GNU bug report logs - #29335
'guix publish' workers occasionally crash

Previous Next

Package: guix;

Reported by: ludo <at> gnu.org (Ludovic Courtès)

Date: Fri, 17 Nov 2017 10:12:01 UTC

Severity: important

Done: ludo <at> gnu.org (Ludovic Courtès)

Bug is archived. No further changes may be made.

Full log


Message #18 received at 29335-done <at> debbugs.gnu.org (full text, mbox):

From: ludo <at> gnu.org (Ludovic Courtès)
To: 29335-done <at> debbugs.gnu.org
Subject: Re: bug#29335: 'guix publish' workers occasionally crash
Date: Sun, 19 Nov 2017 23:48:47 +0100
ludo <at> gnu.org (Ludovic Courtès) skribis:

> On berlin I’ve noticed that the ‘guix publish’ workers would
> occasionally stop working: the main thread would keep replying to HTTP
> requests, but the worker threads would no longer do anything, and would
> leave behind them a bunch of .tmp files in /var/cache/guix/publish.
>
> I captured the output of ‘guix publish’ (guix-0.13.0-8.357ab93) and the
> only clue I have is this:
>
> GET /6kl9ydqmgklcqhxswg6v5isq5n1ih5gp.narinfo
> In guix/workers.scm:
>      74:9  2 (_)
>     78:32  1 (_ srfi-34 #<condition &nix-connection-error [file: "/v…>)
> In unknown file:
>            0 (make-stack #t)
> ERROR: In procedure make-stack:
> ERROR: Throw to key `srfi-34' with args `(#<condition &nix-connection-error [file: "/var/guix/daemon-socket/socket" errno: 9] 3ba2ea0>)'.
> GET /fgiih42mg2sr82mbmzf56grvrf021im6.narinfo

Good news, this is fixed in 85f4f7b79040d982c6a655c898b4cd00d868fa9c.

This could be reproduced by running ‘guix publish’ with 10 workers or
more, and then triggering nar compression en masse with ‘guix weather’.

EBADF was due to a race condition in zlib.scm when closing gzip output
ports:

          ;; 'gzclose' closes the underlying file descriptor.  'close-port'
          ;; calls close(2) and gets EBADF, which we swallow.
          (gzclose gzfile)
          (ignore-EBADF (close-port port)))

There was a window after the ‘gzclose’ call during which the file
descriptor for GZFILE and PORT above could be reused for something else,
and then ‘close-port’ would close it.

Thanks,
Ludo’.




This bug report was last modified 7 years and 281 days ago.

Previous Next


GNU bug tracking system
Copyright (C) 1999 Darren O. Benham, 1997,2003 nCipher Corporation Ltd, 1994-97 Ian Jackson.