Lustre devel - Feb 2007 - [Bug 11659] ptllnd credit overflows on Portals congestion

Please don''t reply to lustre-devel. Instead, comment in Bugzilla by
using the following link:
https://bugzilla.lustre.org/show_bug.cgi?id=11659



Created an attachment (id=9526)
Please don''t reply to lustre-devel. Instead, comment in Bugzilla by
using the following link:
 --> (https://bugzilla.lustre.org/attachment.cgi?id=9526&action=view)
client side syslog/console

Please don''t reply to lustre-devel. Instead, comment in Bugzilla by
using the following link:
https://bugzilla.lustre.org/show_bug.cgi?id=11659



Created an attachment (id=9527)
Please don''t reply to lustre-devel. Instead, comment in Bugzilla by
using the following link:
 --> (https://bugzilla.lustre.org/attachment.cgi?id=9527&action=view)
server syslog

Please don''t reply to lustre-devel. Instead, comment in Bugzilla by
using the following link:
https://bugzilla.lustre.org/show_bug.cgi?id=11659

           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|bugs@clusterfs.com          |eeb@clusterfs.com
             Status|NEW                         |ASSIGNED


Created an attachment (id=9538)
Please don''t reply to lustre-devel. Instead, comment in Bugzilla by
using the following link:
 --> (https://bugzilla.lustre.org/attachment.cgi?id=9538&action=view)
patch to increase resilience to network problems

This patch tracks peer buffer utilisation more closely so that overruns are
handled properly and don''t fail the assertion in kptllnd_rx_done().  

It also includes stronger peer incarnation checks which may help debug
pathological behaviour if the underlying network misbehaves.

Nic, it would be great if you can try this with the sort of network meltdown
that produced the logs attached here.