[Gluster-devel] infinband client io hangs when one brick fails

[Mickey Mazarick]

Sorry to be annoying today, but while I'm reporting bugs, I've noticed 
that when you use ib-verbs a client will hang forever when an afr/unify 
volume goes down.

The logs show:
2008-02-29 06:00:56 W [client-protocol.c:4520:client_protocol_cleanup] 
main4: cleaning up state in transport object 0x51a6e0
2008-02-29 06:00:56 E [client-protocol.c:4572:client_protocol_cleanup] 
main4: forced unwinding frame type(0) op(34) reply=@0x530c40
2008-02-29 06:00:56 C [ib-verbs.c:1458:ib_verbs_disconnect] 
transport/ib-verbs: main4: peer disconnected, cleaning up
2008-02-29 06:00:56 W [client-protocol.c:294:client_protocol_xfer] 
main4: attempting to pipeline request type(0) op(34) with handshake
2008-02-29 06:01:01 W [client-protocol.c:294:client_protocol_xfer] 
main4: attempting to pipeline request type(0) op(34) with handshake
2008-02-29 06:02:01 W [client-protocol.c:294:client_protocol_xfer] 
main4: attempting to pipeline request type(0) op(34) with handshake
....


Any io just waits until that volume is back online. I can stop any brick 
with the same result.
This works great with tcp, ibverbs seems to be hanging only.

Thanks again!

-Mickey Mazarick