See the detailed analysis https://docs.google.com/document/d/1efVAMcEw7-R_KatHHcobcFBlNsre-DoThVHI8AO2SDQ/edit?tab=t.0
I ran extensive benchmarks using synthetic data as well as real WAL segments pulled from the prombench runs.
All benchmarks are here https://github.com/prometheus/prometheus/compare/bwplotka/wal-reuse?expand=1
* optimization(tsdb/wlog): reuse Ref* buffers across WAL watchers' reads
Signed-off-by: bwplotka <bwplotka@gmail.com>
* optimization(tsdb/wlog): avoid expensive error wraps
Signed-off-by: bwplotka <bwplotka@gmail.com>
* optimization(tsdb/wlog): reuse array for filtering
Signed-off-by: bwplotka <bwplotka@gmail.com>
* fmt
Signed-off-by: bwplotka <bwplotka@gmail.com>
* lint fix
Signed-off-by: bwplotka <bwplotka@gmail.com>
* tsdb/record: add test for clear() on histograms
Signed-off-by: bwplotka <bwplotka@gmail.com>
* updated WriteTo with what's currently expected
Signed-off-by: bwplotka <bwplotka@gmail.com>
---------
Signed-off-by: bwplotka <bwplotka@gmail.com>
* feat(tsdb): allow appending to ST capable XOR chunk optionally
Only for float samples as of now. Supports for in-order and out-of-order
samples.
Make sure that on readout the ST capable chunks are returned automatically.
When the chunks are returned as is, this is trivially true.
When a chunk needs to be re-coded due to deletion (tombstone) markers,
we take the encoding of the original chunk.
When a chunk needs to be created from overlapping chunks, we observe
whether ST is zero or not and create the new chunk based on that.
Signed-off-by: György Krajcsovits <gyorgy.krajcsovits@grafana.com>
Initial implementation of https://github.com/prometheus/prometheus/issues/17790.
Only implements ST-per-sample for Counters. Tests and benchmarks updated.
Note: This increases the size of the RefSample object for all users, whether st-per-sample is turned on or not.
Signed-off-by: Owen Williams <owen.williams@grafana.com>
ReduceResolution is currently called before validation during
ingestion. This will cause a panic if there are not enough buckets in
the histogram. If there are too many buckets, the spurious buckets are
ignored, and therefore the error in the input histogram is masked.
Furthermore, invalid negative offsets might cause problems, too.
Therefore, we need to do some minimal validation in reduceResolution.
Fortunately, it is easy and shouldn't slow things down. Sadly, it
requires to return errors, which triggers a bunch of code changes.
Even here is a bright side, we can get rud of a few panics. (Remember:
Don't panic!)
In different news, we haven't done a full validation of histograms
read via remote-read. This is not so much a security concern (as you
can throw off Prometheus easily by feeding it bogus data via
remote-read) but more that remote-read sources might be makeshift and
could accidentally create invalid histograms. We really don't want to
panic in that case. So this commit does not only add a check of the
spans and buckets as needed for resolution reduction but also a full
validation during remote-read.
Signed-off-by: beorn7 <beorn@grafana.com>
* fix: Fix slicelabels corruption when used with proto decoding
Alternative to https://github.com/prometheus/prometheus/pull/16957/
Signed-off-by: bwplotka <bwplotka@gmail.com>
* addressed comments
Signed-off-by: bwplotka <bwplotka@gmail.com>
---------
Signed-off-by: bwplotka <bwplotka@gmail.com>
Reduce the resolution of histograms as needed and ignore invalid
schemas while emitting a warning log.
Signed-off-by: György Krajcsovits <gyorgy.krajcsovits@grafana.com>
See
https://pkg.go.dev/golang.org/x/tools/gopls/internal/analysis/modernize
for details.
This ran into a few issues (arguably bugs in the modernize tool),
which I will fix in the next commit, so that we have transparency what
was done automatically.
Beyond those hiccups, I believe all the changes applied are
legitimate. Even where there might be no tangible direct gain, I would
argue it's still better to use the "modern" way to avoid micro
discussions in tiny style PRs later.
Signed-off-by: beorn7 <beorn@grafana.com>
They are used in multiple repos, so common is a better place for them.
Several packages now don't depend on `model/textparse`, e.g.
`storage/remote`.
Also remove `metadata` struct from `api.go`, since it was identical to
a struct in the `metadata` package.
Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
The 'ToFloat' method on integer histograms currently allocates new memory
each time it is called.
This commit adds an optional *FloatHistogram parameter that can be used
to reuse span and bucket slices. It is up to the caller to make sure the
input float histogram is not used anymore after the call.
Signed-off-by: Filip Petkovski <filip.petkovsky@gmail.com>
* Labels: reduce allocations when creating from TSDB
When reading the WAL, by passing references into the buffer we can avoid
copying strings under `-tags stringlabels`.
Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
When reading the WAL this method is called with buffers from a pool, on
multiple goroutines. Pre-allocating sufficient size avoids slow growth
and many reallocations in `append`.
Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
