Deduplication must respect profiles with identical timestamps within the same series

[kolesnikovae]

At deduplication we discard profiles if their timestamps and labels match:

pyroscope/pkg/querier/select_merge.go

Lines 254 to 258 in 8efd7c1

	if previousTs != profile.Timestamp || phlaremodel.CompareLabelPairs(previousLabels, profile.Labels) != 0 {
	previousTs = profile.Timestamp
	previousLabels = profile.Labels
	next.Keep()
	continue

Sometimes distinct profiles of the same series may have the same timestamps (millisecond precision), which results in partial reads. Potentially, this could be fixed by #2480 (#2443)

/cc @cyriltovena

[cyriltovena]

Presicion should actually be nanoseconds but you're correct, we also have a unique ID that we could use to ultimately decide if this is a real duplicate.

[cyriltovena]

We could also use the IngesterAddr if it's the same as the previous one it's not a dupe, duplicate will only exists across ingester not within.

[kolesnikovae]

I think we can have duplicate timestamps on the same ingester, unfortunately. It comes down to the case when multiple instances form a single series, e.g:

• host A sends {service_name="my-app"} at time X
• host B sends {service_name="my-app"} at time X
• host C sends {service_name="my-app"} at time X

And so on

[kolesnikovae]

#2489 makes this issue way less severe