-
Notifications
You must be signed in to change notification settings - Fork 678
weave_flows metric is fastdp-only, no way to see how many sleeve flows there are #3788
Comments
"flow" is an object in the OVS kernel module used by fastdp; there is no analogous structure for sleeve. But I think you are looking for some indication that peer-to-peer connections have stuck at Does |
Ah, thanks for clarifying. Yes, that's definitely what I'm looking for, and connection type on
That's kind of what I was wondering, but the scale seems off. From what you're implying For example, during one of the instants when we were running Weave 2.6.1, for one of the pods, And then, after the upgrade to 2.6.2, So I suppose that means that no, |
Ah, right: flows are per-MAC whereas connections are per-machine (or peer if you prefer). Would you be interested to try a PR to add the connection-type label? |
perhaps, though I'm not at all familiar with the weave code so it may take some time... I'll poke at it and see 😉 |
Great! Start here: Line 50 in 0c4659c
The status struct has a dictionary |
@bboreham thanks - I was wondering what |
(note: I'm not very familiar with Weave internals, so apologies if I make terminology mistakes below - please correct me!)
We recently ran into the issues fixed in 2.6.2 (#3781, #3783, #3782), and while working through upgrading our clusters were wondering how we could detect this sort of thing in the future.
We use Prometheus for our monitoring, and have the
weave_flows
metric graphed on a Grafana dashboard. It's obvious when looking over a period of time that the number of flows plummets (in our case, from an average of a few thousand per node to below ~100 per node).However, we found no way to measure the number of sleeve flows.
Note that this may be related to #2557, but that may be a big general, and it's been stale for a long time.
What you expected to happen?
A metric to measure the number of sleeve flows should be present, or perhaps a label to indicate which kind of flows are being measured.
What happened?
No indication of what
weave_flows
is actually indicating.How to reproduce it?
Grab the output of the
/metrics
endpoint from a weave process.Anything else we need to know?
Probably not relevant, but we're running on AWS with KOPS.
Versions:
Note: we were running Weave 2.6.1, but have now upgraded to 2.6.2. (can't exec into the weave-net pod, but we're running
weaveworks/weave-npc:2.6.2
).Logs:
(not relevant IMO)
Network:
(not relevant IMO)
The text was updated successfully, but these errors were encountered: