Config file reference

This page describes the different configuration options for Polars on-premises. The config file is a standard TOML file with different sections. Any of the configuration can be overridden using environment variables in the following format: PC_CUBLET__section_name__key.

Top-level configuration

The polars-on-premises binary requires a license which path is provided as a configuration option, listed below. The license itself has the following shape:

{ "params": { "expiry": "2026-01-31T23:59:59Z", "name": "Company" }, "signature": "..." }

Key	Type	Description
cluster_id	string	Logical ID for the cluster; workers and scheduler that share this ID will form a single cluster. e.g. prod-eu-1; must be unique among all clusters.
cublet_id	string	Unique ID for this node (aka "cublet") within the cluster, used for addressing and leader selection. e.g. scheduler, worker_0; must be unique per cluster.
license	path	Absolute path to the Polars on-premises license file required to start the process. e.g. /etc/polars/license.json.
memory_limit	integer	Hard memory budget for all components in this cublet; enforced via cgroups when delegated. e.g. 1073741824 (1 GiB), 10737418240 (10 GiB).

Example:

cluster_id = "polars-cluster-dev"
cublet_id = "scheduler"
license = "/etc/polars/license.json"
memory_limit = 1073741824 # 1 GiB

[scheduler] section

For remote Polars queries without a specific output sink, Polars on-premises can automatically add persistent sink. We call these sinks "anonymous results" sinks. Infrastructure-wise, these sinks are backed by S3-compatible storage accessible from all worker nodes and the Python client. The data written to this location is not automatically deleted, so you need to configure a retention policy for this data yourself.

You may configure the credentials using the options listed below; the key names correspond to the storage_options parameter from the scan_parquet() method (e.g. aws_access_key_id, aws_secret_access_key, aws_session_token, aws_region). We currently only support the AWS keys of the storage_options dictionary, but note that you can use any other cloud provider that supports the S3 API, such as MinIO or DigitalOcean Spaces.

Key	Type	Description
enabled	boolean	Whether the scheduler component runs in this process. true for the leader node, false on pure workers.
allow_shared_disk	boolean	Whether workers are allowed to write to a shared/local disk visible to the scheduler. false for fully remote/storage-only setups, true if you have a shared filesystem.
n_workers	integer	Expected number of workers in this cluster; scheduler waits for the latter to be online before running queries. e.g. 4.
anonymous_result_dst.s3.url	string	Destination for results of queries that do not have an explicit sink. Currently only S3-compatible storages are supported. The bucket must be reachable by scheduler, workers, and client. e.g. s3://bucket/path/to/key.
anonymous_result_dst.s3.aws_endpoint_url	string	Storage option configuration, see scan_parquet().
anonymous_result_dst.s3.aws_region	string	Storage option configuration. e.g. eu-east-1
anonymous_result_dst.s3.aws_access_key_id	string	Storage option configuration.
anonymous_result_dst.s3.aws_secret_access_key	string	Storage option configuration.

Example:

[scheduler]
enabled = true
allow_shared_disk = false
n_workers = 4
anonymous_result_dst.s3.url = "s3://bucket/path/to/key"

[worker] section

During distributed query execution, data may be shuffled between workers. A local path can be provided, but shuffles can also be configured to use S3-compatible storage (accessible from all worker nodes). You may configure the credentials using the options listed below; the key names correspond to the storage_options parameter from the scan_parquet() method (e.g. aws_access_key_id, aws_secret_access_key, aws_session_token, aws_region).

Key	Type	Description
enabled	boolean	Whether the worker component runs in this process. true on worker nodes, false on the dedicated scheduler.
worker_ip	string	Public or routable IP address other workers/scheduler use to reach this worker. e.g. 192.168.1.2.
flight_port	integer	Port for shuffle traffic between workers. e.g. 5052.
service_port	integer	Port on which the worker receives task instructions from the scheduler. e.g. 5053.
heartbeat_interval_secs	integer	Interval for worker heartbeats towards the scheduler, used for liveness and load reporting. e.g. 5.
shuffle_location.local.path	path	Local path where shuffle/intermediate data is stored; fast local SSD is recommended. e.g. /mnt/storage/polars/shuffle.
shuffle_location.s3.url	path	Destination for shuffle/intermediate data. e.g. s3://bucket/path/to/key.
shuffle_location.s3.aws_endpoint_url	string	Storage option configuration, see scan_parquet().
shuffle_location.s3.aws_region	string	Storage option configuration. e.g. eu-east-1
shuffle_location.s3.aws_access_key_id	string	Storage option configuration.
shuffle_location.s3.aws_secret_access_key	string	Storage option configuration.

Example:

[worker]
enabled = true
worker_ip = "192.168.1.2"
flight_port = 5052
service_port = 5053
heartbeat_interval_secs = 5
shuffle_location.local.path = "/mnt/storage/polars/shuffle"

[observatory] section

Key	Type	Description
enabled	boolean	Enable sending/receiving profiling data so clients can call result.await_profile(). true on both scheduler and workers if you want profiles on queries; false to disable.
max_metrics_bytes_total	integer	How many bytes all the worker host metrics will consume in total. If a system-wide memory limit is specified then this is added to the share that the scheduler takes. Note that the worker host metrics is not yet available, so this configuration can be set to 0.

Example:

[observatory]
enabled = true
max_metrics_bytes_total = 0

[static_leader] section

Key	Type	Description
leader_key	string	ID of the leader service; should match the scheduler’s cublet_id. Typically scheduler to match your scheduler node.
public_leader_addr	string	Host/IP at which the leader is reachable from this node. e.g. 192.168.1.1.

Example:

[static_leader]
leader_key = "scheduler"
public_leader_addr = "192.168.1.1"