Re: post 3.1 xlator/io-threads instability on NetBSD (was: NULL frame->conn cause a crash)

[Anand Avati <anand.avati@xxxxxxxxx>]

I'm not sure how good/efficient the pthread implementation in NetBSD is. Also, do you know if there are any NetBSD specifics about using pthread_spinlock() APIs on it? Lot of the issues you have previously described can occur if there is a broken implementation of pthread spinlocks (which seems to get masked without io-threads)

Avati

On Wed, Jul 6, 2011 at 8:41 PM, Emmanuel Dreyfus <manu@xxxxxxxxxx> wrote:

On Tue, Jul 05, 2011 at 09:32:23AM +0000, Emmanuel Dreyfus wrote:
> Let us forget my previous batch of crashes that were with 3.2.1. This
> one iw with today's code from git:

I made some progress on that problem. Removing the io-threads xlator
fixes the stability problem. It also greatly improves performances
while the server load decrease from 14 to 3 for the same operations.

It seems there have been some refactoring of xlatorx/io-threads since 3.1.
It does not use sem_timedwait() anymore, that was replaced by
pthread_cond_timedwait().

Anyone has ideas here?

--
Emmanuel Dreyfus
manu@xxxxxxxxxx

_______________________________________________
Gluster-devel mailing list
Gluster-devel@xxxxxxxxxx
https://lists.nongnu.org/mailman/listinfo/gluster-devel