[Mailman-Users] a corrupt pickle
barry at python.org
Tue Jan 9 00:58:17 CET 2007
-----BEGIN PGP SIGNED MESSAGE-----
On Jan 8, 2007, at 6:36 PM, Mark Sapiro wrote:
> Normally, if the config.pck is corrupt, we should fall back to the
> config.pck.last which should be OK. It is difficult to see how the
> logic could result in both files being corrupt, but strange things
> happen in a crash.
That sucks Ken! BTW, what version were you running?
I don't remember the details but around about 2.1.3->2.1.4 time
frame, we promoted SYNC_AFTER_WRITE to a full fledged mm_cfg.py
variable, but we left it set to No by default. This variable forces
an fsync when the config.pck file is written, which in turn ensures
that the data is flushed to your hard drive (assuming your OS doesn't
lie). Note that this is set to No by default because while being
much more secure about your data, it's a performance killer.
If both files are corrupt, you can thank your hardware. What
probably happened is that a subsequent load of config.pck returned
valid data but it was corrupted on the disk. Maybe the good data was
pulled from the disk buffers -- who knows? But then when Mailman
tried to write the new config.pck file, the old config.pck file was
already hosed and that ended up in config.pck.last.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.5 (Darwin)
-----END PGP SIGNATURE-----
More information about the Mailman-Users