chunkenc.Iterator.AtFloatHistogram may do a shallow copy if
it receives nil as input pointer. This can in turn share the
span slice with multiple histograms in the matrixSelectorHPool,
leading to unexpected errors.
Signed-off-by: György Krajcsovits <gyorgy.krajcsovits@grafana.com>
The basic idea here is that the previous code was always doing
incremental calculation of the mean value, which is more costly and
can be less precise. It protects against overflows, but in most cases,
an overflow doesn't happen anyway.
The other idea applied here is to expand on #14074, where Kahan
summation was applied to sum().
With this commit, the average is calculated in a conventional way
(adding everything up and divide in the end) as long as the sum isn't
overflowing float64. This is combined with Kahan summation so that the
avg aggregation, in most cases, is really equivalent to the sum
aggregation with a following division (which is the user's expectation
as avg is supposed to be syntactic sugar for sum with a following
divison).
If the sum hits ±Inf, the calculation reverts to incremental
calculation of the mean value. Kahan summation is also applied here,
although it cannot fully compensate for the numerical errors
introduced by the incremental mean calculation. (The tests added in
this commit would fail if incremental mean calculation was always
used.)
Signed-off-by: beorn7 <beorn@grafana.com>
The optimizer which detects cases where histogram buckets can be skipped
does not take into account binary expressions. This can lead to buckets
not being decoded if a metric is used with both histogram_fraction/quantile and
histogram_sum/count in the same expression.
Signed-off-by: Filip Petkovski <filip.petkovsky@gmail.com>
Implement histogram statistics decoder
This commit speeds up histogram_count and histogram_sum
functions on native histograms. The idea is to have separate decoders which can be
used by the engine to only read count/sum values from histogram objects. This should help
with reducing allocations when decoding histograms, as well as with speeding up aggregations
like sum since they will be done on floats and not on histogram objects.
Signed-off-by: Filip Petkovski <filip.petkovsky@gmail.com>
---------
Signed-off-by: Filip Petkovski <filip.petkovsky@gmail.com>
Co-authored-by: Anthony Mirabella <a9@aneurysm9.com>
This can give a more precise result, by keeping a separate running
compensation value to accumulate small errors.
See https://en.wikipedia.org/wiki/Kahan_summation_algorithm
Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
The function `rangeEvalTimestampFunctionOverVectorSelector` appeared to be checking histogram size, however the value it used was always 0 due to subtle variable shadowing.
However we don't need to pass sample values to the `timestamp` function, since the latter only cares about timestamps. This also affects peak sample count in statistics, since we are no longer copying histogram samples.
Signed-off-by: Arve Knudsen <arve.knudsen@gmail.com>
This saves memory in other kinds of aggregation.
We don't need `orderedResult` in `aggregationCountValues`; the ordering
is not guaranteed.
Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
They aggregate results in different ways.
topk/bottomk don't consider histograms so can simplify data collection.
Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
This is a cleaner split of responsibilities.
We now check the sample count after calling rangeEvalAgg.
Changed re-use of samples to use `Clone` and `defer`.
Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
Pass it as a float64 not as interface{}.
Make k a simple int, since that is the parameter to make().
Pull invalid quantile warning out of the loop.
Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
The new function `rangeEvalAgg` is mostly a copy of `rangeEval`, but
without `initSeries` which we don't need and inlining the callback to
`aggregation()`.
Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
The existing aggregation function is very long and covers very different
cases.
`aggregationCountValues` is just for `count_values`, which differs from
other aggregations in that it outputs as many series per group as there
are values in the input.
Remove the top-level switch on string parameter type; use the same `Op`
check there as elswehere.
Pull checking parameters out to caller, where it is only executed once.
Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
* add custom buckets to native histogram model
* simple copy for custom bounds
* return errors for unsupported add/sub operations
* add test cases for string and update appendhistogram in scrape to account for new schema
* check fields which are supposed to be unused but may affect results in equals
* allow appending custom buckets histograms regardless of max schema
Signed-off-by: Jeanette Tan <jeanette.tan@grafana.com>
Aggregations discard the metric name, so don't try to
include it in the error message.
Add a test that generates this warning.
Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
Fixes#11708.
If a range vector is fixen in time with the @ modifier, it gets still
moved around for different steps in a range query. Since no additional
points are retrieved from the TSDB, this leads to steadily emptying
the range, leading to the weird behavior described in isse #11708.
This only happens for functions listed in `AtModifierUnsafeFunctions`,
and the only of those that takes a range vector is `predict_linear`,
which is the reason why we see it only for this particular function.
Signed-off-by: beorn7 <beorn@grafana.com>
These functions act on the labels only, so don't need to go step by step
over the samples in a range query.
Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
* Reusing points slice from previous series when the slice is under utilized
* Adding comments on the bench test
Signed-off-by: Alan Protasio <alanprot@gmail.com>
This function is called very frequently when executing PromQL functions,
and we can do it much more efficiently inside Labels.
In the common case that `__name__` comes first in the labels, we simply
re-point to start at the next label, which is nearly free.
`DropMetricName` is now so cheap I removed the cache - benchmarks show
everything still goes faster.
Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
Optimize histogram iterators
Histogram iterators allocate new objects in the AtHistogram and
AtFloatHistogram methods, which makes calculating rates over long
ranges expensive.
In #13215 we allowed an existing object to be reused
when converting an integer histogram to a float histogram. This commit follows
the same idea and allows injecting an existing object in the AtHistogram and
AtFloatHistogram methods. When the injected value is nil, iterators allocate
new histograms, otherwise they populate and return the injected object.
The commit also adds a CopyTo method to Histogram and FloatHistogram which
is used in the BufferedIterator to overwrite items in the ring instead of making
new copies.
Note that a specialized HPoint pool is needed for all of this to work
(`matrixSelectorHPool`).
---------
Signed-off-by: Filip Petkovski <filip.petkovsky@gmail.com>
Co-authored-by: George Krajcsovits <krajorama@users.noreply.github.com>
Restore more efficient version of NewPossibleNonCounterInfo annotation
Signed-off-by: Jeanette Tan <jeanette.tan@grafana.com>
---------
Signed-off-by: Jeanette Tan <jeanette.tan@grafana.com>
In https://github.com/prometheus/prometheus/pull/13276 we started reusing float histogram objects to reduce allocations in PromQL.
That PR introduces a bug where histogram pointers gets copied to the beginning of the histograms slice,
but are still kept in the end of the slice. When a new histogram is read into the last element,
it can overwrite a previous element because the pointer is the same.
This commit fixes the issue by moving outdated points to the end of the slice
so that we don't end up with duplicate pointers in the same buffer. In other words,
the slice gets rotated so that old objects can get reused.
Signed-off-by: Filip Petkovski <filip.petkovsky@gmail.com>
This commit reduces the memory needed to query native histogram objects
by reusing existing HPoint instances.
Signed-off-by: Filip Petkovski <filip.petkovsky@gmail.com>
This function is useful to analyze promQL queries. We want to use this in Mimir to record the time range which the query touches.
I also chose to remove the `Engine` receiver because it was unnecessary, and it makes it easier to use, but happy to refactor that if you disagree.
The function is untested on its own. If you prefer to have unit tests now that its exported, I can look into adding some.
Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>
* Remove NewPossibleNonCounterInfo until it can be made more efficient, and avoid creating empty annotations as much as possible
Signed-off-by: Jeanette Tan <jeanette.tan@grafana.com>
Return annotations (warnings and infos) from PromQL queries
This generalizes the warnings we have already used before (but only for problems with remote read) as "annotations".
Annotations can be warnings or infos (the latter could be false positives). We do not treat them different in the API for now and return them all as "warnings". It would be easy to distinguish them and return infos separately, should that appear useful in the future.
The new annotations are then used to create a lot of warnings or infos during PromQL evaluations. Partially these are things we have wanted for a long time (e.g. inform the user that they have applied `rate` to a metric that doesn't look like a counter), but the new native histograms have created even more needs for those annotations (e.g. if a query tries to aggregate float numbers with histograms).
The annotations added here are not yet complete. A prominent example would be a warning about a range too short for a rate calculation. But such a warnings is more tricky to create with good fidelity and we will tackle it later.
Another TODO is to take annotations into account when evaluating recording rules.
---------
Signed-off-by: Jeanette Tan <jeanette.tan@grafana.com>
promql engine: check unique labels using existing map
ContainsSameLabelset constructs a map with the same hash key as the one used to compile the output of rangeEval, so we can use that one and save work.
Need to hold the timestamp so we can be sure we saw the same series in the same evaluation.
Charles Korn