cpu: Metric 'package_throttles_total' is per package.#657
cpu: Metric 'package_throttles_total' is per package.#657SuperQ merged 4 commits intoprometheus:masterfrom
Conversation
|
@rtreffer @SuperQ This PR has only limited testing (among other things because of Issue #644). Please review it. Also, please notice that I decided to use the label Feedback is welcome! BTW: IMHO the to for consistency. |
|
I think those missing fixtures were explicitly missing for some tests. |
|
@SuperQ Hm, I didn't catch that (but also didn't notice any test failures after adding them). Feel free to drop this hunk from the PR if you want. |
|
Adding the metrics to the e2e output masks the error. 😜 |
|
Any thoughts on my comment regarding the cpu label values? This is something that should be fixed before node-exporter 1.0 - if you agree that it's an inconsistency worth fixing. I can open a separate issue for this if you want. |
|
Yes, please open an issue to talk about the cpu labels. |
|
Oh, the |
|
For the |
|
I've opened issue #661 (cpu collector: Change cpu label values from "cpu0" to "0" ). |
de484f2 to
6ce8f40
Compare
|
I've removed the cpu2/cpu3 fixture changes and added two minor improvements as separate commits. I've also rebased against master. |
|
Can you add a cpu1 fixture to verify that the 2nd cpu in a node is ignored? |
'package_throttles_total' is per package, not per cpu. This also reduces the total number of cpu time series a lot (esp for multi core cpus).
Do not rely on the range index.
This file must be ignored by the cpu collector.
6ce8f40 to
0fc4035
Compare
|
@SuperQ I've added the node 0 cpu1 fixture and rebased against master. |
* cpu: Metric 'package_throttles_total' is per package. 'package_throttles_total' is per package, not per cpu. This also reduces the total number of cpu time series a lot (esp for multi core cpus). * cpu: Better handling of a cpulist edge-case. * cpu: Extract the package number from the directory name. Do not rely on the range index. * cpu: Add package_throttle_count for node0 cpu1 This file must be ignored by the cpu collector.
Signed-off-by: prombot <prometheus-team@googlegroups.com>
'package_throttles_total' is per package, not per cpu. This also reduces
the total number of cpu time series a lot (esp for multi core cpus).
Also add some missing fixtures for cpu2 and cpu3 for consistency.