Be more specific when identifying a sparse histogram

It's a prefectly valid use case to have a sparse histogram with a zero
threshold of zero (i.e. only observations of exactly zero go into the
zero bucket). Even if the current PoC implementation of client_golang
doesn't allow that, such a case should be ingested properly.

However, there is now the edge case af a sparse histogram with a zero
threshold of zero and no observations yet. Such a histogram would look
the same if it was meant to be a conventional histogram. For now, we
ingest this case as a conventional histogram, but the final format
should have means to unambiguously express if a histogram is meant to
be ingested as a sparse histogram or as a conventional histogram.

Signed-off-by: beorn7 <beorn@grafana.com>
This commit is contained in:
beorn7 2021-07-19 19:58:04 +02:00
parent 4fefd7520e
commit c35f138a9a
1 changed files with 17 additions and 1 deletions

View File

@ -309,7 +309,7 @@ func (p *ProtobufParser) Next() (Entry, error) {
p.state = EntryType
case EntryType:
if p.mf.GetType() == dto.MetricType_HISTOGRAM &&
p.mf.GetMetric()[0].GetHistogram().GetSbZeroThreshold() != 0 {
isSparseHistogram(p.mf.GetMetric()[0].GetHistogram()) {
p.state = EntryHistogram
} else {
p.state = EntrySeries
@ -459,3 +459,19 @@ func formatOpenMetricsFloat(f float64) string {
}
return s + ".0"
}
// isSparseHistogram returns false iff the provided histograms has no
// SparseBuckets and a zero threshold of 0 and a zero count of 0. In principle,
// this could still be meant to be a sparse histgram (with a zero threshold of 0
// and no observations yet), but for now, we'll treat this case as a conventional
// histogram.
//
// TODO(beorn7): In the final format, there should be an unambiguous way of
// deciding if a histogram should be ingested as a conventional one or a sparse
// one.
func isSparseHistogram(h *dto.Histogram) bool {
return len(h.GetSbNegative().GetDelta()) > 0 ||
len(h.GetSbPositive().GetDelta()) > 0 ||
h.GetSbZeroCount() > 0 ||
h.GetSbZeroThreshold() > 0
}