#4717 cluster upgrade brings high load (60+)
Jamie, I have to report unsusual behaviour on my herd of SLES 11 and 12 hosts.
Yesterday I did the upgrade to 1.790. As usual, I updated my "Cluster Manager" VM first, and thought that it became slower than usual. The I went to Cluster -> Cluster Webmin Servers and wanted to update "all hosts that don't have it", as I usually do to deploy the update. I stopped the action when the host showed a load of 130+ (normally it is below 1).
After today's update to 1.791 the first page (with the system information) is faster, but the update of all cluster hosts still caused a load of 60+ and most host were displayed as not updated for unknown reason. The swap usage ist up to 70% though (being <10% before the "Cluster Update" started.
Now I'm updating my 20+ hosts by groups of 5 or 5 hosts, an get a load of 10-15 "only". In older versions, the load did not exceed 5 when I performed "Cluster" actions.
Regards, Werner
There was a bug in Webmin 1.790 that almost certainly caused this. Did you still see slowness even after the master system was upgraded to 1.791?
Yes, Jamie, I did, but the load did not go as high as before, only up to ~60 for the update of 20 cluster hosts, not ~130 as before. But most cluster hosts were not updated for "unknown reasons" (maybe timeouts). And I still noticed enormous swapping during that action (~6 GB extra). Updating the cluster hosts in groups (of 3-5 members each) works with a load of ~10 and "only" 2 G additional swap use.
Ok, I think I see the cause of this - a change in Webmin 1.790 is causing an inefficient code path to be used when uploading files from a master system to slaves. I will fix this in the next release.