Avoid a tree of merge objects, which can result in
what I suspect is n^2 calls to Seek when using Without.
With 100k metrics, and a regex of ^$ in BenchmarkHeadPostingForMatchers:
Before:
BenchmarkHeadPostingForMatchers-8 1 51633185216 ns/op 29745528 B/op 200357 allocs/op
After:
BenchmarkHeadPostingForMatchers-8 10 108924996 ns/op 25715025 B/op 101748 allocs/op
Signed-off-by: Brian Brazil <brian.brazil@robustperception.io>
With 1M series:
Before:
BenchmarkHeadPostingForMatchers-8 1 3501996117 ns/op 61311520 B/op 78 allocs/op
After:
BenchmarkHeadPostingForMatchers-8 1 1403072952 ns/op 69261568 B/op 72 allocs/op
This works out as 3X faster, as the above time includes other things.
Signed-off-by: Brian Brazil <brian.brazil@robustperception.io>
It is easy to forget to close the block returned by createPopulatedBlock
which causes failures for windows so instead it returns the block dir
and which can be used by OpenBlock explicitly.
Signed-off-by: Krasi Georgiev <kgeorgie@redhat.com>
The WALFlushInterval is not used anywhere in the code base.
The WAL is not an interface anymore to save some lookup time so can't use NopWAL in the tests. Instead can just pass nil as the code checks for that and it is essentially a noop.
Signed-off-by: Krasi Georgiev <kgeorgie@redhat.com>
This reports the cardinality of each label,
the total number of label pairs,
and how much series worth of time is "uncovered"
by series data. Which is basically how much churn there is.
Signed-off-by: Brian Brazil <brian.brazil@robustperception.io>
closes https://github.com/prometheus/tsdb/issues/471
after implementing the new WAL this metric was missing so adding it again.
Also added it in a test to make sure it works as expected.
Signed-off-by: Krasi Georgiev <kgeorgie@redhat.com>
* refactor NewSegmentsRangeReader to take multi WAL ranges
In case of an error when checkpointing the WAL the error doesn't show
the exact WAL index that is corrupter. this is because it uses
MultiReader to read multiply WAL files.
This refactoring allows the NewSegmentsRangeReader to take more than a
single WAL range and it reads all of the ranges by iterating each one.
this changes the logs from
create checkpoint: read segments: corruption after 4841144384 bytes:...
to
create checkpoint: read segments: corruption in segment
data/wal/00017351 at 123142208: ...
Signed-off-by: Krasi Georgiev <kgeorgie@redhat.com>
* repair wal when the record cannot be decoded
Currently repair is run only when the error happens in the reader.
A corruption can occur after the record is read and when it is decoded.
This change wraps the error at decoding as a CorruptionErr as this error
is expected to trigger a repair.
Signed-off-by: Krasi Georgiev <kgeorgie@redhat.com>
* return an error when the last wal segment record is torn.
this ensures that a repair will be run when the last record in a segment
is torn.
Signed-off-by: Krasi Georgiev <kgeorgie@redhat.com>
* *: support Go modules
Signed-off-by: Simon Pasquier <spasquie@redhat.com>
* Update go.mod and Makefile.common
Signed-off-by: Simon Pasquier <spasquie@redhat.com>
This causes the head to be GCed at startup,
removing any series that were read from the WAL
but have since been written to a block. In
systems with low ingestion rates, this potentially
could be many many hours of data.
Signed-off-by: Brian Brazil <brian.brazil@robustperception.io>
This saves memory, about a quarter of the size of the postings map
itself with high-cardinality labels (not including the post ids).
Signed-off-by: Brian Brazil <brian.brazil@robustperception.io>
This reduces memory by only having to store the string's 16
bytes+map overheard once per label name, rather than duplicating it in every
entry for the label value.
Signed-off-by: Brian Brazil <brian.brazil@robustperception.io>
Reuse the string already allocated for symbols
in the posting tables.
Use a slice for symbols in v2 format.
Move symbol size logic into the index code.
Avoid duplication of lookupSymbol logic.
Signed-off-by: Brian Brazil <brian.brazil@robustperception.io>
This shows up as a hot spot in profiles of queries involving lots of seeks, as each seek creates a new iterator.
Signed-off-by: Tom Wilkie <tom.wilkie@gmail.com>
for tests we need to control when a compaction happens so with this
change automated compaction can be disabled, but allow to run it
manually it tests.
fixes failing tests in : https://github.com/prometheus/tsdb/pull/374
Signed-off-by: Krasi Georgiev <kgeorgie@redhat.com>
unexported NewMemTombstones as this returns unexported memTombstones
type which will not be shows in godoc.
Added missing comments for exported methods.
Removed unused RecordLogger,RecordReader interfaces.
Signed-off-by: Krasi Georgiev <kgeorgie@redhat.com>
Limit the returned `db.Querier` to the requested time range. Preallocate the `baseChunkSeries.lset` and `baseChunkSeries.chks` slices to the previous series' slice sizes to avoid unnecessary grow slice reallocations.
Calculating the modulus in each worker was a hotspot,
and meant that you had more work to do the more cores you had.
This cuts CPU usage (on my 8 core, 4 real core machine) by
33%, and walltime by 3%
Signed-off-by: Brian Brazil <brian.brazil@robustperception.io>
This is read far more than it changes.
This cuts ~14% off walltme and ~27% off CPU for WAL reading.
Signed-off-by: Brian Brazil <brian.brazil@robustperception.io>
With the various goroutines running, the locking
in getByID is notable. This cuts cpu usage by ~25%
and walltime by ~20%.
Signed-off-by: Brian Brazil <brian.brazil@robustperception.io>
