### Describe Your Changes
Currently, vmagent always uses a separate `http.Client` for every group
watcher in Kubernetes SD. With a high number of group watchers this
leads to large amount of opened connections.
This PR adds 2 changes to address this:
- re-use of existing `http.Client` - in case `http.Client` is connecting
to the same API server and uses the same parameters it will be re-used
between group watchers
- HTTP2 support - this allows to reuse connections more efficiently due
to ability of using streaming via existing connections.
See this issue for the details and test results -
https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5971
### Checklist
The following checks are **mandatory**:
- [ ] My change adheres [VictoriaMetrics contributing
guidelines](https://docs.victoriametrics.com/contributing/).
---------
Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com>
Co-authored-by: Roman Khavronenko <roman@victoriametrics.com>
This reverts commit 5a3abfa041.
Reason for revert: exemplars aren't in wide use because they have numerous issues which prevent their adoption (see below).
Adding support for examplars into VictoriaMetrics introduces non-trivial code changes. These code changes need to be supported forever
once the release of VictoriaMetrics with exemplar support is published. That's why I don't think this is a good feature despite
that the source code of the reverted commit has an excellent quality. See https://docs.victoriametrics.com/goals/ .
Issues with Prometheus exemplars:
- Prometheus still has only experimental support for exemplars after more than three years since they were introduced.
It stores exemplars in memory, so they are lost after Prometheus restart. This doesn't look like production-ready feature.
See 0a2f3b3794/content/docs/instrumenting/exposition_formats.md (L153-L159)
and https://prometheus.io/docs/prometheus/latest/feature_flags/#exemplars-storage
- It is very non-trivial to expose exemplars alongside metrics in your application, since the official Prometheus SDKs
for metrics' exposition ( https://prometheus.io/docs/instrumenting/clientlibs/ ) either have very hard-to-use API
for exposing histograms or do not have this API at all. For example, try figuring out how to expose exemplars
via https://pkg.go.dev/github.com/prometheus/client_golang@v1.19.1/prometheus .
- It looks like exemplars are supported for Histogram metric types only -
see https://pkg.go.dev/github.com/prometheus/client_golang@v1.19.1/prometheus#Timer.ObserveDurationWithExemplar .
Exemplars aren't supported for Counter, Gauge and Summary metric types.
- Grafana has very poor support for Prometheus exemplars. It looks like it supports exemplars only when the query
contains histogram_quantile() function. It queries exemplars via special Prometheus API -
https://prometheus.io/docs/prometheus/latest/querying/api/#querying-exemplars - (which is still marked as experimental, btw.)
and then displays all the returned exemplars on the graph as special dots. The issue is that this doesn't work
in production in most cases when the histogram_quantile() is calculated over thousands of histogram buckets
exposed by big number of application instances. Every histogram bucket may expose an exemplar on every timestamp shown on the graph.
This makes the graph unusable, since it is litterally filled with thousands of exemplar dots.
Neither Prometheus API nor Grafana doesn't provide the ability to filter out unneeded exemplars.
- Exemplars are usually connected to traces. While traces are good for some
I doubt exemplars will become production-ready in the near future because of the issues outlined above.
Alternative to exemplars:
Exemplars are marketed as a silver bullet for the correlation between metrics, traces and logs -
just click the exemplar dot on some graph in Grafana and instantly see the corresponding trace or log entry!
This doesn't work as expected in production as shown above. Are there better solutions, which work in production?
Yes - just use time-based and label-based correlation between metrics, traces and logs. Assign the same `job`
and `instance` labels to metrics, logs and traces, so you can quickly find the needed trace or log entry
by these labes on the time range with the anomaly on metrics' graph.
This code adds Exemplars to VMagent and the promscrape parser adhering
to OpenMetrics Specifications. This will allow forwarding of exemplars
to Prometheus and other third party apps that support OpenMetrics specs.
---------
Signed-off-by: Ted Possible <ted_possible@cable.comcast.com>
* vmagent: export `vm_promscrape_scrape_pool_targets` metric to track the number of targets that each scrape_job discovers
* add extra panel for new metric
* fix inconsistent behaviors with prometheus when scraping
1. address https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959. skip job with wrong syntax in `scrape_configs` with error logs instead of exiting;
2. show error messages on vmagent /targets ui if there are wrong auth configs in `scrape_configs`, previously will print error logs and do scrape without auth header;
3. don't send requests if there are wrong auth configs in:
1. vmagent remoteWrite;
2. vmalert datasource/remoteRead/remoteWrite/notifier.
* add changelogs
* address review comments
* fix ut
* lib/promscrape: make concurrency control optional
Before, `-maxConcurrentInserts` was limiting all calls to `promscrape.Parse`
function: during ingestion and scraping. This behavior is incorrect.
Cmd-line flag `-maxConcurrentInserts` should have effect onl on ingestion.
Since both pipelines use the same `promscrape.Parse` function, we extend it
to make concurrency limiter optional. So caller can decide whether concurrency
should be limited or not.
This commit makes c53b5788b4
obsolete.
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* Revert "dashboards: move `Concurrent inserts` panel to Troubleshooting section"
This reverts commit c53b5788b4.
---------
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* lib/promscrape: add metric `vm_promscrape_scrapes_skipped_total`
add metric `vm_promscrape_scrapes_skipped_total`to show whether vmagent skips the scrapes.
This could happen if vmagent is overloaded or target is responding too slow for configured `scrape_interval`.
The follow-up commit should add a corresponding alerting rule and panel to vmagent dashboard.
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* deployment/docker: add `TooManyScrapeSkips` alerting rule for vmagent
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* dashboards: add panels `Scrape duration 0.99 quantile` and `Skipped scrapes` to vmagent dashboard
Signed-off-by: hagen1778 <roman@victoriametrics.com>
---------
Signed-off-by: hagen1778 <roman@victoriametrics.com>
`lib/protoparser/prometheus` is used by various applications,
such as `app/vmalert`. The recent change to the
`lib/protoparser/prometheus` package introduced a new dependency
of `lib/writeconcurrencylimiter` which exposes some metrics.
Because of the dependency, now all applications which have this
dependency also expose these metrics.
Creating a new `lib/protoparser/prometheus/stream` package helps
to remove these metrics from apps which use `lib/protoparser/prometheus`
as dependency.
See https://github.com/VictoriaMetrics/VictoriaMetrics/issues/3761
Signed-off-by: hagen1778 <roman@victoriametrics.com>
Stale series are sent when there is a difference between current
and previous scrapes. Those series which disappeared in the current scrape
are marked as stale and sent to the remote storage.
Sending stale series requires memory allocation and in case when too many
series disappear in the same it could result in noticeable memory spike.
For example, re-deploy of a big fleet of service can result into
excessive memory usage for vmagent, because all the series with old
pod name will be marked as stale and sent to the remote write storage.
This change limits the number of stale series which can be sent at once,
so memory usage remains steady.
https://github.com/VictoriaMetrics/VictoriaMetrics/issues/3668https://github.com/VictoriaMetrics/VictoriaMetrics/issues/3675
Signed-off-by: hagen1778 <roman@victoriametrics.com>
Signed-off-by: hagen1778 <roman@victoriametrics.com>
- Document the fix at docs/CHANGELOG.md
- Limit the concurrency for sendStaleMarkers() function in order to limit its memory usage
when big number of targets disappear and staleness markers are sent
for all the metrics exposed by these targets.
- Make sure that the writeRequestCtx is returned to the pool
when there is no need to send staleness markers.
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/3668
This should reduce the maximum memory usage for processScrapedData() function by 2x.
The only part, which can be IO-bound in the processScrapedData() is pushData() call,
when it buffers data to persistent queue if the remote storage cannot keep up
with the data ingestion speed. In this case it is OK if the scrape pace will be limited.
This should reduce memory usage when scraping big number of targets,
since this limits the summary memory usage during concurrent parsing and relabeling
by the number of available CPU cores.
This fixes handling of values bigger than 2GiB for the following command-line flags:
- -storage.minFreeDiskSpaceBytes
- -remoteWrite.maxDiskUsagePerURL
- Return meta-labels for the discovered targets via promutils.Labels
instead of map[string]string. This improves the speed of generating
meta-labels for discovered targets by up to 5x.
- Remove memory allocations in hot paths during ScrapeWork generation.
The ScrapeWork contains scrape settings for a single discovered target.
This improves the service discovery speed by up to 2x.
The issue was in the `labels := dst[offset:]` line in the beginning of appendExtraLabels() function.
The `dst` may be re-allocated when adding extra labels to it. In this case the addition of `exported_`
prefix to labels inside `labels` slice become invisible in the returned `dst` labels.
While at it, properly handle some corner cases:
- Add additional `exported_` prefix to clashing metric labels with already existing `exported_` prefix.
- Store scraped metric names in `exported___name__` label if scrape target contains `__name__` label.
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/3278
Thanks to @jplanckeel for the initial attempt to fix this issue
at https://github.com/VictoriaMetrics/VictoriaMetrics/pull/3281
Sort labels explicitly after calling the ParsedConfigs.Apply() when needed.
This reduces CPU usage when performing metric-level relabeling, where labels' sorting isn't needed.
Previously the `__address__` label could contain only `host:port` part of the target url,
while the scheme and metrics path were obtained from `__scheme__` and `__metrics_path__`
labels. Now it is possible to set the full url in `__address__` label.
This makes valid the following scrape config, which is frequently used by novice users:
scrape_configs:
- job_name: foo
static_configs:
- targets:
- http://host1/metrics1
- https://host2/metrics2
This reduces scrape duration for targets returning big responses.
The response body was already read into memory in stream parsing mode before this change,
so this commit shouldn't increase memory usage.
ioutil.ReadAll is deprecated since Go1.16 - see https://tip.golang.org/doc/go1.16#ioutil
VictoriaMetrics requires at least Go1.18, so it is OK to switch from ioutil.ReadAll to io.ReadAll.
This is a follow-up for 02ca2342ab
