Julien

10596 seems the user id, not the process ID, so 63 is not the number of files of the processes. Can you provide the accurate number? I think your issue...

(A+B (mod M)) = (A (mod M) + B (mod M)) This is not mathematically correct: 7%2+3%2 != 10%2

Maybe we could better comment the existing calculation. The current form is quite confusing to me as we change the value of offset.

I don't know of any, but changing the formula is a risk of introducing new bugs.

cc @rfratto

If this is available to the API calls we are using now, we could expose it. Is it the case? We should in general try to avoid extra calls.

I would expect that in this case Prometheus would indeed skip 6 hours of data. One the time on the server is moved by multiple hours, our TSDB behaviour can...

It is more complicated, the tsdb must ingest samples in the correct order. If you have ingested data at 6pm, you can't ingest the same samples at 1pm.

> I think it has to do something with the scraping logic. I see `alignedScrapeTime` which remains uneffected by time change and always goes up. @roidelapluie since you worked on...

This is probably a duplicate of #8243 fixed by #8601. @liuliqiang can you update to 2.25.2 or 2.26.0 to test the fix? Thanks!