tcp segment metrics are misleading #123
Comments
brayniac
added a commit
to brayniac/rezolus
that referenced
this issue
Nov 30, 2023
Fixes iopsystems#123 by using /proc/net/snmp to report TCP in/out segments and renaming the eBPF metrics so we can continue to report logical sends/recvs before fragmentation and after reassembly respectively.
Merged
brayniac
added a commit
that referenced
this issue
Nov 30, 2023
Fixes #123 by using /proc/net/snmp to report TCP in/out segments and renaming the eBPF metrics so we can continue to report logical sends/recvs before fragmentation and after reassembly respectively.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
The tcp/transmit/segments and tcp/receive/segments metrics are misleading because we have instrumented closer to the socket level.
This means that for workloads with large payloads we can undercount because we count before fragmentation and count after reassembly.
We should rename these metrics and clarify in the metric descriptions that these are logical sends/recvs and not segment counts. We should augment the existing metrics with rx/tx metrics along the softirq path.
The text was updated successfully, but these errors were encountered: