Björn Rabenstein
Björn Rabenstein
> The validation is necessary to determine whether the name needs to be quoted, or can remain unquoted. This doesn't have to happen at scrape time, but could be done...
We discussed this again at the dev summit today. An important outcome was that we have confirmed that we do not aim for a "secure by default" setup, not only...
Some thoughts about mitigation: I think just reverting #16569 isn't a good mitigation because the tests added in that PR proves that the previous solution was also problematic, arguably more...
I propose to deactivate the failing tests for now, to not confuse arm64 users when doing unrelated dev work, see https://github.com/prometheus/prometheus/pull/16715 .
As a data point: `GOARCH=386 go test` on my `linux/amd64` system passes all the tests.
What is the failure message (i.e. what value does it come up with?).
To clarify: The test above is a new one, not one that already existed and usually succeeds. This new tests is prone to fail on all platforms. I guess arm64...
> IMHO -7.105427357601002e-15 is zero; the bug is in the tolerance. Right, but the calculation was hitting exactly zero with direct mean calculation. (Also, thanks to @crush-on-anechka, I have [just...
With #16773 merged, I think we can close this. Please follow-up here if new weirdnesses beyond the expected show up.
For posterity (in case somebody wonders if we can go back to pure incremental mean calculation after #16895): If I go back to f71daa797786b8a98a82c98f86f71bb709e5c609 (prior to re-introducing direct mean calculation,...