load_kafka

Loads a byte stream from an Apache Kafka topic.

load_kafka topic:string, [count=int, exit=bool, offset=int|string, options=record,
           aws_iam=record, aws_region=string, commit_batch_size=int, commit_timeout=duration]

Description

The load_kafka operator reads bytes from a Kafka topic.

The implementation uses the official librdkafka from Confluent and supports all configuration options. You can specify them via options parameter as {key: value, ...}.

The operator injects the following default librdkafka configuration values in case no configuration file is present, or when the configuration does not include them:

bootstrap.servers: localhost
client.id: tenzir
group.id: tenzir
enable.auto.commit: false (This option cannot be changed)

`topic: string`

The Kafka topic to use.

`count = int (optional)`

Exit successfully after having consumed count messages.

`exit = bool (optional)`

Exit successfully after having received the last message.

Without this option, the operator waits for new messages after consuming the last one.

`offset = int|string (optional)`

The offset to start consuming from. Possible values are:

"beginning": first offset
"end": last offset
"stored": stored offset
<value>: absolute offset
-<value>: relative offset from end

The default is "stored".

`options = record (optional)`

A record of key-value configuration options for librdkafka, e.g., {"auto.offset.reset" : "earliest", "enable.partition.eof": true}.

The load_kafka operator passes the key-value pairs directly to librdkafka. Consult the list of available configuration options to configure Kafka according to your needs.

We recommend factoring these options into the plugin-specific kafka.yaml so that they are independent of the load_kafka arguments.

`commit_batch_size = int (optional)`

The operator commits offsets after receiving commit_batch_size messages to improve throughput. If you need to ensure exactly-once semantics for your pipeline, set this option to 1 to commit every message individually.

Defaults to 1000.

`commit_timeout = duration (optional)`

A timeout after which the operator commits messages, even if it accepted fewer than commit_batch_size. This helps with long-running, low-volume pipelines.

Defaults to 10s.

`aws_iam = record (optional)`

Configures explicit AWS credentials or IAM role assumption. If not specified, the operator uses the default AWS credential chain.

{
  region: string,        // AWS region for API requests.
  access_key_id: string, // AWS access key ID.
  secret_access_key: string, // AWS secret access key.
  session_token: string, // session token for temporary credentials.
  assume_role: string,   // ARN of IAM role to assume.
  session_name: string,  // session name for role assumption.
  external_id: string,   // external ID for role assumption.
  web_identity: record,  // OIDC web identity token configuration.
}

The access_key_id and secret_access_key must be specified together. If neither is specified, the operator uses the default AWS credential chain:

Environment variables (AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY)
Shared credentials file (~/.aws/credentials)
IAM role for Amazon EC2 or ECS task role
Instance metadata service

Web identity authentication

The web_identity option enables OIDC-based authentication using the AWS AssumeRoleWithWebIdentity API. This lets you authenticate with AWS resources using OpenID Connect tokens from external identity providers like Azure, Google Cloud, or custom OIDC endpoints.

When web_identity is specified, you must also provide assume_role with the ARN of the IAM role configured to trust your identity provider.

The web_identity record accepts the following fields:

{
  token_file: string,       // path to file containing the JWT token.
  token_endpoint: {         // HTTP endpoint configuration.
    url: string,            // endpoint URL to fetch tokens from.
    headers: record,        // HTTP headers for the request.
    path: string,           // JSON path to extract token from response.
  },
  token: string,            // direct token value.
}

Exactly one of token_file, token_endpoint, or token must be specified:

token_file: Path to a file containing the JWT token. This is the standard approach for Kubernetes workload identity (EKS, AKS, GKE) where the platform mounts a token file into the pod.
token_endpoint: Configuration for fetching tokens from an HTTP endpoint. Use this for Azure IMDS or similar metadata services. The nested record contains:
- url (required): The HTTP endpoint URL that returns a token, such as http://169.254.169.254/metadata/identity/oauth2/token?... for Azure IMDS.
- headers: HTTP headers to include in the token request. For Azure IMDS, you typically need {Metadata: "true"}.
- path: JSON path to extract the token from the endpoint response. Defaults to .access_token. Set to null for endpoints that return the token as plain text.
token: Direct token value as a string. Useful for testing or when the token is available from another source.

Credentials are automatically refreshed before expiration, with exponential backoff retry logic for transient failures.

`aws_region = string (optional)`

The AWS region used to construct the MSK authentication URL. Required when connecting to MSK with IAM authentication.

Examples

Read 100 JSON messages from the topic `tenzir`

load_kafka "tenzir", count=100
read_json

Read Zeek Streaming JSON logs starting at the beginning

load_kafka "zeek", offset="beginning"
read_zeek_json

load_kafka

Description

topic: string

count = int (optional)

exit = bool (optional)

offset = int|string (optional)

options = record (optional)

commit_batch_size = int (optional)

commit_timeout = duration (optional)

aws_iam = record (optional)