Configuration

A node reads its settings from the defaults, then a TOML file given with -c, then MINK_* environment variables. An unknown setting fails startup.

postgres = "postgres://mink:mink@db:5432/mink"
warehouse = "s3://lake/mink"

[kafka]
bootstrap = ["kafka-1:9092", "kafka-2:9092"]

[storage]
region = "us-east-1"

[schema_registry]
type = "confluent"
url = "http://registry:8081"

Every setting has a variable: MINK_ and its name in capitals, with __ between a table and its keys. Lists take their TOML form.

MINK_POSTGRES=postgres://mink:mink@db:5432/mink
MINK_KAFKA__BOOTSTRAP='["kafka-1:9092", "kafka-2:9092"]'
MINK_KAFKA__SASL__PASSWORD=secret
MINK_SCHEMA_REGISTRY__TYPE=confluent
MINK_SCHEMA_REGISTRY__URL=http://registry:8081

Durations are a number and a unit: ms, s, m or h. Sizes are bytes. Every duration and size must be positive.

Roles and listeners

Setting	Default	Purpose
roles	["catalog", "s3", "indexer", "admin"]	Roles this node runs. See Overview.
catalog_address	127.0.0.1:8181	Where the REST catalog listens
s3_address	127.0.0.1:9100	Where the S3 endpoint listens
admin_address	127.0.0.1:9180	Where the admin API, the console and /metrics listen
admin_remote	false	Allow an admin_address that is not loopback. The admin API has no authentication.

Addresses others reach

Setting	Default	Purpose
catalog_endpoint	http://127.0.0.1:8181	The catalog URL as engines reach it, shown on the console's Connect page
s3_endpoint	http://127.0.0.1:9100	The S3 endpoint URL as engines reach it, handed out by the catalog
advertised_address	http://{s3_address}	The URL other nodes forward data file reads to
advertised_admin_address	http://{admin_address}	The URL other nodes forward admin requests to

Services

Setting	Default	Purpose
postgres	postgres://mink:mink@localhost:5432/mink	Postgres connection URL. TLS is set in the URL: sslmode=verify-full checks the server's certificate against sslrootcert, or the Mozilla CA certificates without it. require encrypts without checking the certificate.
kafka.bootstrap	["localhost:9092"]	Kafka brokers to bootstrap from
kafka.tls	none	TLS to the brokers, verified against the system's CA certificates. {} turns it on.
kafka.tls.ca	none	A PEM file of more CA certificates to trust
kafka.sasl.mechanism	none	PLAIN, SCRAM-SHA-256 or SCRAM-SHA-512, as Kafka names them
kafka.sasl.username, kafka.sasl.password	none	The SASL credentials
warehouse	s3://mink/warehouse	Where Iceberg metadata lives, and the root of every table location
storage.endpoint	none	An S3-compatible store such as RustFS or MinIO, addressed in path style. Without it, AWS S3 in storage.region.
storage.region	us-east-1	The bucket's region
storage.credentials	none	{ access_key_id, secret_access_key }. Without them, AWS's own sources: the AWS_* environment variables, web identity, the ECS task role, EKS Pod Identity, or the instance profile.
schema_registry.type	none	confluent, for Confluent Schema Registry or a registry with its API, or glue, for AWS Glue Schema Registry.
schema_registry.url	none	For confluent: the registry's URL
schema_registry.username, schema_registry.password	none	For confluent: Basic authentication
schema_registry.name	none	For glue: the registry's name. A record's schema must be a version of the schema named after its topic.
schema_registry.region	none	For glue: the registry's region. Without it, the AWS SDK's own sources, such as AWS_REGION. Credentials come from AWS's own sources, as for storage.credentials.
namespace	kafka	The catalog's one namespace

Indexing

Setting	Default	Purpose
history	all	How far backfill goes: all, none, or { since = "<RFC 3339 time>" }
file_bytes	134217728	A range of records becomes files when their key, value and header bytes reach this
file_window	60s	A range ends when its timestamps span this, or its first record was read this long ago
backfill_offsets	100000	Offsets per backfill chunk
retention_check	60s	How often indexers check Kafka's earliest offsets
granularity	hour	Time partitioning for tables added without one: hour, day, month or none
max_partitions	unlimited	The most partition leases one node holds
lease_ttl	30s	Lease lifetime. Leases renew every third of it.

Snapshots and retention

Setting	Default	Purpose
snapshot_interval	10s	How often a writer commits
snapshot_retention	24h	How long snapshots are kept
collection_interval	10m	How often a writer deletes unused metadata objects
collection_grace	1h	How long an unused object waits before deletion
assembly_delay	10m	How long after its end a time bucket is assembled
assembled_bytes	1073741824	The largest assembled file
statistics_interval	1h	How often a writer publishes a statistics file. Must be shorter than snapshot_retention.

Cache

Setting	Default	Purpose
cache_bytes	1073741824	Memory for built data files and metadata objects
cache_dir	none	Directory of the disk tier. Without it, the cache has no disk tier.
cache_disk_bytes	53687091200	Size of the disk tier
cache_admission	reused	Which blocks go to disk: reused, fetched or evicted. See Reads.
max_builds	the number of CPUs	Builds the node runs at once

Console

Setting	Default	Purpose
duckdb_engine	none, jsDelivr	Base URL of the DuckDB-Wasm files the SQL page loads
duckdb_extensions	none, DuckDB's own	DuckDB extension repository for the SQL page. A mirror holds iceberg, httpfs, parquet, avro and json.