Prometheus remote write sink

Warnings

High cardinality metric names and labels are discouraged by Prometheus as they can provide performance and reliability problems. You should consider alternative strategies to reduce the cardinality. Vector offers a tag_cardinality_limit transform as a way to protect against this.

Configuration

Example configurations

{
  "sinks": {
    "my_sink_id": {
      "type": "prometheus_remote_write",
      "inputs": [
        "my-source-or-transform-id"
      ],
      "endpoint": "https://localhost:8087/api/v1/write"
    }
  }
}

[sinks.my_sink_id]
type = "prometheus_remote_write"
inputs = [ "my-source-or-transform-id" ]
endpoint = "https://localhost:8087/api/v1/write"

sinks:
  my_sink_id:
    type: prometheus_remote_write
    inputs:
      - my-source-or-transform-id
    endpoint: https://localhost:8087/api/v1/write

{
  "sinks": {
    "my_sink_id": {
      "type": "prometheus_remote_write",
      "inputs": [
        "my-source-or-transform-id"
      ],
      "compression": "snappy",
      "default_namespace": "service",
      "endpoint": "https://localhost:8087/api/v1/write",
      "tenant_id": "my-domain"
    }
  }
}

[sinks.my_sink_id]
type = "prometheus_remote_write"
inputs = [ "my-source-or-transform-id" ]
compression = "snappy"
default_namespace = "service"
endpoint = "https://localhost:8087/api/v1/write"
tenant_id = "my-domain"

sinks:
  my_sink_id:
    type: prometheus_remote_write
    inputs:
      - my-source-or-transform-id
    compression: snappy
    default_namespace: service
    endpoint: https://localhost:8087/api/v1/write
    tenant_id: my-domain

acknowledgements

optional object

Controls how acknowledgements are handled for this sink.

See End-to-end Acknowledgements for more information on how event acknowledgement is handled.

acknowledgements.enabled

optional bool

Whether or not end-to-end acknowledgements are enabled.

When enabled for a sink, any source connected to that sink, where the source supports end-to-end acknowledgements as well, waits for events to be acknowledged by all connected sinks before acknowledging them at the source.

Enabling or disabling acknowledgements at the sink level takes precedence over any global acknowledgements configuration.

auth

optional object

Authentication strategies.

auth.access_key_id

required string literal

The AWS access key ID.

Relevant when: strategy = "aws"

Examples

"AKIAIOSFODNN7EXAMPLE"

auth.assume_role

required string literal

The ARN of an IAM role to assume.

Relevant when: strategy = "aws"

Examples

"arn:aws:iam::123456789098:role/my_role"

auth.credentials_file

required string literal

Path to the credentials file.

Relevant when: strategy = "aws"

Examples

"/my/aws/credentials"

auth.external_id

optional string literal

The optional unique external ID in conjunction with role to assume.

Relevant when: strategy = "aws"

Examples

"randomEXAMPLEidString"

auth.imds

optional object

Configuration for authenticating with AWS through IMDS.

Relevant when: strategy = "aws"

auth.imds.connect_timeout_seconds

optional uint

Connect timeout for IMDS.

default: 1 (seconds)

auth.imds.max_attempts

optional uint

Number of IMDS retries for fetching tokens and metadata.

default: 4

auth.imds.read_timeout_seconds

optional uint

Read timeout for IMDS.

default: 1 (seconds)

auth.load_timeout_secs

optional uint

Timeout for successfully loading any credentials, in seconds.

Relevant when the default credentials chain or assume_role is used.

Relevant when: strategy = "aws"

Examples

auth.password

required string literal

Basic authentication password.

Relevant when: strategy = "basic"

auth.profile

optional string literal

The credentials profile to use.

Used to select AWS credentials from a provided credentials file.

Relevant when: strategy = "aws"

Examples

"develop"

default: default

auth.region

optional string literal

The AWS region to send STS requests to.

If not set, this defaults to the configured region for the service itself.

Relevant when: strategy = "aws"

Examples

"us-west-2"

auth.secret_access_key

required string literal

The AWS secret access key.

Relevant when: strategy = "aws"

Examples

"wJalrXUtnFEMI/K7MDENG/bPxRfiCYEXAMPLEKEY"

auth.strategy

required string literal enum

The authentication strategy to use.

Enum options

Option Description

aws Amazon Prometheus Service-specific authentication.

basic HTTP Basic Authentication.

Option	Description
`aws`	Amazon Prometheus Service-specific authentication.
`basic`	HTTP Basic Authentication.
`bearer`	Bearer authentication. A bearer token (OAuth2, JWT, etc) is passed as-is.

bearer

Bearer authentication.

A bearer token (OAuth2, JWT, etc) is passed as-is.

Examples

"aws"

"basic"

"bearer"

auth.token

required string literal

The bearer token to send.

Relevant when: strategy = "bearer"

auth.user

required string literal

Basic authentication username.

Relevant when: strategy = "basic"

aws

optional object

Configuration of the region/endpoint to use when interacting with an AWS service.

aws.endpoint

optional string literal

Custom endpoint for use with AWS-compatible services.

Examples

"http://127.0.0.0:5000/path/to/service"

aws.region

optional string literal

The AWS region of the target service.

Examples

"us-east-1"

batch

optional object

The batch config for remote write.

batch.aggregate

optional bool

Whether or not to aggregate metrics within a batch.

default: true

batch.max_bytes

optional uint

The maximum size of a batch that is processed by a sink.

This is based on the uncompressed size of the batched events, before they are serialized/compressed.

batch.max_events

optional uint

The maximum size of a batch before it is flushed.

default: 1000 (events)

batch.timeout_secs

optional float

The maximum age of a batch before it is flushed.

default: 1 (seconds)

buckets

optional [float]

Default buckets to use for aggregating distribution metrics into histograms.

Array float

default: [0.005 0.01 0.025 0.05 0.1 0.25 0.5 1 2.5 5 10]

buffer

optional object

Configures the buffering behavior for this sink.

More information about the individual buffer types, and buffer behavior, can be found in the Buffering Model section.

buffer.max_events

optional uint

The maximum number of events allowed in the buffer.

Relevant when: type = "memory"

default: 500

buffer.max_size

required uint

The maximum size of the buffer on disk.

Must be at least ~256 megabytes (268435488 bytes).

Relevant when: type = "disk"

buffer.type

optional string literal enum

The type of buffer to use.

Enum options

Option Description

Option	Description
`disk`	Events are buffered on disk. This is less performant, but more durable. Data that has been synchronized to disk will not be lost if Vector is restarted forcefully or crashes. Data is synchronized to disk every 500ms.
`memory`	Events are buffered in memory. This is more performant, but less durable. Data will be lost if Vector is restarted forcefully or crashes.

disk

Events are buffered on disk.

This is less performant, but more durable. Data that has been synchronized to disk will not be lost if Vector is restarted forcefully or crashes.

Data is synchronized to disk every 500ms.

memory

Events are buffered in memory.

This is more performant, but less durable. Data will be lost if Vector is restarted forcefully or crashes.

default: memory

buffer.when_full

optional string literal enum

Event handling behavior when a buffer is full.

Enum options

Option Description

Option	Description
`block`	Wait for free space in the buffer. This applies backpressure up the topology, signalling that sources should slow down the acceptance/consumption of events. This means that while no data is lost, data will pile up at the edge.
`drop_newest`	Drops the event instead of waiting for free space in buffer. The event will be intentionally dropped. This mode is typically used when performance is the highest priority, and it is preferable to temporarily lose events rather than cause a slowdown in the acceptance/consumption of events.

block

Wait for free space in the buffer.

This applies backpressure up the topology, signalling that sources should slow down the acceptance/consumption of events. This means that while no data is lost, data will pile up at the edge.

drop_newest

Drops the event instead of waiting for free space in buffer.

The event will be intentionally dropped. This mode is typically used when performance is the highest priority, and it is preferable to temporarily lose events rather than cause a slowdown in the acceptance/consumption of events.

default: block

compression

optional string literal enum

Compression configuration.

All compression algorithms use the default compression level unless otherwise specified.

Enum options string literal

Option	Description
`gzip`	Gzip compression.
`none`	No compression.
`snappy`	Snappy compression.
`zlib`	Zlib compression.
`zstd`	Zstandard compression.

default: snappy

default_namespace

optional string literal

The default namespace for any metrics sent.

This namespace is only used if a metric has no existing namespace. When a namespace is present, it is used as a prefix to the metric name, and separated with an underscore (_).

It should follow the Prometheus naming conventions.

Examples

"service"

endpoint

required string literal

The endpoint to send data to.

The endpoint should include the scheme and the path to write to.

Examples

"https://localhost:8087/api/v1/write"

healthcheck

optional object

Healthcheck configuration.

healthcheck.enabled

optional bool

Whether or not to check the health of the sink when Vector starts up.

default: true

inputs

required [string]

A list of upstream source or transform IDs.

Wildcards (*) are supported.

See configuration for more info.

Array string literal

Examples

[
  "my-source-or-transform-id",
  "prefix-*"
]

proxy

optional object

Proxy configuration.

Configure to proxy traffic through an HTTP(S) proxy when making external requests.

Similar to common proxy configuration convention, you can set different proxies to use based on the type of traffic being proxied. You can also set specific hosts that should not be proxied.

proxy.enabled

optional bool

Enables proxying support.

default: true

proxy.http

optional string literal

Proxy endpoint to use when proxying HTTP traffic.

Must be a valid URI string.

Examples

"http://foo.bar:3128"

proxy.https

optional string literal

Proxy endpoint to use when proxying HTTPS traffic.

Must be a valid URI string.

Examples

"http://foo.bar:3128"

proxy.no_proxy

optional [string]

A list of hosts to avoid proxying.

Multiple patterns are allowed:

Pattern	Example match
Domain names	`example.com` matches requests to `example.com`
Wildcard domains	`.example.com` matches requests to `example.com` and its subdomains
IP addresses	`127.0.0.1` matches requests to `127.0.0.1`
CIDR blocks	`192.168.0.0/16` matches requests to any IP addresses in this range
Splat	`*` matches all hosts

quantiles

optional [float]

Quantiles to use for aggregating distribution metrics into a summary.

Array float

default: [0.5 0.75 0.9 0.95 0.99]

request

optional object

Middleware settings for outbound requests.

Various settings can be configured, such as concurrency and rate limits, timeouts, retry behavior, etc.

Note that the retry backoff policy follows the Fibonacci sequence.

request.adaptive_concurrency

optional object

Configuration of adaptive concurrency parameters.

These parameters typically do not require changes from the default, and incorrect values can lead to meta-stable or unstable performance and sink behavior. Proceed with caution.

request.adaptive_concurrency.decrease_ratio

optional float

The fraction of the current value to set the new concurrency limit when decreasing the limit.

Valid values are greater than 0 and less than 1. Smaller values cause the algorithm to scale back rapidly when latency increases.

Note that the new limit is rounded down after applying this ratio.

default: 0.9

request.adaptive_concurrency.ewma_alpha

optional float

The weighting of new measurements compared to older measurements.

Valid values are greater than 0 and less than 1.

ARC uses an exponentially weighted moving average (EWMA) of past RTT measurements as a reference to compare with the current RTT. Smaller values cause this reference to adjust more slowly, which may be useful if a service has unusually high response variability.

default: 0.4

request.adaptive_concurrency.initial_concurrency

optional uint

The initial concurrency limit to use. If not specified, the initial limit will be 1 (no concurrency).

It is recommended to set this value to your service’s average limit if you’re seeing that it takes a long time to ramp up adaptive concurrency after a restart. You can find this value by looking at the adaptive_concurrency_limit metric.

default: 1

request.adaptive_concurrency.max_concurrency_limit

optional uint

The maximum concurrency limit.

The adaptive request concurrency limit will not go above this bound. This is put in place as a safeguard.

default: 200

request.adaptive_concurrency.rtt_deviation_scale

optional float

Scale of RTT deviations which are not considered anomalous.

Valid values are greater than or equal to 0, and we expect reasonable values to range from 1.0 to 3.0.

When calculating the past RTT average, we also compute a secondary “deviation” value that indicates how variable those values are. We use that deviation when comparing the past RTT average to the current measurements, so we can ignore increases in RTT that are within an expected range. This factor is used to scale up the deviation to an appropriate range. Larger values cause the algorithm to ignore larger increases in the RTT.

default: 2.5

request.concurrency

optional string literal enum uint

Configuration for outbound request concurrency.

This can be set either to one of the below enum values or to a positive integer, which denotes a fixed concurrency limit.

Enum options

Option Description

adaptive Concurrency will be managed by Vector’s Adaptive Request Concurrency feature.

Option	Description
`adaptive`	Concurrency will be managed by Vector’s Adaptive Request Concurrency feature.
`none`	A fixed concurrency of 1. Only one request can be outstanding at any given time.

none

A fixed concurrency of 1.

Only one request can be outstanding at any given time.

default: adaptive

request.rate_limit_duration_secs

optional uint

The time window used for the rate_limit_num option.

default: 1 (seconds)

request.rate_limit_num

optional uint

The maximum number of requests allowed within the rate_limit_duration_secs time window.

default: 9.223372036854776e+18 (requests)

request.retry_attempts

optional uint

The maximum number of retries to make for failed requests.

default: 9.223372036854776e+18 (retries)

request.retry_initial_backoff_secs

optional uint

The amount of time to wait before attempting the first retry for a failed request.

After the first retry has failed, the fibonacci sequence is used to select future backoffs.

default: 1 (seconds)

request.retry_jitter_mode

optional string literal enum

The jitter mode to use for retry backoff behavior.

Enum options

Option Description

Option	Description
`Full`	Full jitter. The random delay is anywhere from 0 up to the maximum current delay calculated by the backoff strategy. Incorporating full jitter into your backoff strategy can greatly reduce the likelihood of creating accidental denial of service (DoS) conditions against your own systems when many clients are recovering from a failure state.
`None`	No jitter.

Full

Full jitter.

The random delay is anywhere from 0 up to the maximum current delay calculated by the backoff strategy.

Incorporating full jitter into your backoff strategy can greatly reduce the likelihood of creating accidental denial of service (DoS) conditions against your own systems when many clients are recovering from a failure state.

None No jitter.

default: Full

request.retry_max_duration_secs

optional uint

The maximum amount of time to wait between retries.

default: 30 (seconds)

request.timeout_secs

optional uint

The time a request can take before being aborted.

Datadog highly recommends that you do not lower this value below the service’s internal timeout, as this could create orphaned requests, pile on retries, and result in duplicate data downstream.

default: 60 (seconds)

tenant_id

optional string template

The tenant ID to send.

If set, a header named X-Scope-OrgID is added to outgoing requests with the value of this setting.

This may be used by Cortex or other remote services to identify the tenant making the request.

Note: This parameter supports Vector's template syntax, which enables you to use dynamic per-event values.

Examples

"my-domain"

tls

optional object

TLS configuration.

tls.alpn_protocols

optional [string]

Sets the list of supported ALPN protocols.

Declare the supported ALPN protocols, which are used during negotiation with peer. They are prioritized in the order that they are defined.

tls.ca_file

optional string literal

Absolute path to an additional CA certificate file.

The certificate must be in the DER or PEM (X.509) format. Additionally, the certificate can be provided as an inline string in PEM format.

Examples

"/path/to/certificate_authority.crt"

tls.crt_file

optional string literal

Absolute path to a certificate file used to identify this server.

The certificate must be in DER, PEM (X.509), or PKCS#12 format. Additionally, the certificate can be provided as an inline string in PEM format.

If this is set, and is not a PKCS#12 archive, key_file must also be set.

Examples

"/path/to/host_certificate.crt"

tls.key_file

optional string literal

Absolute path to a private key file used to identify this server.

The key must be in DER or PEM (PKCS#8) format. Additionally, the key can be provided as an inline string in PEM format.

Examples

"/path/to/host_certificate.key"

tls.key_pass

optional string literal

Passphrase used to unlock the encrypted key file.

This has no effect unless key_file is set.

Examples

"${KEY_PASS_ENV_VAR}"

"PassWord1"

tls.server_name

optional string literal

Server name to use when using Server Name Indication (SNI).

Only relevant for outgoing connections.

Examples

"www.example.com"

tls.verify_certificate

optional bool

Enables certificate verification. For components that create a server, this requires that the client connections have a valid client certificate. For components that initiate requests, this validates that the upstream has a valid certificate.

If enabled, certificates must not be expired and must be issued by a trusted issuer. This verification operates in a hierarchical manner, checking that the leaf certificate (the certificate presented by the client/server) is not only valid, but that the issuer of that certificate is also valid, and so on until the verification process reaches a root certificate.

Do NOT set this to false unless you understand the risks of not verifying the validity of certificates.

tls.verify_hostname

optional bool

Enables hostname verification.

If enabled, the hostname used to connect to the remote host must be present in the TLS certificate presented by the remote host, either as the Common Name or as an entry in the Subject Alternative Name extension.

Only relevant for outgoing connections.

Do NOT set this to false unless you understand the risks of not verifying the remote hostname.

Telemetry

intentional

True if the events were discarded intentionally, like a filter transform, or false if due to an error.

pid optional

The process ID of the Vector instance.

component_errors_total

counter

The total number of errors encountered by this component.

component_id

The Vector component ID.

component_kind

The Vector component kind.

component_type

The Vector component type.

error_type

The type of the error

host optional

The hostname of the system Vector is running on.

pid optional

The process ID of the Vector instance.

stage

The stage within the component at which the error occurred.

component_received_event_bytes_total

counter

The number of event bytes accepted by this component either from tagged origins like file and uri, or cumulatively from other origins.

component_id

The Vector component ID.

component_kind

The Vector component kind.

component_type

The Vector component type.

container_name optional

The name of the container from which the data originated.

file optional

The file from which the data originated.

host optional

The hostname of the system Vector is running on.

mode optional

The connection mode used by the component.

peer_addr optional

The IP from which the data originated.

peer_path optional

The pathname from which the data originated.

pid optional

The process ID of the Vector instance.

pod_name optional

The name of the pod from which the data originated.

uri optional

The sanitized URI from which the data originated.

component_received_events_count

histogram

A histogram of the number of events passed in each internal batch in Vector’s internal topology.

Note that this is separate than sink-level batching. It is mostly useful for low level debugging performance issues in Vector due to small internal batches.

component_id

The Vector component ID.

component_kind

The Vector component kind.

component_type

The Vector component type.

container_name optional

The name of the container from which the data originated.

file optional

The file from which the data originated.

host optional

The hostname of the system Vector is running on.

mode optional

The connection mode used by the component.

peer_addr optional

The IP from which the data originated.

peer_path optional

The pathname from which the data originated.

pid optional

The process ID of the Vector instance.

pod_name optional

The name of the pod from which the data originated.

uri optional

The sanitized URI from which the data originated.

component_received_events_total

counter

The number of events accepted by this component either from tagged origins like file and uri, or cumulatively from other origins.

component_id

The Vector component ID.

component_kind

The Vector component kind.

component_type

The Vector component type.

container_name optional

The name of the container from which the data originated.

file optional

The file from which the data originated.

host optional

The hostname of the system Vector is running on.

mode optional

The connection mode used by the component.

peer_addr optional

The IP from which the data originated.

peer_path optional

The pathname from which the data originated.

pid optional

The process ID of the Vector instance.

pod_name optional

The name of the pod from which the data originated.

uri optional

The sanitized URI from which the data originated.

component_sent_bytes_total

counter

The number of raw bytes sent by this component to destination sinks.

component_id

The Vector component ID.

component_kind

The Vector component kind.

component_type

The Vector component type.

endpoint optional

The endpoint to which the bytes were sent. For HTTP, this will be the host and path only, excluding the query string.

file optional

The absolute path of the destination file.

host optional

The hostname of the system Vector is running on.

pid optional

The process ID of the Vector instance.

protocol

The protocol used to send the bytes.

region optional

The AWS region name to which the bytes were sent. In some configurations, this may be a literal hostname.

component_sent_event_bytes_total

counter

The total number of event bytes emitted by this component.

component_id

The Vector component ID.

component_kind

The Vector component kind.

component_type

The Vector component type.

host optional

The hostname of the system Vector is running on.

output optional

The specific output of the component.

pid optional

The process ID of the Vector instance.

component_sent_events_total

counter

The total number of events emitted by this component.

component_id

The Vector component ID.

component_kind

The Vector component kind.

component_type

The Vector component type.

host optional

The hostname of the system Vector is running on.

output optional

The specific output of the component.

pid optional

The process ID of the Vector instance.

utilization

gauge

A ratio from 0 to 1 of the load on a component. A value of 0 would indicate a completely idle component that is simply waiting for input. A value of 1 would indicate a that is never idle. This value is updated every 5 seconds.

component_id

The Vector component ID.

component_kind

The Vector component kind.

component_type

The Vector component type.

host optional

The hostname of the system Vector is running on.

pid optional

The process ID of the Vector instance.

How it works

Buffers and batches

This component buffers & batches data as shown in the diagram above. You’ll notice that Vector treats these concepts differently, instead of treating them as global concepts, Vector treats them as sink specific concepts. This isolates sinks, ensuring services disruptions are contained and delivery guarantees are honored.

Batches are flushed when 1 of 2 conditions are met:

The batch age meets or exceeds the configured timeout_secs.
The batch size meets or exceeds the configured max_bytes or max_events.

Buffers are controlled via the buffer.* options.

Compression schemes

Officially according to the Prometheus Remote-Write specification, the only supported compression scheme is Snappy. However, there are a number of other implementations that do support other schemes. Thus Vector also supports using Gzip and Zstd.

Duplicate tag names

Multiple tags with the same name are invalid within Prometheus and Prometheus will reject a metric with duplicate tag names. When sending a tag with multiple values for each name, Vector will only send the last value specified.

Health checks

Health checks ensure that the downstream service is accessible and ready to accept data. This check is performed upon sink initialization. If the health check fails an error will be logged and Vector will proceed to start.

Require health checks

If you’d like to exit immediately upon a health check failure, you can pass the --require-healthy flag:

vector --config /etc/vector/vector.yaml --require-healthy

Disable health checks

If you’d like to disable health checks for this sink you can set the healthcheck option to false.

Rate limits & adaptive concurrency

Adaptive Request Concurrency (ARC)

Adaptive Request Concurrency is a feature of Vector that does away with static concurrency limits and automatically optimizes HTTP concurrency based on downstream service responses. The underlying mechanism is a feedback loop inspired by TCP congestion control algorithms. Checkout the announcement blog post,

We highly recommend enabling this feature as it improves performance and reliability of Vector and the systems it communicates with. As such, we have made it the default, and no further configuration is required.

Static concurrency

If Adaptive Request Concurrency is not for you, you can manually set static concurrency limits by specifying an integer for request.concurrency:

sinks:
	my-sink:
		request:
			concurrency: 10

Rate limits

In addition to limiting request concurrency, you can also limit the overall request throughput via the request.rate_limit_duration_secs and request.rate_limit_num options.

sinks:
	my-sink:
		request:
			rate_limit_duration_secs: 1
			rate_limit_num: 10

These will apply to both adaptive and fixed request.concurrency values.

Retry policy

Vector will retry failed requests (status == 429, >= 500, and != 501). Other responses will not be retried. You can control the number of retry attempts and backoff rate with the request.retry_attempts and request.retry_backoff_secs options.

State

This component is stateful, meaning its behavior changes based on previous inputs (events). State is not preserved across restarts, therefore state-dependent behavior will reset between restarts and depend on the inputs (events) received since the most recent restart.

Transport Layer Security (TLS)

Vector uses OpenSSL for TLS protocols due to OpenSSL’s maturity. You can enable and adjust TLS behavior via the tls.* options and/or via an OpenSSL configuration file. The file location defaults to /usr/local/ssl/openssl.cnf or can be specified with the OPENSSL_CONF environment variable.