* add hook to allow head compaction to create multiple output blocks
Signed-off-by: Ben Ye <benye@amazon.com>
* change Compact interface; remove BlockPopulator changes
Signed-off-by: Ben Ye <benye@amazon.com>
* rebase main
Signed-off-by: Ben Ye <benye@amazon.com>
* fix lint
Signed-off-by: Ben Ye <benye@amazon.com>
* fix unit test
Signed-off-by: Ben Ye <benye@amazon.com>
* address feedbacks; add unit test
Signed-off-by: Ben Ye <benye@amazon.com>
* Apply suggestions from code review
Signed-off-by: Ganesh Vernekar <ganeshvern@gmail.com>
* Update tsdb/compact_test.go
Signed-off-by: Ganesh Vernekar <ganeshvern@gmail.com>
---------
Signed-off-by: Ben Ye <benye@amazon.com>
Signed-off-by: Ganesh Vernekar <ganeshvern@gmail.com>
Co-authored-by: Ganesh Vernekar <ganeshvern@gmail.com>
* expose hook in tsdb to allow customizing compactor
Signed-off-by: Ben Ye <benye@amazon.com>
* address comment
Signed-off-by: Ben Ye <benye@amazon.com>
---------
Signed-off-by: Ben Ye <benye@amazon.com>
Now the error will include the timestamp and the existing and new values.
When you are trying to track down the source of this error, it can be
useful to see that the values are close, or alternating, or something
else.
Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
use it in loadDataAsQueryable to make sure the RO Head doesn't truncate or cut new chunks in data/chunks_head/.
add a -sandbox-dir-root flag to "promtool tsdb dump/dump-openmetrics" to control the root of that sandbox dirrectory.
Signed-off-by: machine424 <ayoubmrini424@gmail.com>
Signed-off-by: Jonathan Halterman <jonathan@grafana.com>
Signed-off-by: Jonathan Halterman <jhalterman@gmail.com>
Co-authored-by: Jesus Vazquez <jesusvazquez@users.noreply.github.com>
* Stop compactions if there's a block to write
db.Compact() checks if there's a block to write with HEAD chunks before calling db.compactBlocks().
This is to ensure that if we need to write a block then it happens ASAP, otherwise memory usage might keep growing.
But what can also happen is that we don't need to write any block, we start db.compactBlocks(),
compaction takes hours, and in the meantime HEAD needs to write out chunks to a block.
This can be especially problematic if, for example, you run Thanos sidecar that's uploading block,
which requires that compactions are disabled. Then you disable Thanos sidecar and re-enable compactions.
When db.compactBlocks() is finally called it might have a huge number of blocks to compact, which might
take a very long time, during which HEAD cannot write out chunks to a new block.
In such case memory usage will keep growing until either:
- compactions are finally finished and HEAD can write a block
- we run out of memory and Prometheus gets OOM-killed
This change adds a check for pending HEAD block writes inside db.compactBlocks(), so that
we bail out early if there are still compactions to run, but we also need to write a new
block.
Also add a test for compactBlocks.
---------
Signed-off-by: Łukasz Mierzwa <l.mierzwa@gmail.com>
Signed-off-by: Lukasz Mierzwa <lukasz@cloudflare.com>
* TSDB: Don't compact the head block when empty
Don't compact the Head block if there have not yet been any samples
appended.
Previously, the logic for determining if the head should be compacted
relied on the default values for min and max time and integer overflow
when they were checked in `Head.compactable()`. The check in
`Head.compactable()` effectively did `math.MinInt64 - math.MaxInt64`
which overflowed and wrapped to `1`. Since `1` is less than `1.5`
times the chunk range, compaction did not happen. This was the correct
behavior but relying on overflow wrapping is surprising.
This change add a method for checking if the min and max time for the
head is unset and uses it to short-circuit compaction in that case.
It also replaces several explicit checks for the default value to
determine if the head has not yet had any samples added.
Signed-off-by: Nick Pillitteri <nick.pillitteri@grafana.com>
For instance `require.NoError` will print the unexpected error; we don't
need to include it in the message.
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>
Digging around the TSDB code and I've found that this flag is unused so
let's remove it.
Signed-off-by: Giedrius Statkevičius <giedrius.statkevicius@vinted.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>
The ChunkReader interface's Chunk() has been changed to ChunkOrIterable().
This is a precursor to OOO native histogram support - with OOO native histograms, the chunks.Meta passed to Chunk() can result in multiple chunks being returned rather than just a single chunk (e.g. if oooMergedChunk has a counter reset in the middle).
To support this, ChunkOrIterable() requires either a single chunk or an iterable to be returned. If an iterable is returned, the caller has the responsibility of converting the samples from the iterable into possibly multiple chunks. The OOOHeadChunkReader now returns an iterable rather than a chunk to prepare for the native histograms case. Also as a beneficial side effect, oooMergedChunk and boundedChunk has been simplified as they only need to implement the Iterable interface now, not the full Chunk interface.
---------
Signed-off-by: Fiona Liao <fiona.y.liao@gmail.com>
Co-authored-by: George Krajcsovits <krajorama@users.noreply.github.com>
* Add failing test.
Signed-off-by: Charles Korn <charles.korn@grafana.com>
* Don't run OOO head garbage collection while reads are running.
Signed-off-by: Charles Korn <charles.korn@grafana.com>
* Add further test cases for different order of operations.
Signed-off-by: Charles Korn <charles.korn@grafana.com>
* Ensure all queriers are closed if `DB.blockChunkQuerierForRange()` fails.
Signed-off-by: Charles Korn <charles.korn@grafana.com>
* Ensure all queriers are closed if `DB.Querier()` fails.
Signed-off-by: Charles Korn <charles.korn@grafana.com>
* Invert error handling in `DB.Querier()` and `DB.blockChunkQuerierForRange()` to make it clearer
Signed-off-by: Charles Korn <charles.korn@grafana.com>
* Ensure that queries that touch OOO data can't block OOO head garbage collection forever.
Signed-off-by: Charles Korn <charles.korn@grafana.com>
* Address PR feedback: fix parameter name in comment
Co-authored-by: Jesus Vazquez <jesusvazquez@users.noreply.github.com>
Signed-off-by: Charles Korn <charleskorn@users.noreply.github.com>
* Address PR feedback: use `lastGarbageCollectedMmapRef`
Signed-off-by: Charles Korn <charles.korn@grafana.com>
* Address PR feedback: ensure pending reads are cleaned up if creating an OOO querier fails
Signed-off-by: Charles Korn <charles.korn@grafana.com>
---------
Signed-off-by: Charles Korn <charles.korn@grafana.com>
Signed-off-by: Charles Korn <charleskorn@users.noreply.github.com>
Co-authored-by: Jesus Vazquez <jesusvazquez@users.noreply.github.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>
* Additionally wrap WBL replay error
Although WBL replay is already wrapped with errLoadWbl,
there are other errors that can happen during a WBL replay.
We should not try to repair WAL in those cases.
This commit additionally wraps the final error in Head.Init again
with errLoadWbl so that WBL replay errors can be identified properly.
Signed-off-by: Ganesh Vernekar <ganeshvern@gmail.com>
Signed-off-by: Jesus Vazquez <jesusvzpg@gmail.com>
Co-authored-by: Jesus Vazquez <jesusvzpg@gmail.com>
Signed-off-by: Levi Harrison <git@leviharrison.dev>
* Additionally wrap WBL replay error
Although WBL replay is already wrapped with errLoadWbl,
there are other errors that can happen during a WBL replay.
We should not try to repair WAL in those cases.
This commit additionally wraps the final error in Head.Init again
with errLoadWbl so that WBL replay errors can be identified properly.
Signed-off-by: Ganesh Vernekar <ganeshvern@gmail.com>
Signed-off-by: Jesus Vazquez <jesusvzpg@gmail.com>
Co-authored-by: Jesus Vazquez <jesusvzpg@gmail.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>
Add a chunk size limit in bytes
This creates a hard cap for XOR chunks of 1024 bytes.
The limit for histogram chunk is also 1024 bytes, but it is a soft limit as a histogram has a dynamic size, and even a single one could be larger than 1024 bytes.
This also avoids cutting new histogram chunks if the existing chunk has fewer than 10 histograms yet. In that way, we are accepting "jumbo chunks" in order to have at least 10 histograms in a chunk, allowing compression to kick in.
Signed-off-by: Justin Lei <justin.lei@grafana.com>
So far, `ValidateHistogram` would not detect if the count did not
include the count in the zero bucket. This commit fixes the problem
and updates all the tests that have been undetected offenders so far.
Note that this problem would only ever create false negatives, so we
never falsely rejected to store a histogram because of it.
On the other hand, `ValidateFloatHistogram` has been to strict with
the count being at least as large as the sum of the counts in all the
buckets. Float precision issues could create false positives here, see
products of PromQL evaluations, it's actually quite hard to put an
upper limit no the floating point imprecision. Users could produce the
weirdest expressions, maxing out float precision problems. Therefore,
this commit simply removes that particular check from
`ValidateFloatHistogram`.
Signed-off-by: beorn7 <beorn@grafana.com>
Simlar to cleanup of WAL files on startup, cleanup temporary
chunk_snapshot dirs. This prevents storage space leaks due to terminated
snapshots on shutdown.
Signed-off-by: SuperQ <superq@gmail.com>
Currently memSeries holds a single head chunk in-memory and a slice of mmapped chunks.
When append() is called on memSeries it might decide that a new headChunk is needed to use for given append() call.
If that happens it will first mmap existing head chunk and only after that happens it will create a new empty headChunk and continue appending
our sample to it.
Since appending samples uses write lock on memSeries no other read or write can happen until any append is completed.
When we have an append() that must create a new head chunk the whole memSeries is blocked until mmapping of existing head chunk finishes.
Mmapping itself uses a lock as it needs to be serialised, which means that the more chunks to mmap we have the longer each chunk might wait
for it to be mmapped.
If there's enough chunks that require mmapping some memSeries will be locked for long enough that it will start affecting
queries and scrapes.
Queries might timeout, since by default they have a 2 minute timeout set.
Scrapes will be blocked inside append() call, which means there will be a gap between samples. This will first affect range queries
or calls using rate() and such, since the time range requested in the query might have too few samples to calculate anything.
To avoid this we need to remove mmapping from append path, since mmapping is blocking.
But this means that when we cut a new head chunk we need to keep the old one around, so we can mmap it later.
This change makes memSeries.headChunk a linked list, memSeries.headChunk still points to the 'open' head chunk that receives new samples,
while older, yet to be mmapped, chunks are linked to it.
Mmapping is done on a schedule by iterating all memSeries one by one. Thanks to this we control when mmapping is done, since we trigger
it manually, which reduces the risk that it will have to compete for mmap locks with other chunks.
Signed-off-by: Łukasz Mierzwa <l.mierzwa@gmail.com>
