Add Ip and Tcp metric collection#11170
Conversation
|
The |
albertvaka
left a comment
There was a problem hiding this comment.
Other Linux-only metrics explicitly say "Linux-only" in the metadata. Should we do the same here? Having so many Linux-specific metrics we should probably start looking for a way to mention that that scales better, but 🤷
Also, are these available on all Linux systems (even on older distros)? If they are not, what does the check do in that case? At the very least it shouldn't blow up but ideally it should not report them at all.
a2359de to
39934e2
Compare
|
I've added the (Linux Only) to the new metric metadata. |
sarah-witt
left a comment
There was a problem hiding this comment.
Thanks, looks good! I just had two nits about metrics metadata
5d892fe to
bd38403
Compare
|
@bonnefoa @DrynnBavis I think we need the metadata update to also include this PR in addition to #11156 |
bd38403 to
7d95878
Compare
7d95878 to
fe036a1
Compare
fe036a1 to
d555324
Compare
What does this PR do?
Collect additional IP and TCP metrics from
/proc/net/{snmp,netstat}.Motivation
For IP, we are interested in errors (checksum issues, invalid headers) and detailed informations about fragmentation. Knowing when an instance is fragmenting sent packets or has to reassemble fragmented packet can be a good indicator of a MTU issue somewhere in the network path.
For TCP, the connection metrics (ActiveOpens, PassiveOpens, AttemptFails...) will provide better visibility to detect connection churn.
system.net.tcp4.openingonly gives a snapshot of connections in an open state and is expensive to compute on instance with a lot of socket to parse.Additional Notes
Review checklist (to be filled by reviewers)
changelog/andintegration/labels attached