can you try to use the latest beta version of pmm-server?
it has updated grafana to the latest version -
I analyzed HAR files and it looks very strange.
failed response size is exactly 200000 bytes…
successful response size is much bigger - 6376716
looks like some firewall or proxy cut packages if they bigger that 6.5 or 7 megabytes.
Can you try to fetch exactly the same url (which you dumped to HAR file) locally on PMM Server and remotely from you computer?
if result be the same (200000 bytes), it is definitely not network issue.
I have done more than that. I have been able to temporary deploy firefox on the Docker host and the Memory Graph fails there as well, same as on any other browser from any other machine.
Unless there is something weird on the container network stack, I would exclude a truncation of the TCP stream…
is “Load Average” graph is also affected?
is “Memory Usage” still broken if open it alone in “fullscreen” mode (left click on title, “View” button) ?
can you dump failed request one more time please?
is this issue connected to exact host or to any random 68 hosts?
can you choose first 67 hosts in the list without issues?
can you choose last 67 hosts in the list without issues?
can you choose 67 hosts in the middle of the list without issues?
Hi Mykola,
actually I made a mistake while counting, the critical threshold is 68, sorry for that. Basically, the memory graph works correctly with 68 hosts selected (top, bottom, middle or punched card style doesn’t matter), fails with 69 onwards.
I have tested version 1.1.1 without any luck, too.
We have 58 MySQL instances running on 16 MySQL VM, plus the PMM server is monitored, of course.
I have tried to access data directly in Prometheus and I see all the metrics for all nodes without any trouble, hence the issue seems related to Grafana.
Playing around selecting and deselecting hosts from the list, the problem appears “around” 68 hosts. Depending on the set, sometimes 67 is enough to trigger the problem, other times you need 69. So, it does not seem to be related to the number of hosts rather than the data volume to be accessed.