2024-01-31 12:21:41

by Charles Hedrick

[permalink] [raw]
Subject: effect of delegations

We have two production NFS servers, both running Ubuntu 22.04 with 5.15. We had been using NFS 4 where possible, with delegations disabled because of bugs that were around a couple of years ago. This year we decided to enable delegation, since there haven't been as many bugs recently.

CPU time in system state had been very low. Suddenly it went to 30%, more continuous on one server than the other. Performance degraded by an order of magnitude on that server. Turning off delegation went back to good performance.

We'll report this through Ubuntu, though I'm not sure whether diagnosis is going to be practical, since we can't really mess with these two servers. I'm hoping the report might be useful in any case as one data point.

(We've had to back off to NFS v3 on one of the servers, because of the typical NFSv4 permanent client hangs, i.e.hangs that survive rebooting the clients. These are almost certainly due to the very high level of lock activity on that server. Again, we'll see if the Ubuntu folks can help diagnose it.)