Making durations and number literals the same

[darshanime]

Design doc: https://docs.google.com/document/d/1LaZfknXuuRWGtQSbULoMtclQhuLUMrdwg15wMvoBvCQ/edit#

Signed-off-by: darshanime deathbullet@gmail.com

[roidelapluie]

Thank you. I expect that Beorn will want to look at this one, but you can expect some delays due to the summer period.

[LeviHarrison]

According to your design doc, the following should work: rate(prometheus_http_requests_total[0.25]), but instead I get this error: Error executing query: 1:37: parse error: strconv.Atoi: parsing "0.25": invalid syntax. Otherwise seems good!

[juliusv]

I tried predict_linear(foo[5h], 1h), but I get:

parse error: trailing commas not allowed in function call args

I also left some comments on the design doc on questions I had.

[juliusv]

One thing I'm interested in is how this will affect the printing of PromQL expressions once they have been parsed.

It would be sad if the input was:

predict_linear(foo[1h], 4h)

...and we could only serialize it out again as:

predict_linear(foo[1h], 14400)

Not sure if this is the plan already, but would it make sense to store the original tokens in the parsed AST somehow so that we can render them out again in a way that makes most sense to the user?

[beorn7]

One thing I'm interested in is how this will affect the printing of PromQL expressions once they have been parsed.
[…]

That's a good point, but perhaps we can solve that later, once the main functionality is implemented.

I think the comments above( #9138 (comment) , #9138 (comment) ) are real issues that need to be fixed.

[darshanime]

thank you, will address 👍
marking as draft to avoid notifications in the meantime...

[beorn7]

This might come from my ignorance of parser internals, but wouldn't it be easier to just get rid of the DURATION type altogether and replace it with NUMBER everywhere? And then parse each number as a duration.

[valyala]

FYI, VictoriaMetrics supports using durations instead of numeric values and vice versa in MetricsQL. See, for example, how does predict_linear(sum(rate(vm_http_requests_total[300])) offset 3600, 3h) work in VictoriaMetrics playground. It would be great if Prometheus could support this too.

[beorn7]

We discussed this during our bug scrub. I think my point above is still valid and hasn't been addressed. @darshanime are you still up to working on this?

This general topic comes up quite often, so we should finish this up one way or another.

[beorn7]

Rebased and adjusted the version (as this will land in 2.54, not 2.53).

[bboreham]

/prombench main

[prombot]

⏱️ Welcome to Prometheus Benchmarking Tool. ⏱️

Compared versions: PR-9138 and main

After successful deployment, the benchmarking results can be viewed at:

• Prometheus Meta
• Prombench Dashboard
• Grafana Explorer, Loki logs
• Parca profiles (e.g. in-use memory)

Other Commands:
To stop benchmark: /prombench cancel
To restart benchmark: /prombench restart main

[bboreham]

/prombench cancel

[prombot]

Benchmark cancel is in progress.

[bboreham]

Sorry, it's pointless comparing against main when the PR has been merged.

[prombot]

Benchmark tests are running for 3 days! If this is intended ignore this message otherwise you can cancel it by commenting: /prombench cancel

[bboreham]

/prombench cancel

[prombot]

Benchmark cancel is in progress.