[Openais] latest openais tree problems

Mark Haverkamp markh at osdl.org
Wed Jan 4 13:53:40 PST 2006


Steve,

I updated my view and started testing.  I have seen problems in
determining the primary component.  I have my four node setup and
started aisexec on all four.  Then I started my publish/subscribe tests.
So far so good.  Next I killed aisexec on one node.  The other three
said that they weren't in the primary component and stopped.  I then
restarted the node that I killed.  All the nodes now say that they are
in the primary component.  But, even though the publish tests all
started sending once again, three of the four nodes's subscription tests
aren't seeing events delivered.  If I kill the subscription test and
restart it, though, it does see delivered events.  What happens now when
a node is non-primary?  Do the open connections from the exec to the
applications get closed, or just go dormant until the node is back in
the cluster?

This behavior seems to be related to the on-going event traffic.  If the
node isn't busy sending and receiving events when I kill one node the
other three seem to recover OK.

Mark.
-- 
Mark Haverkamp <markh at osdl.org>




More information about the Openais mailing list