[LU-2494] error: get_param: /proc/{fs,sys}/{lnet,lustre}/obdfilter/*/mds_sync: Found no match

• Type: Bug
• Resolution: Fixed
• Priority: Blocker
• Fix Version/s: Lustre 2.4.0
• Affects Version/s: Lustre 2.4.0
• Labels: HB
• Environment: Lustre Branch: master
• Severity: 3
• Assignee: Bob Glossman (Inactive)
• Reporter: Jian Yu

Description

The following error message occurred in many tests which call wait_mds_ost_sync:

CMD: fat-intel-3vm8 /usr/sbin/lctl get_param -n obdfilter.*.mds_sync
fat-intel-3vm8: error: get_param: /proc/{fs,sys}/{lnet,lustre}/obdfilter/*/mds_sync: Found no match

https://maloo.whamcloud.com/test_sets/b9a24768-44a8-11e2-8c8b-52540035b04c
https://maloo.whamcloud.com/test_sets/5419a010-44aa-11e2-8c8b-52540035b04c
https://maloo.whamcloud.com/test_sets/c923effc-44ad-11e2-8c8b-52540035b04c
https://maloo.whamcloud.com/test_sets/34f7f994-44ae-11e2-8c8b-52540035b04c
https://maloo.whamcloud.com/test_sets/d0434e5c-44af-11e2-8c8b-52540035b04c

Issue Links

• is related to: LU-2620 Failure on test suite replay-ost-single test_6: test_6 failed with 1

Comments

[Andreas Dilger]

Bob,
while it is good to fix the test-framework.sh to check for the new parameter, but this parameter does not exist on older servers. Since the test-framework.sh runs from the client (which may be newer or older than the server), it needs to be able to work with the old-style /proc parameters as well.

Could you please make another patch that adds a check for the old mds_sync file if old_sync_processed does not exist.

[Bob Glossman (Inactive)]

change framework to look at old_sync_processed
http://review.whamcloud.com/4885

[Alex Zhuravlev]

right, because OSS does not drive recovery any more. it's all done on MDS and from OSS point of view this is a regular request.
so, instead of looking for signs on OSS side, we should watch for all OSPs to get old_sync_processed=1

[Peter Jones]

Alex could you please comment on this? Thanks Peter

[Bob Glossman (Inactive)]

This /proc var seems to have been entirely done away with in the conversion to osd. In previous revs there was an entry in lprocfs_filter_obd_vars[]

        { "mds_sync",     lprocfs_filter_rd_mds_sync, 0, 0},

This read out the value of obd->u.filter.fo_mds_ost_sync

In current code it looks like obd->u.filter.fo_mds_ost_sync still exists, but nobody ever sets or clears it.

There is no "mds_sync" entry in the closest equivalent I can find to lprocfs_filter_obd_vars[], namely lprocfs_osd_obd_vars[].

Is there some other proc var now someplace else in /proc that fetches similar info from some new layer? The function seems to be lost from obdfilter entirely.

[Peter Jones]

Bob

Could you please look into this one?

Thanks

Peter