Horton

Re: pg11+: pg_ls_*dir LIMIT 1: temporary files .. not closed at end-of-transaction

Justin Pryzby <pryzby@telsasoft.com>

View thread View on postgr.es
From: Justin Pryzby <pryzby@telsasoft.com>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: pgsql-hackers@postgresql.org, Fabien COELHO <coelho@cri.ensmp.fr>, Alvaro Herrera <alvherre@2ndquadrant.com>, David Steele <david@pgmasters.net>, "Bossart, Nathan" <bossartn@amazon.com>, Thomas Munro <thomas.munro@gmail.com>
Date: 2020-03-29T20:12:15Z
Lists: pgsql-hackers 
On Sun, Mar 29, 2020 at 01:22:04PM -0400, Tom Lane wrote:
> Justin Pryzby <pryzby@telsasoft.com> writes:
> > On Sun, Mar 29, 2020 at 12:37:05PM -0400, Tom Lane wrote:
> >> After looking at the callers of pg_ls_dir_files, and noticing that
> >> it's already defined to ignore anything that's not a regular file,
> >> I think switching to lstat makes sense.
> 
> > Yea, only pg_ls_dir() shows special file types (and currently the others even
> > hide dirs).
> 
> > The essence of your patch is to ignore ENOENT, but you also changed to use
> > lstat(), which seems unrelated.  That means we'll now hide (non-broken)
> > symlinks.  Is that intentional/needed ?
> 
> Well, the following comment says "ignore anything but regular files",
> so I'm supposing that that is the behavior that we actually want here
> and failed to implement correctly.  There might be scope for
> additional directory-reading functions, but I'd think you'd want
> more information (such as the file type) returned from anything
> that doesn't act this way.

Maybe pg_stat_file() deserves similar attention ?  Right now, it'll fail on a
broken link.  If we changed it to lstat(), then it'd work, but it'd also show
metadata for the *link* rather than its target.

Patch proposed as v14-0001 patch here may be relevant:
https://www.postgresql.org/message-id/20200317190401.GY26184%40telsasoft.com
-    indicating if it is a directory.  Typical usages include:
+    indicating if it is a directory (or a symbolic link to a directory).
...

> In practice, since these directories shouldn't contain symlinks,
> it's likely moot.  The only place in PG data directories where
> we actually expect symlinks is pg_tablespace ... and that contains
> symlinks to directories, so that this function would ignore them
> anyway.

I wouldn't hesitate to make symlinks, at least in log.  It's surprising when
files are hidden, but I won't argue about the best behavior here.

I'm thinking of distributions or local configurations that use
/var/log/postgresql.  I didn't remember or didn't realize, but it looks like
debian's packages use logging_collector=off and then launch postmaster with
2> /var/log/postgres/...  It seems reasonable to do something like:
log/huge-querylog.csv => /zfs/compressed/...

-- 
Justin

Commits

  1. Teach pg_ls_dir_files() to ignore ENOENT failures from stat().

  2. Remove useless pfree()s at the ends of various ValuePerCall SRFs.

  3. Avoid holding a directory FD open across assorted SRF calls.

  4. Document pg_ls_*dir hiding of directories and special files

  5. Avoid holding a directory FD open across pg_ls_dir_files() calls.

  6. Adjust nodeFunctionscan.c to reset transient memory context between calls