Content Switching Module (CSM): Sessions stuck in closing state

[thorsten.steffen]

Hi guys,

I'm troubleshooting a CSM with 4.3(1).

We migrated a real server from windows to linux, since then lot's of sessions stuck in closing state until the idle timer expires.

I traced the traffic from the client to the csm and from the csm to the real server what I saw is the following behaviour at the end of a session:

windows real server: client sends FIN -> real server acks the FIN -> real server sends FIN -> client sends RST -> real server sends RST -> CSM deletes session from session table

The linux real server mostly behaves like the windows real server but sometimes behaves as follows:

client sends FIN -> real server sends  FIN, but does not ack the FIN sequence number of the client -> real server sends RST -> client sends RST -> CSM does not delete  session from session table, it stucks in closing state

Does anybody know how I can solve this problem of half-closed sessions other than decreasing the idle timeout?

Is it possible to configure CSM to close and delete sessions not only on FIN-ACK packets but also on RST packets?

Or to use a different fast idle timeout when sessions are half-closed?

Many thanks in advance,

Thorsten

[Fabrizio Pedracini]

Hi Thorsten,

I remember seeing the "CSM_FAST_FIN_TIMEOUT" having a wrong value set by the CSM in some cases. The value was set to 0 instead of 10 seconds or more. Normally this timeout tells after how many seconds we reset the connection where a FIN has been detected.

Might be worth to double check the "show mod csm X variable" to verify if that is the case.

hope this helps,

Fabrizio

[thorsten.steffen]

Hi Fabrizio,

CSM_FAST_FIN_TIMEOUT is already set to 10 seconds.

What I found out while testing is that CSM_FAST_FIN_TIMEOUT works with configurations where idle timeout is configured > 0.

But when idle timeout is configured to 0 (what meas no idle timeout) the csm seems not to use CSM_FAST_FIN_TIMEOUT: Half closed sessions will stuck in closing state.

Any idea why cms behaves this way?

Regards,

Thorsten

[Fabrizio Pedracini · Accepted solution]

Hi Thorsten,

this is as per design. If the idle timeout is set to 0 it means never timeout, so in these cases the CSM_FAST_FIN_TIMEOUT won't be used. Using a very high idle timeout (non 0 though) will cause the CSM to use the

CSM_FAST_FIN_TIMEOUT when applicable.

Fabrizio

[thorsten.steffen]

Hi Fabrizio,

is there any piece of documention in cco where I can find a statement concerning this behaviour?

Thorsten

[Fabrizio Pedracini]

Hi Thorsten,

I can't find any pointer on cco regarding this behavior.

Fabrizio