[LU-178] server can do replays after recovery

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.1.0
• Affects Version/s: Lustre 2.1.0
• Severity: 3
• Assignee: Niu Yawei (Inactive)
• Reporter: Mikhail Pershin

Description

Niu has found in LU-128 that there is race possible in target_handle_connect(). The server sets RECOVERING flag in reply but right after that the recovery is ended, so client is evicted and new connection is established. The client will start replaying over established connection and server accepts them fully.

We have two problems here:
1) The race itself to be fixed so client shouldn't get RECOVERING flag for new connection.
2) The server must deny replay requests during normal processing otherwise it is possible to break transaction flow by malformed client which will send replays without recovery.

I am going to add patches in gerrit for both problems.

Comments

[Niu Yawei (Inactive)]

There is no intention to backport the fix to b1_8. This can be closed.

[Niu Yawei (Inactive)]

Yes, I think so, at least current b1_8 has the same problem.

[Peter Jones]

Niu,

Do I understand correctly that this is a long-standing issue with recovery on 1.8.x?

Peter

[Niu Yawei (Inactive)]

Looks b1_8 has the same problem, we should port it to b1)8 as well.

[Peter Jones]

Is any further work required or can this ticket be marked as resolved?

[Build Master (Inactive)]

Integrated in lustre-master » i686,server,el5,ofa #110
LU-178 prevent replays after recovery on server

Oleg Drokin : 5d4ae6c905b8e8a38a6e4d3195f550a1e2b37ba8
Files :

• lustre/ptlrpc/client.c
• lustre/include/obd_support.h
• lustre/mdt/mdt_handler.c
• lustre/ldlm/ldlm_lib.c
• lustre/tests/replay-single.sh
• lustre/ptlrpc/service.c

[Build Master (Inactive)]

Integrated in lustre-master » i686,server,el5,inkernel #110
LU-178 prevent replays after recovery on server

Oleg Drokin : 5d4ae6c905b8e8a38a6e4d3195f550a1e2b37ba8
Files :

• lustre/tests/replay-single.sh
• lustre/ptlrpc/client.c
• lustre/mdt/mdt_handler.c
• lustre/ldlm/ldlm_lib.c
• lustre/include/obd_support.h
• lustre/ptlrpc/service.c

[Build Master (Inactive)]

Integrated in lustre-master » i686,server,el6,inkernel #110
LU-178 prevent replays after recovery on server

Oleg Drokin : 5d4ae6c905b8e8a38a6e4d3195f550a1e2b37ba8
Files :

• lustre/include/obd_support.h
• lustre/ptlrpc/service.c
• lustre/mdt/mdt_handler.c
• lustre/ldlm/ldlm_lib.c
• lustre/tests/replay-single.sh
• lustre/ptlrpc/client.c

[Build Master (Inactive)]

Integrated in lustre-master » i686,client,el5,ofa #110
LU-178 prevent replays after recovery on server

Oleg Drokin : 5d4ae6c905b8e8a38a6e4d3195f550a1e2b37ba8
Files :

• lustre/ldlm/ldlm_lib.c
• lustre/mdt/mdt_handler.c
• lustre/include/obd_support.h
• lustre/ptlrpc/service.c
• lustre/tests/replay-single.sh
• lustre/ptlrpc/client.c

[Build Master (Inactive)]

Integrated in lustre-master » x86_64,server,el6,inkernel #110
LU-178 prevent replays after recovery on server

Oleg Drokin : 5d4ae6c905b8e8a38a6e4d3195f550a1e2b37ba8
Files :

• lustre/include/obd_support.h
• lustre/ldlm/ldlm_lib.c
• lustre/ptlrpc/client.c
• lustre/ptlrpc/service.c
• lustre/mdt/mdt_handler.c
• lustre/tests/replay-single.sh