[Gluster-devel] 0-poll: index not found for fd

[Emmanuel Dreyfus]

Emmanuel Dreyfus <manu at netbsd.org> wrote:

> I guess I know need to track event_pool->reg[] usage. Hints appreciated.

On a brick that cause the disconnect, here is event_poll initialization at
startup time:

[2013-04-25 04:31:55.493219] E [event-poll.c:189:event_register_poll]  
    0-XXXmanu: event_pool = 0xbb741080, event_pool->reg[1] = 7
[2013-04-25 04:31:55.496896] E [event-poll.c:189:event_register_poll] 
    0-XXXmanu: event_pool = 0xbb741080, event_pool->reg[2] = 11
[2013-04-25 04:31:55.504432] E [event-poll.c:189:event_register_poll] 
    0-XXXmanu: event_pool = 0xbb741080, event_pool->reg[3] = 13
[2013-04-25 04:31:59.179970] E [event-poll.c:189:event_register_poll] 
    0-XXXmanu: event_pool = 0xbb741080, event_pool->reg[4] = 8
[2013-04-25 04:32:22.723378] E [event-poll.c:189:event_register_poll] 
    0-XXXmanu: event_pool = 0xbb741080, event_pool->reg[5] = 12

Then before the brick close the TCP connexion with the client:

[2013-04-25 10:43:06.119896] E [event-poll.c:256:event_unregister_poll] 
    0-XXXmanu: event_pool = 0xbb741080, event_pool->used = 5,    
    event_pool->reg[6] = 12 -> 3
[2013-04-25 10:43:06.136103] E [event-poll.c:348:event_dispatch_poll_handler]
    0-poll: index not found for fd=12 (idx_hint=5 event_pool=0xbb741080)

It seems odd to me that idx = 6 while event_pool->used = 5. Does it makes
sense?

-- 
Emmanuel Dreyfus
http://hcpnet.free.fr/pubz
manu at netbsd.org