Обсуждение: basebackup: add missing deflateEnd() in gzip compression sink

On Fri, Mar 20, 2026 at 10:02:44AM -0700, Jianghua Yang wrote:
> While reviewing the backup compression backends, I noticed that the gzip
> sink (basebackup_gzip.c) never calls deflateEnd(), unlike the lz4 and
> zstd sinks which properly free their compression contexts.
>
> This means each archive (one per tablespace) leaks ~256KB of zlib
> internal state until the backup's memory context is destroyed. On the
> ERROR path, the zlib context is not released at all since there is no
> gzip-specific cleanup callback.

Yes, you are right on this one.  This code could be better in cleaning
up the resources it is working on (and note that all the other
gzip-related code paths call the end() part if doing an init() or an
init2()).  Leaking that once per archive may not look like a big deal,
but this is backend-side code we are talking about, and on a
persistent replication connection it is not really cool to lose the
context of this data with garbage coming from zlib piling up, so based
on the argument of the permanent connection my take is that this
cleanup warrants a backpatch.
--
Michael

  Thanks for reviewing, Michael.

  > based on the argument of the permanent connection my take is that
  > this cleanup warrants a backpatch.

  Agreed. The current code leaks zlib internal state on every archive,
  and on a long-lived replication connection those allocations just
  pile up with no cleanup until the connection ends.

Michael Paquier <michael@paquier.xyz> 于2026年3月20日周五 23:09写道：

On Fri, Mar 20, 2026 at 10:02:44AM -0700, Jianghua Yang wrote:
> While reviewing the backup compression backends, I noticed that the gzip
> sink (basebackup_gzip.c) never calls deflateEnd(), unlike the lz4 and
> zstd sinks which properly free their compression contexts.
>
> This means each archive (one per tablespace) leaks ~256KB of zlib
> internal state until the backup's memory context is destroyed. On the
> ERROR path, the zlib context is not released at all since there is no
> gzip-specific cleanup callback.

Yes, you are right on this one.  This code could be better in cleaning
up the resources it is working on (and note that all the other
gzip-related code paths call the end() part if doing an init() or an
init2()).  Leaking that once per archive may not look like a big deal,
but this is backend-side code we are talking about, and on a
persistent replication connection it is not really cool to lose the
context of this data with garbage coming from zlib piling up, so based
on the argument of the permanent connection my take is that this
cleanup warrants a backpatch.
--
Michael

On Sat, Mar 21, 2026 at 02:22:25PM -0700, Jianghua Yang wrote:
> v1-0001: basebackup: add missing deflateEnd() calls in gzip compression
> sink

After double-checking the whole code, I agree that this is a good
practice to have in the tree.  However, the issue is not worth
bothering in back-branches as the server-side base backup gzip code
relies on allocation and free callbacks, with zlib internals doing
nothing with fds or more persistent states as far as I have read its
code.  For the current use, we'd bloat this data once per tablespace
in a single base backup, safe even if the connection is persistent
(missed that in my first message).

What I am more worried about are future callers of this code, though,
and we care about having a end() call for each matching init[2]() call
in the tree in all the places that rely on gzip internals.  So that's
a good practice on consistency ground, at least.  For these reasons,
applied that on HEAD.

> v1-0002: pg_basebackup: add missing close() for incremental manifest
> file

This one does not matter.  This resource is for a backup manifest and
we are talking about a single one for a single invocation of the
binary.
--
Michael