A lot of this code was hacked together, literally during a
hackathon. This commit intends not to change the code substantially,
but just make the code obey the usual style practices.
A (possibly incomplete) list of areas:
* Generally address linter warnings.
* The `pgk` directory is deprecated as per dev-summit. No new packages should
be added to it. I moved the new `pkg/histogram` package to `model`
anticipating what's proposed in #9478.
* Make the naming of the Sparse Histogram more consistent. Including
abbreviations, there were just too many names for it: SparseHistogram,
Histogram, Histo, hist, his, shs, h. The idea is to call it "Histogram" in
general. Only add "Sparse" if it is needed to avoid confusion with
conventional Histograms (which is rare because the TSDB really has no notion
of conventional Histograms). Use abbreviations only in local scope, and then
really abbreviate (not just removing three out of seven letters like in
"Histo"). This is in the spirit of
https://github.com/golang/go/wiki/CodeReviewComments#variable-names
* Several other minor name changes.
* A lot of formatting of doc comments. For one, following
https://github.com/golang/go/wiki/CodeReviewComments#comment-sentences
, but also layout question, anticipating how things will look like
when rendered by `godoc` (even where `godoc` doesn't render them
right now because they are for unexported types or not a doc comment
at all but just a normal code comment - consistency is queen!).
* Re-enabled `TestQueryLog` and `TestEndopints` (they pass now,
leaving them disabled was presumably an oversight).
* Bucket iterator for histogram.Histogram is now created with a
method.
* HistogramChunk.iterator now allows iterator recycling. (I think
@dieterbe only commented it out because he was confused by the
question in the comment.)
* HistogramAppender.Append panics now because we decided to treat
staleness marker differently.
Signed-off-by: beorn7 <beorn@grafana.com>
We are re-enabling HTTP 2 again. There has been a few bugfixes upstream
in go, and we have also enabled ReadIdleTimeout.
Fix#7588Fix#9068
Signed-off-by: Julien Pivotto <roidelapluie@inuits.eu>
* Allow to tune the scrape tolerance
In most of the classic monitoring use cases, a few milliseconds
difference can be omitted.
In Prometheus, a few millisecond difference can however make a big
difference.
Currently, Prometheus will ignore up to 2 ms difference in the
alignments.
It turns out that for users who can afford a 10ms difference, there is a
lot of resources and disk space to win, as shown in this graph, which
shows the bytes / samples over a production Prometheus server. You can
clearly see the switch from 2ms to 10ms tolerance.
This pull request enables the adjustment of the scrape timestamp
alignment tolerance.
Signed-off-by: Julien Pivotto <roidelapluie@inuits.eu>
* Fix golint
Signed-off-by: Julien Pivotto <roidelapluie@inuits.eu>
* Refactor: extract function to make scrapeLoop for testing
Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
* Add benchmarks for ScrapeLoopAppend
For Prometheus and OpenMetrics
Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
* Create less garbage when parsing metrics
Exemplar escapes to heap due to being passed through text-parser
interface, but we can reduce the impact by hoisting it out of the loop
and resetting it after every use.
(Note the cost was paid on every line even when exemplars were disabled)
Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
* Create less garbage when parsing OpenMetrics
After calling parseLVals() we always append the return value, so pass in
what we want to append it to and save garbage.
Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
This adds a new metric exposing per target scrape sample_limit value. Metrics are only exposed if extra-scrape-metrics feature flag is enabled.
scrape_sample_limit will make it easy to monitor and alert on targets getting close to configured sample_limit, which is important given than exceeding sample_limit results in the entire scrape results being rejected.
Signed-off-by: Łukasz Mierzwa <l.mierzwa@gmail.com>
Add a new built-in metric `scrape_timeout_seconds` to allow monitoring
of the ratio of scrape duration to the scrape timeout. Hide behind a
feature flag to avoid additional cardinality by default.
Signed-off-by: SuperQ <superq@gmail.com>
This "brings back" protobuf parsing, with the only goal to play with
the new sparse histograms.
The Prom-2.x style parser is highly adapted to the structure of the
Prometheus text format (and later OpenMetrics). Some jumping through
hoops is required to feed protobuf into it.
This is not meant to be a model for the final implementation. It
should just enable sparse histogram ingestion at a reasonable
efficiency.
Following known shortcomings and flaws:
- No tests yet.
- Summaries and legacy histograms, i.e. without sparse buckets, are
ignored.
- Staleness doesn't work (but this could be fixed in the appender, to
be discussed).
- No tricks have been tried that would be similar to the tricks the
text parsers do (like direct pointers into the HTTP response
body). That makes things weird here. Tricky optimizations only make
sense once the final format is specified, which will almost
certainly not be the old protobuf format. (Interestingly, I expect
this implementation to be in fact much more efficient than the
original protobuf ingestion in Prom-1.x.)
- This is using a proto3 version of metrics.proto (mostly to be
consistent with the other protobuf uses). However, proto3 sees no
difference between an unset field. We depend on that to distinguish
between an unset timestamp and the timestamp 0 (1970-01-01, 00:00:00
UTC). In this experimental code, we just assume that timestamp is
never specified and therefore a timestamp of 0 always is interpreted
as "not set".
Signed-off-by: beorn7 <beorn@grafana.com>
* Write exemplars to the WAL and send them over remote write.
Signed-off-by: Callum Styan <callumstyan@gmail.com>
* Update example for exemplars, print data in a more obvious format.
Signed-off-by: Callum Styan <callumstyan@gmail.com>
* Add metrics for remote write of exemplars.
Signed-off-by: Callum Styan <callumstyan@gmail.com>
* Fix incorrect slices passed to send in remote write.
Signed-off-by: Callum Styan <callumstyan@gmail.com>
* We need to unregister the new metrics.
Signed-off-by: Callum Styan <callumstyan@gmail.com>
* Address review comments
Signed-off-by: Callum Styan <callumstyan@gmail.com>
* Order of exemplar append vs write exemplar to WAL needs to change.
Signed-off-by: Callum Styan <callumstyan@gmail.com>
* Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars
Signed-off-by: Martin Disibio <mdisibio@gmail.com>
* Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use
Signed-off-by: Martin Disibio <mdisibio@gmail.com>
* Condense sample/exemplar delivery tests to parameterized sub-tests
Signed-off-by: Martin Disibio <mdisibio@gmail.com>
* Rename test methods for clarity now that they also handle exemplars
Signed-off-by: Martin Disibio <mdisibio@gmail.com>
* Rename counter variable. Fix instances where metrics were not updated correctly
Signed-off-by: Martin Disibio <mdisibio@gmail.com>
* Add exemplars to LoadWAL benchmark
Signed-off-by: Callum Styan <callumstyan@gmail.com>
* last exemplars timestamp metric needs to convert value to seconds with
ms precision
Signed-off-by: Callum Styan <callumstyan@gmail.com>
* Process exemplar records in a separate go routine when loading the WAL.
Signed-off-by: Callum Styan <callumstyan@gmail.com>
* Address review comments related to clarifying comments and variable
names. Also refactor sample/exemplar to enqueue prompb types.
Signed-off-by: Callum Styan <callumstyan@gmail.com>
* Regenerate types proto with comments, update protoc version again.
Signed-off-by: Callum Styan <callumstyan@gmail.com>
* Put remote write of exemplars behind a feature flag.
Signed-off-by: Callum Styan <callumstyan@gmail.com>
* Address some of Ganesh's review comments.
Signed-off-by: Callum Styan <callumstyan@gmail.com>
* Move exemplar remote write feature flag to a config file field.
Signed-off-by: Callum Styan <callumstyan@gmail.com>
* Address Bartek's review comments.
Signed-off-by: Callum Styan <callumstyan@gmail.com>
* Don't allocate exemplar buffers in queue_manager if we're not going to
send exemplars over remote write.
Signed-off-by: Callum Styan <callumstyan@gmail.com>
* Add ValidateExemplar function, validate exemplars when appending to head
and log them all to WAL before adding them to exemplar storage.
Signed-off-by: Callum Styan <callumstyan@gmail.com>
* Address more reivew comments from Ganesh.
Signed-off-by: Callum Styan <callumstyan@gmail.com>
* Add exemplar total label length check.
Signed-off-by: Callum Styan <callumstyan@gmail.com>
* Address a few last review comments
Signed-off-by: Callum Styan <callumstyan@gmail.com>
Co-authored-by: Martin Disibio <mdisibio@gmail.com>
* scrape: add label limits per scrape
Add three new limits to the scrape configuration to provide some
mechanism to defend against unbound number of labels and excessive
label lengths. If any of these limits are broken by a sample from a
scrape, the whole scrape will fail. For all of these configuration
options, a zero value means no limit.
The `label_limit` configuration will provide a mechanism to bound the
number of labels per-scrape of a certain sample to a user defined limit.
This limit will be tested against the sample labels plus the discovery
labels, but it will exclude the __name__ from the count since it is a
mandatory Prometheus label to which applying constraints isn't
meaningful.
The `label_name_length_limit` and `label_value_length_limit` will
prevent having labels of excessive lengths. These limits also skip the
__name__ label for the same reasons as the `label_limit` option and will
also make the scrape fail if any sample has a label name/value length
that exceed the predefined limits.
Signed-off-by: Damien Grisonnet <dgrisonn@redhat.com>
* scrape: add metrics and alert to label limits
Add three gauge, one for each label limit to easily access the
limit set by a certain scrape target.
Also add a counter to count the number of targets that exceeded the
label limits and thus were dropped. This is useful for the
`PrometheusLabelLimitHit` alert that will notify the users that scraping
some targets failed because they had samples exceeding the label limits
defined in the scrape configuration.
Signed-off-by: Damien Grisonnet <dgrisonn@redhat.com>
* scrape: apply label limits to __name__ label
Apply limits to the __name__ label that was previously skipped and
truncate the label names and values in the error messages as they can be
very very long.
Signed-off-by: Damien Grisonnet <dgrisonn@redhat.com>
* scrape: remove label limits gauges and refactor
Remove `prometheus_target_scrape_pool_label_limit`,
`prometheus_target_scrape_pool_label_name_length_limit`, and
`prometheus_target_scrape_pool_label_value_length_limit` as they are not
really useful since we don't have the information on the labels in it.
Signed-off-by: Damien Grisonnet <dgrisonn@redhat.com>
This moves the label lookup into TSDB, whilst still keeping the cached-ref optimisation for repeated Appends.
This makes the API easier to consume and implement. In particular this change is motivated by the scrape-time-aggregation work, which I don't think is possible to implement without it as it needs access to label values.
Signed-off-by: Tom Wilkie <tom.wilkie@gmail.com>
Manager.reload takes the mutex that would make it safe, however
releases it before the goroutines spawned are finished with it.
Thus more explicit locking of scrapePool.Sync/stop/reload is needed.
Signed-off-by: Brian Brazil <brian.brazil@robustperception.io>
Don't lock for all of Sync/stop/reload as that holds up /metrics and the
UI when they want a list of active/dropped targets. Instead take
advantage of the fact that Sync/stop/reload cannot be called
concurrently by the scrape Manager and lock just on the targets
themselves.
Signed-off-by: Brian Brazil <brian.brazil@robustperception.io>
This also fixes a bug in query_log_file, which now is relative to the config file like all other paths.
Signed-off-by: Andy Bursavich <abursavich@gmail.com>
When I started wotking on target_limit, scrapeAndReport did not exist
yet. Then I simply rebased my work without thinking.
It appears that there is a lot that can be inline if I defer() the
report.
Signed-off-by: Julien Pivotto <roidelapluie@inuits.eu>
* Separate scrape add error checking out into it's own function.
Signed-off-by: Callum Styan <callumstyan@gmail.com>
* pass sampleLimitError to checkAddError instead of returning an error
Signed-off-by: Callum Styan <callumstyan@gmail.com>
* Return bool, error from checkAddError so we can properly handle
ErrNotFound for AddFast. This should in theory never happen, but the
previous code path handled this case. Adds a test for this, which master
passes and the previous commit fails.
Signed-off-by: Callum Styan <callumstyan@gmail.com>
* Address comment changes.
Signed-off-by: Callum Styan <callumstyan@gmail.com>
* Move sampleAdded inside the loop iteration within append, since that's
the only block the variable is used in.
Signed-off-by: Callum Styan <callumstyan@gmail.com>
This fixes#6992, which was introduced by #6777. There was an
intermediate component which translated TSDB errors into storage errors,
but that component was deleted and this bug went unnoticed, until we
were watching at the Prombench results. Without this, scrape will fail
instead of dropping samples or using "Add" when the series have been
garbage collected.
Signed-off-by: Julien Pivotto <roidelapluie@inuits.eu>
With defer having less of a performance penalty, there is no reason
not to do those crucial operations via defer.
Context: With isolation in place, if we forget to Commit/Rollback, the
low watermark will get stuck forever.
The current code should not have any bugs, but moving to defer helps
to avoid future bugs.
This is also moving the `closeAppend` in the `Commit` implementation
itself to defer. If logging to the WAL fails, we would have missed the
`closeAppend`.
Signed-off-by: beorn7 <beorn@grafana.com>
This is most likely due to an endpoint not producing valid
metrics output, which we should treat the same as a failed
scrape, and thus not spam the application logs with it.
Signed-off-by: Brian Brazil <brian.brazil@robustperception.io>
* [comments] change word ‘wheter’ to ‘whether’
Signed-off-by: fuling <fuling.lgz@alibaba-inc.com>
* [comments] change word ‘wheter’ to ‘whether’
Signed-off-by: fuling <fuling.lgz@alibaba-inc.com>
* tsdb: don't allow ingesting empty labelsets
When we ingest an empty labelset in the head, further blocks can not be
compacted, with the error:
```
level=error ts=2020-02-27T21:26:58.379Z caller=db.go:659 component=tsdb
msg="compaction failed" err="persist head block: write compaction:
add series: out-of-order series added with label set \"{}\" / prev:
\"{}\""
```
We should therefore reject those invalid empty labelsets upfront.
This can be reproduced with the following:
```
cat << END > prometheus.yml
scrape_configs:
- job_name: 'prometheus'
scrape_interval: 1s
basic_auth:
username: test
password: test
metric_relabel_configs:
- regex: ".*"
action: labeldrop
static_configs:
- targets:
- 127.0.1.1:9090
END
./prometheus --storage.tsdb.min-block-duration=1m
```
And wait a few minutes.
Signed-off-by: Julien Pivotto <roidelapluie@inuits.eu>
This is part of https://github.com/prometheus/prometheus/pull/5882 that can be done to simplify things.
All todos I added will be fixed in follow up PRs.
* querier.Querier, querier.Appender, querier.SeriesSet, and querier.Series interfaces merged
with storage interface.go. All imports that.
* querier.SeriesIterator replaced by chunkenc.Iterator
* Added chunkenc.Iterator.Seek method and tests for xor implementation (?)
* Since we properly handle SelectParams for Select methods I adjusted min max
based on that. This should help in terms of performance for queries with functions like offset.
* added Seek to deletedIterator and test.
* storage/tsdb was removed as it was only a unnecessary glue with incompatible structs.
No logic was changed, only different source of abstractions, so no need for benchmarks.
Signed-off-by: Bartlomiej Plotka <bwplotka@gmail.com>
To test the implementation of our metric metadata API, we need to represent various states of metadata in the scrape metadata store. That is currently not possible as the interface and method to set the store are private.
This changes the interface, list and get methods, and the SetMetadaStore function to be public.
Incidentally, the scrapeCache implementation needs to be renamed to match the new signature.
Signed-off-by: gotjosh <josue@grafana.com>
When using both a label and the suffix+label in the
relabel config. It's possible that Prometheus remove
the suffx+label for no obvious reason. It's due to a
collision when merging labels from target and from
the sample.
Signed-off-by: Geoffrey Beausire <g.beausire@criteo.com>
* Update go.mod dependencies before release
Signed-off-by: Julius Volz <julius.volz@gmail.com>
* Add issue for showing query warnings in promtool
Signed-off-by: Julius Volz <julius.volz@gmail.com>
* Revert json-iterator back to 1.1.6
It produced errors when marshaling Point values with special float
values.
Signed-off-by: Julius Volz <julius.volz@gmail.com>
* Fix expected step values in promtool tests after client_golang update
Signed-off-by: Julius Volz <julius.volz@gmail.com>
* Update generated protobuf code after proto dep updates
Signed-off-by: Julius Volz <julius.volz@gmail.com>
Fixes https://github.com/prometheus/common/issues/36
Move logic handling this into the labels package,
so all the cases are handled in one place and we're
less likely to have this come up again.
Signed-off-by: Brian Brazil <brian.brazil@robustperception.io>
This is an estimate of churn, with series being added to the cache being
considered churn. This will have both false positives (e.g. series
appearing and disappearing) and false negatives (e.g. series hit
sample_limit, but still created in head block), but should be generally
useful as-is.
Relevant docs live in another repo.
Signed-off-by: Brian Brazil <brian.brazil@robustperception.io>
From the documentation:
> The default HTTP client's Transport may not
> reuse HTTP/1.x "keep-alive" TCP connections if the Body is
> not read to completion and closed.
This effectively enable keep-alive for the fixed requests.
Signed-off-by: Romain Baugue <romain.baugue@elwinar.com>
* Reload certificates from disk automatically
This change bumps github.com/prometheus/common to include
https://github.com/prometheus/common/pull/173
Signed-off-by: Simon Pasquier <spasquie@redhat.com>
* scrape: close idle connections on reload/stop
Signed-off-by: Simon Pasquier <spasquie@redhat.com>
* use v0.3.0 tag
Signed-off-by: Simon Pasquier <spasquie@redhat.com>
Now that we're not losing the scrape cache across failed
scrape, a scrape that continually failed but had varying
series or metadata (e.g. timestamps in metric names,
plus hitting smaple_limit) would grow the cache indefinitely.
Add some code to catch that, and flush the cache anyway.
Signed-off-by: Brian Brazil <brian.brazil@robustperception.io>
i) Uses the more idiomatic Wrap and Wrapf methods for creating nested errors.
ii) Fixes some incorrect usages of fmt.Errorf where the error messages don't have any formatting directives.
iii) Does away with the use of fmt package for errors in favour of pkg/errors
Signed-off-by: tariqibrahim <tariq181290@gmail.com>
* scrape: Add global jitter for HA server
Covers issue in https://github.com/prometheus/prometheus/pull/4926#issuecomment-449039848
where the HA setup become a problem for targets unable to be scraped simultaneously.
The new jitter per server relies on the hostname and external labels which necessarily to be uniq.
As before, scrape offset will be calculated with regard the absolute time, so even
restart/reload doesn't change scrape time per scrape target + prometheus instance.
Use fqdn if possible, otherwise fall back to the hostname. It adds extra random seed
to calculate server hash to be distinguish on machines with the same hostname, but
different DC.
Signed-off-by: Aleksei Semiglazov <xjewer@gmail.com>
* scrape: catch errors when creating HTTP clients
This change makes sure that no scrape pool is created with a nil HTTP
client.
Signed-off-by: Simon Pasquier <spasquie@redhat.com>
* Address Tariq's comment
Signed-off-by: Simon Pasquier <spasquie@redhat.com>
* Address Brian's comment
Signed-off-by: Simon Pasquier <spasquie@redhat.com>
* *: use latest release of staticcheck
It also fixes a couple of things in the code flagged by the additional
checks.
Signed-off-by: Simon Pasquier <spasquie@redhat.com>
* Use official release of staticcheck
Also run 'go list' before staticcheck to avoid failures when downloading packages.
Signed-off-by: Simon Pasquier <spasquie@redhat.com>
* buf Has been declared in scrape.go in line 785, I think it is unnecessary to declare a new variable again here.
Signed-off-by: arugakiWei <arugaki.wei@daocloud.io>
* delete the buf in line 785 because it is never used.
Signed-off-by: arugakiWei <arugaki.wei@daocloud.io>
* *: remove use of golang.org/x/net/context
Signed-off-by: Simon Pasquier <spasquie@redhat.com>
* scrape: fix TestTargetScrapeScrapeCancel
Signed-off-by: Simon Pasquier <spasquie@redhat.com>
Pass content type down to text parser.
Add layer of indirection in front of text parser,
and rename to avoid future clashes.
Signed-off-by: Brian Brazil <brian.brazil@robustperception.io>
The scrape manage receiver's channel now just saves the target sets
and another backgorund runner updates the scrape loops every 5 seconds.
This is so that the scrape manager doesn't block the receiving channel
when it does the long background reloading of the scrape loops.
Active and dropped targets are now saved in each scrape pool instead of
the scrape manager. This is mainly to avoid races when getting the
targets via the web api.
When reloading the scrape loops now happens in parallel to speed up the
final disared state and this also speeds up the prometheus's shutting
down.
Also updated some funcs signatures in the web package for consistency.
Signed-off-by: Krasi Georgiev <kgeorgie@redhat.com>
There are many more (mostly finalizers like Close/Stop/etc.), but most of
the others seemed like one couldn't do much about them anyway.
Signed-off-by: Julius Volz <julius.volz@gmail.com>
This adds a per-target cache of scraped metadata. The metadata is only
available for the lifecycle of the attached target. An API endpoint allows
to select metadata by metric name and a label selection of targets.
Signed-off-by: Fabian Reinartz <freinartz@google.com>
Extends the parser to allow retrieving metadata.
The lexer now yields proper tokens that are fed into a hand-written
parser on top.
Signed-off-by: Fabian Reinartz <freinartz@google.com>
This commit avoids passing the full scrape configuration down to the
scrape loop to fix data races when the scrape configuration is being
reloaded.
Signed-off-by: Simon Pasquier <spasquie@redhat.com>
Fix race by properly locking access to scrape pools. Use separate mutex for information needed by UI so that UI isn't blocked when targets are being updated.
The semantics of honor_labels are that if a target exposes
and empty label it will override the target labels. This PR
fixes that by once again distinguishing between empty labels
and missing labels in this one use case.
Beyond that empty labels should be pruned and not added to storage,
which this also fixes.
Fixes#3841
DrAuYueng