[ES-1292925] Fix metrics with reusable counter resets #107
Conversation
Signed-off-by: Yi Jin <[email protected]>
| var it adjustableSeriesIterator | ||
| if m.isCounter { | ||
| it = &counterErrAdjustSeriesIterator{Iterator: r.Iterator(nil)} | ||
| } else { | ||
| it = &noopAdjustableSeriesIterator{Iterator: r.Iterator(nil)} | ||
| } |
There was a problem hiding this comment.
You are applying counter adjustments to the raw data samples before deduplication. This is more complicated than applying the adjustments to the single time series after deduplication. The adjustments will intervene with quorum-based deduplication logic. I'm concerned it may introduce other edge cases.
There was a problem hiding this comment.
Can you add another test case such that one raw time series has a large gap with a reset, but the other two have complete data?
replica 0: [[1000, 10], [10000, 8], [11000, 10]]
replica 1: [[1000, 10], [2000, 0], [3000, 1], [4000, 2], [5000, 3], [6000, 4], [7000, 5], [8000, 6], [9000, 7], [10000, 8], [11000, 10]
replica 2: [[1000, 10], [2000, 0], [3000, 1], [4000, 2], [5000, 3], [6000, 4], [7000, 5], [8000, 6], [9000, 7], [10000, 8], [11000, 10]
There was a problem hiding this comment.
added new test cases, the reason i've done it this way is to reuse counterErrAdjustSeriesIterator which you need to call adjustAtValue() somewhere, passing a merged time series to original newDedupSeries doesn't work
| // feed the merged series into dedup series which apply counter adjustment | ||
| return NewMergedSeries(s.lset, repl, s.f) | ||
| } | ||
| if s.deduplicationFunc == AlgorithmChain { |
There was a problem hiding this comment.
fyi @yuchen-db, I've also added the prometheus implementation, but it doesn't work for a number of unit tests if you wonder
jnyi
changed the title
jnyi
force-pushed
the
time-series-gaps
branch
from
e012181 to
33ec6e3
Compare
jnyi
force-pushed
the
time-series-gaps
branch
from
33ec6e3 to
ccb205c
Compare
Signed-off-by: Yi Jin <[email protected]>
jnyi
force-pushed
the
time-series-gaps
branch
3 times, most recently
from
0b0fcfe to
7c98625
Compare
pkg/query/querier.go
Outdated
| partialResponseStrategy := storepb.PartialResponseStrategy_ABORT | ||
| if opts.GroupReplicaPartialResponseStrategy { | ||
| level.Debug(logger).Log("msg", "Enabled group-replica partial response strategy in newQuerierInternal") | ||
| level.Info(logger).Log("msg", "Enabled group-replica partial response strategy in newQuerierInternal") |
There was a problem hiding this comment.
This will be very chatty. I intentionally changed it to Debug previously.
There was a problem hiding this comment.
Every query will print this log.
There was a problem hiding this comment.
found i actually doesn't need this log line, it is logged here already:
level.Info(logger).Log("msg", "databricks querier features", "opts", fmt.Sprintf("%+v", opts))
pkg/query/querier.go
Outdated
| } else if partialResponse { | ||
| partialResponseStrategy = storepb.PartialResponseStrategy_WARN | ||
| } | ||
| level.Info(logger).Log("msg", "Deduplication algorithm applied", "func", opts.DeduplicationFunc) |
There was a problem hiding this comment.
Every query will print this log.
pkg/query/querier.go
Outdated
| partialResponseStrategy := storepb.PartialResponseStrategy_ABORT | ||
| if opts.GroupReplicaPartialResponseStrategy { | ||
| level.Debug(logger).Log("msg", "Enabled group-replica partial response strategy in newQuerierInternal") | ||
| level.Info(logger).Log("msg", "Enabled group-replica partial response strategy in newQuerierInternal") |
There was a problem hiding this comment.
Every query will print this log.
Signed-off-by: Yi Jin <[email protected]>
jnyi
force-pushed
the
time-series-gaps
branch
from
7c98625 to
ab31d53
Compare
This PR tries to fix when a reusable counter metrics resets:
Added a number of unit tested and make sure they pass when doing counter functions.
Also unit tested if not doing counter functions the original time series is returned
Pending integration tests, sent for early review and feedbacks
Integration tests work as expected:
Changes
Verification