[LU-3859] grant shrinker floods OST and produce a large load Created: 29/Aug/13  Updated: 11/Dec/18  Resolved: 11/Dec/18

Status: Resolved
Project: Lustre
Component/s: None
Affects Version/s: None
Fix Version/s: None

Type: Bug Priority: Critical
Reporter: Alexey Lyashkov Assignee: WC Triage
Resolution: Cannot Reproduce Votes: 0
Labels: None
Environment:

xyratex lustre (mostly 2.1 based)


Issue Links:
Related
is related to LU-6951 sanity test_27m: @@@@@@ FAIL: OST0 wa... Closed
is related to LU-8708 Grant shrinking disabled all the time Resolved
is related to LU-7015 Grant space and reserved blocks perce... Resolved
Severity: 3
Rank (Obsolete): 10011

 Description   

While invested a sort of LNet bugs i tried to dump a rpc cache on OST after fail and find most of rpc in cache is grant shrink requests - it's produce a large load to the routers and OST's.

ffff8807b9cc3440 rpc_cache               1024        313       896    224     4k
# grep -c RQF_OST_SET_GRANT_INFO \&\ log 
304

so 304 from 313 RPC's is flood from grant shrinker.



 Comments   
Comment by Andreas Dilger [ 30/Aug/13 ]

We don't actually have OBD_CONNECT_GRANT_SHRINK enabled on the clients in our version of Lustre.

I agree that it shouldn't be flooding the server with these requests - grant shrinking should only happen occasionally, like on a ping. The default interval GRANT_SHRINK_INTERVAL is 1200s, but this would be sent from multiple clients if they are idle and not using the grant. It should only be a one-time event.

Comment by Andreas Dilger [ 11/Dec/18 ]

No activity here in 5 years and this is likely fixed with LU-8708.

Generated at Sat Feb 10 01:37:32 UTC 2024 using Jira 9.4.14#940014-sha1:734e6822bbf0d45eff9af51f82432957f73aa32c.