Documentation Grafana Agent Flow mode Reference Components reference pyroscope.scrape

pyroscope.scrape

pyroscope.scrape collects pprof performance profiles for a given set of HTTP targets.

pyroscope.scrape mimcks the scraping behavior of prometheus.scrape. Similarly to how Prometheus scrapes metrics via HTTP, pyroscope.scrape collects profiles via HTTP requests.

Unlike Prometheus, which usually only scrapes one /metrics endpoint per target, pyroscope.scrape may need to scrape multiple endpoints for the same target. This is because different types of profiles are scraped on different endpoints. For example, “mutex” profiles may be scraped on a /debug/pprof/delta_mutex HTTP endpoint, whereas memory consumption may be scraped on a /debug/pprof/allocs HTTP endpoint.

The profile paths, protocol scheme, scrape interval, scrape timeout, query parameters, as well as any other settings can be configured within pyroscope.scrape.

The pyroscope.scrape component regards a scrape as successful if it responded with an HTTP 200 OK status code and returned the body of a valid pprof profile.

If a scrape request fails, the debug UI for pyroscope.scrape will show:

• Detailed information about the failure.
• The time of the last successful scrape.
• The labels last used for scraping.

The scraped performance profiles can be forwarded to components such as pyroscope.write via the forward_to argument.

Multiple pyroscope.scrape components can be specified by giving them different labels.

Usage

river Copy

pyroscope.scrape "LABEL" {
  targets    = TARGET_LIST
  forward_to = RECEIVER_LIST
}

Arguments

pyroscope.scrape starts a new scrape job to scrape all of the input targets. Multiple scrape jobs can be started for a single input target when scraping multiple profile types.

The list of arguments that can be used to configure the block is presented below.

Any omitted arguments take on their default values. If conflicting arguments are being passed (for example, configuring both bearer_token and bearer_token_file), then pyroscope.scrape will fail to start and will report an error.

The following arguments are supported:

At most, one of the following can be provided:

• bearer_token argument.
• bearer_token_file argument.
• basic_auth block.
• authorization block.
• oauth2 block.

job_name argument

job_name defaults to the component’s unique identifier.

For example, the job_name of pyroscope.scrape "local" { ... } will be "pyroscope.scrape.local".

targets argument

The list of targets can be provided statically, dynamically, or a combination of both.

The special __address__ label must always be present and corresponds to the <host>:<port> that is used for the scrape request.

Labels starting with a double underscore (__) are treated as internal, and are removed prior to scraping.

The special label service_name is required and must always be present. If it is not specified, pyroscope.scrape will attempt to infer it from either of the following sources, in this order:

1. __meta_kubernetes_pod_annotation_pyroscope_io_service_name which is a pyroscope.io/service_name pod annotation.
2. __meta_kubernetes_namespace and __meta_kubernetes_pod_container_name
3. __meta_docker_container_name
4. __meta_dockerswarm_container_label_service_name or __meta_dockerswarm_service_name

If service_name is not specified and could not be inferred, then it is set to unspecified.

The following labels are automatically injected to the scraped profiles so that they can be linked to a scrape target:

scrape_interval argument

The scrape_interval typically refers to the frequency with which Grafana Agent Flow collects performance profiles from the monitored targets. It represents the time interval between consecutive scrapes or data collection events. This parameter is important for controlling the trade-off between resource usage and the freshness of the collected data.

If scrape_interval is short:

• Advantages:
  • Fewer profiles may be lost if the application being scraped crashes.
• Disadvantages:
  • Greater consumption of CPU, memory, and network resources during scrapes and remote writes.
  • The backend database (Pyroscope) will consume more storage space.

If scrape_interval is long:

• Advantages:
  • Lower resource consumption.
• Disadvantages:
  • More profiles may be lost if the application being scraped crashes.
  • If the delta argument is set to true, the batch size of each remote write to Pyroscope may be bigger. The Pyroscope database may need to be tuned with higher limits.
  • If the delta argument is set to true, there is a larger risk of reaching the HTTP server timeouts of the application being scraped.

For example, consider this situation:

• pyroscope.scrape is configured with a scrape_interval of "60s".
• The application being scraped is running an HTTP server with a timeout of 30 seconds.
• Any scrape HTTP requests where the delta argument is set to true will fail, because they will attempt to run for 59 seconds.

Blocks

The following blocks are supported inside the definition of pyroscope.scrape:

The > symbol indicates deeper levels of nesting. For example, oauth2 > tls_config refers to a tls_config block defined inside an oauth2 block.

Any omitted blocks take on their default values. For example, if profile.mutex is not specified in the config, the defaults documented in profile.mutex will be used.

basic_auth block

authorization block

oauth2 block

tls_config block

profiling_config block

The profiling_config block configures the profiling settings when scraping targets.

The following arguments are supported:

profile.memory block

The profile.memory block collects profiles on memory consumption.

The following arguments are supported:

For more information about the delta argument, see the delta argument section.

profile.block block

The profile.block block collects profiles on process blocking.

The following arguments are supported:

For more information about the delta argument, see the delta argument section.

profile.goroutine block

The profile.goroutine block collects profiles on the number of goroutines.

The following arguments are supported:

For more information about the delta argument, see the delta argument section.

profile.mutex block

The profile.mutex block collects profiles on mutexes.

The following arguments are supported:

For more information about the delta argument, see the delta argument section.

profile.process_cpu block

The profile.process_cpu block collects profiles on CPU consumption for the process.

The following arguments are supported:

For more information about the delta argument, see the delta argument section.

profile.fgprof block

The profile.fgprof block collects profiles from an fgprof endpoint.

The following arguments are supported:

For more information about the delta argument, see the delta argument section.

profile.godeltaprof_memory block

The profile.godeltaprof_memory block collects profiles from godeltaprof memory endpoint. The delta is computed on the target.

The following arguments are supported:

profile.godeltaprof_mutex block

The profile.godeltaprof_mutex block collects profiles from godeltaprof mutex endpoint. The delta is computed on the target.

The following arguments are supported:

profile.godeltaprof_block block

The profile.godeltaprof_block block collects profiles from godeltaprof block endpoint. The delta is computed on the target.

The following arguments are supported:

profile.custom block

The profile.custom block allows for collecting profiles from custom endpoints. Blocks must be specified with a label:

river Copy

profile.custom "PROFILE_TYPE" {
  enabled = true
  path    = "PROFILE_PATH"
}

Multiple profile.custom blocks can be specified. Labels assigned to profile.custom blocks must be unique across the component.

The following arguments are supported:

When the delta argument is true, a seconds query parameter is automatically added to requests. The seconds used will be equal to scrape_interval - 1.

clustering block

When Grafana Agent Flow is using clustering, and enabled is set to true, then this pyroscope.scrape component instance opts-in to participating in the cluster to distribute scrape load between all cluster nodes.

Clustering causes the set of targets to be locally filtered down to a unique subset per node, where each node is roughly assigned the same number of targets. If the state of the cluster changes, such as a new node joins, then the subset of targets to scrape per node will be recalculated.

When clustering mode is enabled, all Grafana Agents participating in the cluster must use the same configuration file and have access to the same service discovery APIs.

If Grafana Agent Flow is not running in clustered mode, this block is a no-op.

Common configuration

delta argument

When the delta argument is false, the pprof HTTP query will be instantaneous.

When the delta argument is true:

• The pprof HTTP query will run for a certain amount of time.
• A seconds parameter is automatically added to the HTTP request.
• The seconds used will be equal to scrape_interval - 1. For example, if scrape_interval is "15s", seconds will be 14 seconds. If the HTTP endpoint is /debug/pprof/profile, then the HTTP query will become /debug/pprof/profile?seconds=14

Exported fields

pyroscope.scrape does not export any fields that can be referenced by other components.

Component health

pyroscope.scrape is only reported as unhealthy if given an invalid configuration.

Debug information

pyroscope.scrape reports the status of the last scrape for each configured scrape job on the component’s debug endpoint.

Debug metrics

• pyroscope_fanout_latency (histogram): Write latency for sending to direct and indirect components.

Examples

Default endpoints of static targets

The following example sets up a scrape job of a statically configured list of targets - Grafana Agent itself and Pyroscope. The scraped profiles are sent to pyroscope.write which remote writes them to a Pyroscope database.

river Copy

pyroscope.scrape "local" {
  targets = [
    {"__address__" = "localhost:4100", "service_name"="pyroscope"},
    {"__address__" = "localhost:12345", "service_name"="agent"},
  ]

  forward_to = [pyroscope.write.local.receiver]
}

pyroscope.write "local" {
  endpoint {
    url = "http://pyroscope:4100"
  }
}

These endpoints will be scraped every 15 seconds:

Copy

http://localhost:4100/debug/pprof/allocs
http://localhost:4100/debug/pprof/block
http://localhost:4100/debug/pprof/goroutine
http://localhost:4100/debug/pprof/mutex
http://localhost:4100/debug/pprof/profile?seconds=14

http://localhost:12345/debug/pprof/allocs
http://localhost:12345/debug/pprof/block
http://localhost:12345/debug/pprof/goroutine
http://localhost:12345/debug/pprof/mutex
http://localhost:12345/debug/pprof/profile?seconds=14

Note that seconds=14 is added to the /debug/pprof/profile endpoint, because:

• The delta argument of the profile.process_cpu block is true by default.
• scrape_interval is "15s" by default.

Also note that the /debug/fgprof endpoint will not be scraped, because the enabled argument of the profile.fgprof block is false by default.

Default endpoints of dynamic targets

river Copy

discovery.http "dynamic_targets" {
  url = "https://example.com/scrape_targets"
  refresh_interval = "15s"
}

pyroscope.scrape "local" {
  targets = [discovery.http.dynamic_targets.targets]

  forward_to = [pyroscope.write.local.receiver]
}

pyroscope.write "local" {
  endpoint {
    url = "http://pyroscope:4100"
  }
}

Default endpoints of static and dynamic targets

river Copy

discovery.http "dynamic_targets" {
  url = "https://example.com/scrape_targets"
  refresh_interval = "15s"
}

pyroscope.scrape "local" {
  targets = concat([
    {"__address__" = "localhost:4040", "service_name"="pyroscope"},
    {"__address__" = "localhost:12345", "service_name"="agent"},
  ], discovery.http.dynamic_targets.targets)

  forward_to = [pyroscope.write.local.receiver]
}

pyroscope.write "local" {
  endpoint {
    url = "http://pyroscope:4100"
  }
}

Enabling and disabling profiles

river Copy

pyroscope.scrape "local" {
  targets = [
    {"__address__" = "localhost:12345", "service_name"="agent"},
  ]

  profiling_config {
    profile.fgprof {
      enabled = true
    }
    profile.block {
      enabled = false
    }
    profile.mutex {
      enabled = false
    }
  }

  forward_to = [pyroscope.write.local.receiver]
}

These endpoints will be scraped every 15 seconds:

Copy

http://localhost:12345/debug/pprof/allocs
http://localhost:12345/debug/pprof/goroutine
http://localhost:12345/debug/pprof/profile?seconds=14
http://localhost:12345/debug/fgprof?seconds=14

These endpoints will NOT be scraped because they are explicitly disabled:

Copy

http://localhost:12345/debug/pprof/block
http://localhost:12345/debug/pprof/mutex

Compatible components

pyroscope.scrape can accept arguments from the following components:

• Components that export Targets
• Components that export Pyroscope ProfilesReceiver