Feature flags

Here is a list of features that are disabled by default since they are breaking changes or are considered experimental. Their behaviour can change in future releases which will be communicated via the release changelog.

You can enable them using the --enable-feature flag with a comma separated list of features. They may be enabled by default in future versions.

Expand environment variables in external labels


Replace ${var} or $var in the external_labels values according to the values of the current environment variables. References to undefined variables are replaced by the empty string. The $ character can be escaped by using $$.

Remote Write Receiver


The remote write receiver allows Prometheus to accept remote write requests from other Prometheus servers. More details can be found here.

Activating the remote write receiver via a feature flag is deprecated. Use --web.enable-remote-write-receiver instead. This feature flag will be ignored in future versions of Prometheus.

Exemplars storage


OpenMetrics introduces the ability for scrape targets to add exemplars to certain metrics. Exemplars are references to data outside of the MetricSet. A common use case are IDs of program traces.

Exemplar storage is implemented as a fixed size circular buffer that stores exemplars in memory for all series. Enabling this feature will enable the storage of exemplars scraped by Prometheus. The config file block storage/exemplars can be used to control the size of circular buffer by # of exemplars. An exemplar with just a trace_id=<jaeger-trace-id> uses roughly 100 bytes of memory via the in-memory exemplar storage. If the exemplar storage is enabled, we will also append the exemplars to WAL for local persistence (for WAL duration).

Memory snapshot on shutdown


This takes the snapshot of the chunks that are in memory along with the series information when shutting down and stores it on disk. This will reduce the startup time since the memory state can be restored with this snapshot and m-mapped chunks without the need of WAL replay.

Extra scrape metrics


When enabled, for each instance scrape, Prometheus stores a sample in the following additional time series:

  • scrape_timeout_seconds. The configured scrape_timeout for a target. This allows you to measure each target to find out how close they are to timing out with scrape_duration_seconds / scrape_timeout_seconds.
  • scrape_sample_limit. The configured sample_limit for a target. This allows you to measure each target to find out how close they are to reaching the limit with scrape_samples_post_metric_relabeling / scrape_sample_limit. Note that scrape_sample_limit can be zero if there is no limit configured, which means that the query above can return +Inf for targets with no limit (as we divide by zero). If you want to query only for targets that do have a sample limit use this query: scrape_samples_post_metric_relabeling / (scrape_sample_limit > 0).
  • scrape_body_size_bytes. The uncompressed size of the most recent scrape response, if successful. Scrapes failing because body_size_limit is exceeded report -1, other scrape failures report 0.

New service discovery manager


When enabled, Prometheus uses a new service discovery manager that does not restart unchanged discoveries upon reloading. This makes reloads faster and reduces pressure on service discoveries' sources.

Users are encouraged to test the new service discovery manager and report any issues upstream.

In future releases, this new service discovery manager will become the default and this feature flag will be ignored.

Prometheus agent


When enabled, Prometheus runs in agent mode. The agent mode is limited to discovery, scrape and remote write.

This is useful when you do not need to query the Prometheus data locally, but only from a central remote endpoint.

Per-step stats


When enabled, passing stats=all in a query request returns per-step statistics. Currently this is limited to totalQueryableSamples.

When disabled in either the engine or the query, per-step statistics are not computed at all.



When enabled, GOMAXPROCS variable is automatically set to match Linux container CPU quota.



When enabled, the GOMEMLIMIT variable is automatically set to match the Linux container memory limit. If there is no container limit, or the process is running outside of containers, the system memory total is used.

There is also an additional tuning flag, --auto-gomemlimit.ratio, which allows controlling how much of the memory is used for Prometheus. The remainder is reserved for memory outside the process. For example, kernel page cache. Page cache is important for Prometheus TSDB query performance. The default is 0.9, which means 90% of the memory limit will be used for Prometheus.

No default scrape port


When enabled, the default ports for HTTP (:80) or HTTPS (:443) will not be added to the address used to scrape a target (the value of the __address_ label), contrary to the default behavior. In addition, if a default HTTP or HTTPS port has already been added either in a static configuration or by a service discovery mechanism and the respective scheme is specified (http or https), that port will be removed.

Native Histograms


When enabled, Prometheus will ingest native histograms (formerly also known as sparse histograms or high-res histograms). Native histograms are still highly experimental. Expect breaking changes to happen (including those rendering the TSDB unreadable).

Native histograms are currently only supported in the traditional Prometheus protobuf exposition format. This feature flag therefore also enables a new (and also experimental) protobuf parser, through which all metrics are ingested (i.e. not only native histograms). Prometheus will try to negotiate the protobuf format first. The instrumented target needs to support the protobuf format, too, and it needs to expose native histograms. The protobuf format allows to expose classic and native histograms side by side. With this feature flag disabled, Prometheus will continue to parse the classic histogram (albeit via the text format). With this flag enabled, Prometheus will still ingest those classic histograms that do not come with a corresponding native histogram. However, if a native histogram is present, Prometheus will ignore the corresponding classic histogram, with the notable exception of exemplars, which are always ingested. To keep the classic histograms as well, enable scrape_classic_histograms in the scrape job.

Note about the format of le and quantile label values:

In certain situations, the protobuf parsing changes the number formatting of the le labels of classic histograms and the quantile labels of summaries. Typically, this happens if the scraped target is instrumented with client_golang provided that promhttp.HandlerOpts.EnableOpenMetrics is set to false. In such a case, integer label values are represented in the text format as such, e.g. quantile="1" or le="2". However, the protobuf parsing changes the representation to float-like (following the OpenMetrics specification), so the examples above become quantile="1.0" and le="2.0" after ingestion into Prometheus, which changes the identity of the metric compared to what was ingested before via the text format.

The effect of this change is that alerts, recording rules and dashboards that directly reference label values as whole numbers such as le="1" will stop working.

Aggregation by the le and quantile labels for vectors that contain the old and new formatting will lead to unexpected results, and range vectors that span the transition between the different formatting will contain additional series. The most common use case for both is the quantile calculation via histogram_quantile, e.g. histogram_quantile(0.95, sum by (le) (rate(histogram_bucket[10m]))). The histogram_quantile function already tries to mitigate the effects to some extent, but there will be inaccuracies, in particular for shorter ranges that cover only a few samples.

Ways to deal with this change either globally or on a per metric basis:

  • Fix references to integer le, quantile label values, but otherwise do nothing and accept that some queries that span the transition time will produce inaccurate or unexpected results. This is the recommended solution, to get consistently normalized label values. Also Prometheus 3.0 is expected to enforce normalization of these label values.
  • Use metric_relabel_config to retain the old labels when scraping targets. This should only be applied to metrics that currently produce such labels.
      - source_labels:
          - quantile
        target_label: quantile
        regex: (\d+)\.0+
      - source_labels:
          - le
          - __name__
        target_label: le
        regex: (\d+)\.0+;.*_bucket

OTLP Receiver


The OTLP receiver allows Prometheus to accept OpenTelemetry metrics writes. Prometheus is best used as a Pull based system, and staleness, up metric, and other Pull enabled features won't work when you push OTLP metrics.

Experimental PromQL functions


Enables PromQL functions that are considered experimental and whose name or semantics could change.

Created Timestamps Zero Injection


Enables ingestion of created timestamp. Created timestamps are injected as 0 valued samples when appropriate. See PromCon talk for details.

Currently Prometheus supports created timestamps only on the traditional Prometheus Protobuf protocol (WIP for other protocols). As a result, when enabling this feature, the Prometheus protobuf scrape protocol will be prioritized (See scrape_config.scrape_protocols settings for more details).

Besides enabling this feature in Prometheus, created timestamps need to be exposed by the application being scraped.

Concurrent evaluation of independent rules


By default, rule groups execute concurrently, but the rules within a group execute sequentially; this is because rules can use the output of a preceding rule as its input. However, if there is no detectable relationship between rules then there is no reason to run them sequentially. When the concurrent-rule-eval feature flag is enabled, rules without any dependency on other rules within a rule group will be evaluated concurrently. This has the potential to improve rule group evaluation latency and resource utilization at the expense of adding more concurrent query load.

The number of concurrent rule evaluations can be configured with --rules.max-concurrent-rule-evals, which is set to 4 by default.

This documentation is open-source. Please help improve it by filing issues or pull requests.