a long time for tar to
> extract it.
>
> So I am proposing the attached patch, which sends the base tablespace
> first, and then all the other external tablespaces afterwards, thus
> having base_backup be the first file in the tar in all cases. Does
> anybody see a problem with that?

Please add this to commitfest.postgresql.org so it doesn't get forgotten.

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Tom Lane 2017-02-26 17:07:45 Re: Should logtape.c blocks be of type long?
Previous Message Robert Haas 2017-02-26 16:49:21 Re: mat views stats