pcp
pcp copied to clipboard
performancecopilot
Performance Co-Pilot
As specified in readme, agent accepts messages which specify labels using hashtag like so `metric:5|c|#tagX:X,tagW:W` but it throws away messages which do include # but without any tags specified eg....
https://lwn.net/Articles/875945/ ClusterFuzzLite [1] looks promising and supports GitHub Actions [2]. [1] https://google.github.io/clusterfuzzlite/ [2] https://google.github.io/clusterfuzzlite/running-clusterfuzzlite/github-actions/
The `kernel.cpu.util.sys` derived metric is defined as: ``` kernel.cpu.util.sys = (100 * rate(kernel.all.cpu.sys)) / hinv.ncpu ``` It would be great if the same expression is also supported with pmseries: ```...
without `rate()`: ``` $ curl -s "http://localhost:44322/series/values?start=$(( $(date +%s)-60 ))&finish=$(date +%s)&interval=15&series=$(pmseries "kernel.all.sysfork")" | jq [ { "series": "0da9ac73a95ada435faa8e1d1afdd73e0370d2d9", "timestamp": 1636730856274.284, "value": "1750473" }, { "series": "0da9ac73a95ada435faa8e1d1afdd73e0370d2d9", "timestamp": 1636730866301.775, "value": "1750499"...
https://packages.debian.org/sid/pcp we have a hard dependency on bpftrace, but a weak dependency (suggests) on python3-bpfcc. imho bpftrace shouldn't be a hard dependency, as probably many users don't need the bpftrace...
after pmlogger_check launches a pmlogger it uses pmlc to wait for it to get started : ``` 554 if echo "connect $1" | pmlc 2>&1 | grep "Unable to connect"...
Review pmlogconf config files to (a) avoid duplicate metrics being logged, and (b) reference leaf pmns names rather than non-leaves where that is feasible. This may help reduce pmlogger config...
