[release 2.54] [BUGFIX] Scraping: allow multiple samples on same series

[bboreham]

So long as they specify timestamps. We don't check that the timestamps are different.

Fixes #14503

Extend test, and use client_golang/prometheus/testutil to simplify metric check.

[machine424]

lgtm!

[machine424]

If we could add a test stating that we assume with this fix test_metric 1\ntest_metric 1 1001 not to get flagged as well, it'd be great.

[machine424]

Actually, if they arrive in this order test_metric 1 1001\ntest_metric 1, they'll still be flagged.
I'm not sure if this is a valid use case though.

[bboreham]

This came up on the issue - I don’t think we should make guarantees about mixing timestamp and no-timestamp.

At low probability, the specified timestamp would end up equal to the unspecified one.

[Masmiiadm]

Hello , will this change be merged to 2.53.0 LTS version

[bboreham]

No we will not change 2.53.0. There should be a patch release with #14740.

[Masmiiadm]

@bboreham
Following the bug fix [[release 2.54] [BUGFIX] , I updated Prometheus from 2.53.0 to version 2.54.1 (kube-prometheus-stack Helm chart v63.0.0), but the issue of 'Error on Ingesting Samples with Different Values but Same Timestamp' still persists.
Is there a flag to enable or any configuration to add in order to allow multiple samples on the same series?
ts=2024-10-11T13:29:45.818Z caller=scrape.go:1754 level=warn component="scrape manager" scrape_pool=consul-appli target=http://x.x.x.x:8080/metrics msg="Error on ingesting samples with different value but same timestamp" num_dropped=1320
ts=2024-10-11T13:29:48.830Z caller=scrape.go:1744 level=warn component="scrape manager" scrape_pool=serviceMonitor/monitoring/monitoring-k8s-kube-state-metrics/0 target=http://10.139.28.168:8080/metrics msg="Error on ingesting samples with different value but same timestamp" num_dropped=25

I downgraded to v2.47 and there is no scrapping errors .

[bboreham]

Please open an issue and include the metrics being scraped.