[Sks-devel] sks recon failed with Sys_error("Bad file descriptor") during reload

[Daniel Kahn Gillmor]

I'm running sks 1.1.3 on a debian system, and it uses the logrotate
package to ensure that the logs don't get out of control.

this means that they rotate daily, and when they rotate, the system
invokes:

    /etc/init.d/sks reload > /dev/null

This just sends a SIGHUP to both the db and the recon processes.

However, back on June 6th, i find this during the sighup:

2013-06-06 06:26:54 Added 1 hash-updates. Caught up to 1370514358.812619
2013-06-06 06:26:54 Raising Sys.Break -- PTree may be corrupted: Sys_error("Bad 
file descriptor")
2013-06-06 06:26:54 <get_missing_keys.catchup> callback interrupted by break.
2013-06-06 06:27:20 DB closed

And then recon dies without being restarted, and my keyserver falls out
of sync with the rest of the gossip network.

Clearly, one way to work around this would be to use a better logging
infrastructure than the crappy sysvinit+logrotate combination, and
another would be to run the daemons under some sort of process
supervision so that if they do fail they would at least be automagically
restarted after a sane period of time.

But sks shouldn't be crashing during a HUP either, right?  any thoughts
about what i might be able to do to debug this better or craft this into
a better report?

      --dkg

pgp6k8PZdfSa2.pgp
Description: PGP signature