[LU-5655] lhsmtool_posix (copy tool agent) does not provide facility to un register

• Status: Resolved
• Project: Lustre
• Affects Version/s: Lustre 2.5.0, Lustre 2.6.0, Lustre 2.7.0
• Type: Improvement
• Priority: Critical
• Reporter: Vinayak Hariharmath (Inactive)
• Assignee: Bruno Faccini (Inactive)
• Resolution: Cannot Reproduce
• Labels: HSM

Environment

centos 6.5

• Epic: client
• Project: HSM

Description

steps :
1. Mounted lustre
2. enabled hsm

    # lctl set_param mdt.lustre-MDT0000.hsm_control=enabled

3. Started copy tool daemon (only one copy tool agent can run on a client )

    #lhsmtool_posix --daemon --hsm-root /tmp/HSM --archive=1 /mnt/lustre/

4. Only way to stop agent is to send TERM signal to agent as per lustre manual. So I killed it (as I wanted to run modified copy tool agent)

    # ps -ef | grep lhs
       root      4017     1  0 16:54 ?        00:00:00 lhsmtool_posix --daemon --hsm-root /tmp/HSM --archive=1 /mnt/lustre/
       root      4045  2110  0 16:55 pts/1    00:00:00 grep lhs

5. Now I tried to start new copy tool agent

     #lhsmtool_posix --daemon --hsm-root /tmp/HSM --archive=1 /mnt/lustre/ 

But got below message from kernel

    Sep 11 16:55:34 localhost kernel: Lustre: HSM agent b150c068-22f2-83cd-21b0-2b4e76a3082a already registered.

Comments

[Vinayak Hariharmath (Inactive)]

I feel once the daemon is killed , it should get unregistered.

[Andreas Dilger]

Is this a new problem in 2.6.0 or does the same problem exist in 2.5?

[Robert Read (Inactive)]

I'm not able to reproduce this on 2.5.3. Whenever I kill the daemon, it stops running and the is no longer registered.

It's not clear in the description that the ps in step 4 was run before or after you killed the copytool. If it is after, then clearly the daemon is still running for some reason and so it is still registered.

[Vinayak Hariharmath (Inactive)]

ps in step 4 was run before I killed copy tool after that I killed it. Later I tried to start copy tool again but I got below message. Using b2_6 branch.

Oct 14 17:53:59 localhost kernel: Lustre: HSM agent 0f4d5578-4a43-df67-a8b1-ce31a2c2cd3a already registered

[Robert Read (Inactive)]

I'm seeing this on b2_6 as well, so this appears to be new in 2.6.0.

[Jodi Levi (Inactive)]

Bruno,
Can you please have a look at this one?
Thank you!

[Bruno Faccini (Inactive)]

Ok. I just wonder if copytool death and unregister could not take longer than before in b2_6 and then if this ticket could be related to LU-5622 ? But then, I also wonder why this does not currently affect auto-tests sanity-hsm results ?

[Bruno Faccini (Inactive)]

BTW I am unable to reproduce it with master/b2_6 builds.
Robert, can you help me and detail the exact versions ad configuration you used to be able to reproduce ?

[Robert Read (Inactive)]

I just tried again on a recent master build and wasn't able to reproduce it, either. I don't recall exact version of 2.6 I had been using, but the configuration would have been a single node setup using llmount.sh plus an additional mountpoint for the copytool.

[Bruno Faccini (Inactive)]

Vinayak, are you still able to reproduce the problem at your site?
If yes can you help and provide more infos, because I am unable to reproduce for the moment?
It would be nice if before running again your test/reproducer you can enable the full Lustre debug-log mask on both the CLient/Agent and MDS nodes, and then to provide them, can you do this ?

[Vinayak Hariharmath (Inactive)]

Sure. I will verify it on my side and update the bug. Sorry I was stuck with other work and could not able update it.

[Bruno Faccini (Inactive)]

Vinayak, Any update ? Did you spend some time for more testing about this issue ? Could it be, as I suspected above, that the copytool unregister could take longer ?

[Vinayak Hariharmath (Inactive)]

Hi

Yes I have spent bit time on it to verify.
The issue is not observed on master but still dmesg gives

Lustre: HSM agent 277f9613-7a7d-1caa-5849-0b5ffe11fdbb already registered

but there is no problem with copy tool which works fine (Earlier with the above error message, copy tool was failing to start after killing it). I guess it only a grammar correction. Dmesg logs attached.

[Bruno Faccini (Inactive)]

Hello Vinayak,
Do you agree if close this ticket as "Cannot reproduce" ?

[Vinayak Hariharmath (Inactive)]

Hi Bruno,

We can close this as "Cannot reproduce", but there is one more ticket LU-5216 looks bit similar to this issue. I am trying to reproduce it and draw some relation with this.

Thanks

[Bruno Faccini (Inactive)]

According to the description in LU-5216, then we may easily think that if both tickets are related, this one would only be a consequence of LU-5216 due to some orphan copytool thread still present or related data structure deferred/missing cleanup.
So closing this one.