odd-collector (generic)

Generic ODD Collector — 41 pull adapters for databases, data warehouses, BI tools, streams, and MLOps platforms.

Status: Stable. Released as a tagged Docker image; the underlying SDK is the same one all odd-collector-* collectors share.

odd-collector is the general-purpose pull collector. It bundles 41 adapters covering relational databases, data warehouses, NoSQL stores, message brokers, BI tools, MLOps platforms, and a few catalog / orchestration sources. One container instance can host any combination of those adapters as plugins, including multiple plugins of the same type pointing at different sources.

For the broader pull-vs-push picture and the shared collector configuration schema, start at the Integrations hub. For deployment-side detail (build, Docker, env vars), see Build and run ODD Collectors.

Supported adapters

The 41 adapters registered in odd_collector/domain/plugin.py (PLUGIN_FACTORY). Every adapter has per-field documentation below — three (postgresql, snowflake, kafka) get longer deep-dive spotlights with deployment guidance and feature notes; the remaining 38 are catalogued in the per-adapter configuration reference section.

Type literal	Source system	Spotlighted below
airbyte	Airbyte
cassandra	Apache Cassandra
ckan	CKAN
clickhouse	ClickHouse
cockroachdb	CockroachDB
couchbase	Couchbase
cubejs	Cube.js
databricks	Databricks (Unity Catalog)
dbt	dbt Cloud (catalog import)
druid	Apache Druid
duckdb	DuckDB
elasticsearch	Elasticsearch
feast	Feast feature store
fivetran	Fivetran
hive	Apache Hive
kafka	Apache Kafka	✓
kubeflow	Kubeflow Pipelines
metabase	Metabase
mlflow	MLflow
mode	Mode Analytics
mongodb	MongoDB
mssql	Microsoft SQL Server
mysql	MySQL / MariaDB
neo4j	Neo4j
odbc	Generic ODBC source
odd_adapter	Another ODD Platform (federated)
opensearch	OpenSearch
oracle	Oracle Database
postgresql	PostgreSQL (incl. pgvector)	✓
presto	Presto
redash	Redash
redshift	Amazon Redshift
scylladb	ScyllaDB
singlestore	SingleStore
snowflake	Snowflake	✓
sqlite	SQLite
superset	Apache Superset
tableau	Tableau
tarantool	Tarantool
trino	Trino
vertica	Vertica

The canonical YAML for each adapter lives at odd-collectors/odd-collector/config_examples/ — one file per adapter, named after the type literal. The Pydantic models that define the accepted fields live at odd-collectors/odd-collector/odd_collector/domain/plugin.py; read those when an example field is unclear.

The repo's top-level README's "Implemented adapters" table lags behind the code — databricks, couchbase, opensearch, and oracle are present in PLUGIN_FACTORY but missing from the README table at the time of writing. Use the type literal table above (or PLUGIN_FACTORY in plugin.py) as the authoritative inventory.

Installation

docker pull ghcr.io/opendatadiscovery/odd-collector:latest

Mount a collector_config.yaml at /app/collector_config.yaml inside the container. A reference Compose snippet is in the generic collector README and a from-source build flow is in Build and run ODD Collectors.

Minimal config

The smallest collector_config.yaml that runs the collector:

platform_host_url: http://localhost:8080
token: <COLLECTOR_TOKEN>          # see Token and datasource registration on the hub
default_pulling_interval: 10      # minutes; omit to run once and exit
plugins:
  - type: postgresql
    name: warehouse_main
    host: pg.internal
    port: 5432
    database: warehouse
    user: odd_reader
    password: !ENV ${PG_PASSWORD}

The shared top-level fields (platform_host_url, token, default_pulling_interval, plugins, plus the optional connection_timeout_seconds / chunk_size / misfire_grace_time / max_instances / verify_ssl) are documented once at Build and run ODD Collectors → Full configuration reference. Only the plugins[*] shape varies per adapter — the rest of this page covers that.

Multiple plugins in one container

plugins is a list — add as many entries as you need, mixing types freely. Two plugins of the same type (e.g. several PostgreSQL databases on different hosts) is the common pattern:

plugins:
  - type: postgresql
    name: warehouse_eu
    host: pg-eu.internal
    port: 5432
    database: warehouse
    user: odd_reader
    password: !ENV ${PG_EU_PASSWORD}
  - type: postgresql
    name: warehouse_us
    host: pg-us.internal
    port: 5432
    database: warehouse
    user: odd_reader
    password: !ENV ${PG_US_PASSWORD}
  - type: snowflake
    name: dwh_snowflake
    account: ab12345.eu-central-1
    warehouse: COMPUTE_WH
    database: PROD
    user: ODD_READER
    password: !ENV ${SNOWFLAKE_PASSWORD}

Each plugin's name must be unique within the file — the collector uses it to log per-plugin progress and to wire each plugin to its own scheduled job. The default_pulling_interval applies to every plugin uniformly; per-plugin overrides are not supported.

Spotlight: PostgreSQL (type: postgresql)

Pulls schemas, tables, columns, foreign-key relationships, and (with pgvector installed in the source) vector indexes. PostgreSQL tables containing at least one vector-typed column are classified as the Vector Store dataset type — see Vector Store metadata for the user-facing classification, the dedicated icon in the catalog, and the Vector column data type rendering on the Structure tab.

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
host	string	yes	—	PostgreSQL server hostname.
port	integer	no	5432	TCP port.
database	string	yes	—	Database to scan; one plugin = one database.
user	string	yes	—	Login. The user needs read on the system catalogs you want indexed.
password	string (Secret)	yes	empty	Password. Use !ENV ${VAR} to source from an environment variable.
schemas_filter.include	list of regex	no	[".*"]	Schemas to include.
schemas_filter.exclude	list of regex	no	[]	Schemas to drop after include matches.

Source: PostgreSQLPlugin in odd_collector/domain/plugin.py; reference YAML at config_examples/postgresql.yaml.

plugins:
  - type: postgresql
    name: warehouse_main
    host: pg.internal
    port: 5432
    database: warehouse
    user: odd_reader
    password: !ENV ${PG_PASSWORD}
    schemas_filter:
      include: ["public", "analytics_.*"]
      exclude: ["analytics_archive_.*"]

The PostgreSQL adapter extracts foreign-key relationships and emits them as ENTITY_RELATIONSHIP entities — these render as ERD edges on the dataset detail page in the catalog. Cross-schema foreign keys are supported.

Spotlight: Snowflake (type: snowflake)

Pulls databases, schemas, tables, views, columns, and foreign-key relationships.

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
account	string	yes	—	Snowflake account identifier (e.g. ab12345.eu-central-1). The adapter derives the host as {ACCOUNT}.snowflakecomputing.com.
warehouse	string	yes	—	Compute warehouse used for the catalog query.
database	string	yes	—	Database to scan.
user	string	yes	—	Snowflake login.
password	string (Secret)	yes	—	Password.
schemas_filter.include	list of regex	no	[".*"]	Schemas to include.
schemas_filter.exclude	list of regex	no	[]	Schemas to drop after include matches.

Source: SnowflakePlugin in odd_collector/domain/plugin.py; reference YAML at config_examples/snowflake.yaml.

plugins:
  - type: snowflake
    name: dwh_snowflake
    account: ab12345.eu-central-1
    warehouse: COMPUTE_WH
    database: PROD
    user: ODD_READER
    password: !ENV ${SNOWFLAKE_PASSWORD}
    schemas_filter:
      include: [".*"]
      exclude: ["TEMP_.*", "SCRATCH"]

Like PostgreSQL, the Snowflake adapter extracts foreign-key constraints and emits ENTITY_RELATIONSHIP entities.

Spotlight: Kafka (type: kafka)

Pulls Kafka topics and (when a Confluent-compatible Schema Registry is reachable) the registered schemas.

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
host	string	yes	—	Bootstrap broker host.
port	integer	yes	—	Bootstrap broker port.
broker_conf	dict	yes	—	Passed to confluent_kafka.AdminClient — e.g. SASL credentials, SSL settings.
schema_registry_conf	dict	no	{}	Passed to the Schema Registry client — e.g. URL, basic auth. When empty, the adapter does not query the registry.

Source: KafkaPlugin in odd_collector/domain/plugin.py; reference YAML at config_examples/kafka.yaml.

plugins:
  - type: kafka
    name: events_kafka
    host: kafka.internal
    port: 9092
    broker_conf:
      security.protocol: SASL_SSL
      sasl.mechanism: PLAIN
      sasl.username: !ENV ${KAFKA_USER}
      sasl.password: !ENV ${KAFKA_PASSWORD}
    schema_registry_conf:
      url: https://schema-registry.internal:8081
      basic.auth.user.info: !ENV ${SR_USER_INFO}

Per-adapter configuration reference

The three spotlights above cover the deployment-shape questions; this section enumerates the per-field config schema for the remaining 38 adapters. Field names, types, and defaults are sourced from the Pydantic plugin classes in odd_collector/domain/plugin.py; each adapter links to its config_examples/{type}.yaml reference YAML where one exists. Two adapters — mode and opensearch — have no upstream config example; their tables come from the Pydantic model alone and the per-section note flags the gap.

Common shapes used across the families below:

• BasePlugin — every plugin carries name (required, operator-chosen, unique within the file). The optional metadata fields description and namespace are accepted by every plugin and omitted from the per-adapter tables to save space.

• DatabasePlugin base — adds host: str (required), port: str (required, often overridden to int by subclasses), database: str | null (optional in the base; many subclasses redeclare it as required), user: str (required), password: str (required, redeclared by most subclasses as SecretStr with an empty default).

• WithHost — adds only host: str. WithPort — adds only port: str. Both are mixed in by adapters that don't fit the full DatabasePlugin shape.

Each table below repeats every field the adapter accepts so that an entry is self-contained — operators don't need to chase the inheritance chain in plugin.py.

Relational databases

Microsoft SQL Server (type: mssql)

Pulls schemas, tables, views, and columns from a Microsoft SQL Server / Azure SQL Server source via the SQL catalog views.

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
host	string	yes	—	SQL Server host.
port	integer	yes	—	TCP port (typical: 1433).
database	string	yes	—	Database to scan; one plugin = one database.
user	string	yes	—	SQL login.
password	string (Secret)	no	empty	Password. Use !ENV ${VAR} to source from an environment variable.

Source: MSSQLPlugin in odd_collector/domain/plugin.py; reference YAML at config_examples/mssql.yaml.

plugins:
  - type: mssql
    name: prod_sqlserver
    host: sqlserver.internal
    port: 1433
    database: warehouse
    user: odd_reader
    password: !ENV ${MSSQL_PASSWORD}

MySQL / MariaDB (type: mysql)

Pulls schemas, tables, views, and columns. Compatible with MariaDB.

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
host	string	yes	—	MySQL server hostname.
port	integer	yes	—	TCP port (typical: 3306).
database	string	yes	—	Database to scan.
user	string	yes	—	Login.
password	string (Secret)	no	empty	Password.
ssl_disabled	boolean	no	false	When true, disables TLS to the server — typically used only for local development against an unencrypted instance.

Source: MySQLPlugin in odd_collector/domain/plugin.py; reference YAML at config_examples/mysql.yaml.

plugins:
  - type: mysql
    name: legacy_billing
    host: mysql.internal
    port: 3306
    database: billing
    user: odd_reader
    password: !ENV ${MYSQL_PASSWORD}

ClickHouse (type: clickhouse)

Pulls databases, tables, and columns from a ClickHouse cluster.

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
host	string	yes	—	ClickHouse server hostname.
port	integer or null	yes	—	HTTP (8123) or native (9000) port. The Pydantic model accepts null, but every reference example provides an explicit value.
database	string or null	no	—	Database to scan. When unset, the connection's default database is used.
user	string	yes	—	Login.
password	string (Secret)	yes	—	Password.
secure	boolean	no	false	Toggles TLS on the connection. Set to true for ClickHouse Cloud or any TLS-fronted deployment.
verify	boolean	no	true	Whether to verify the server certificate when secure: true. Set to false only for self-signed certs on local clusters.
server_hostname	string or null	no	null	Optional hostname for SNI / certificate validation; defaults to the value of host.
query_limit	integer or null	no	0	Optional row cap applied to internal catalog queries. 0 means no limit.

Source: ClickhousePlugin in odd_collector/domain/plugin.py; reference YAML at config_examples/clickhouse.yaml.

plugins:
  - type: clickhouse
    name: clickhouse_main
    host: clickhouse.internal
    port: 8123
    database: default
    user: default
    password: !ENV ${CLICKHOUSE_PASSWORD}
    secure: false

Amazon Redshift (type: redshift)

Pulls schemas, tables, views, and columns from an Amazon Redshift cluster.

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
host	string	yes	—	Redshift cluster endpoint ({cluster}.{region}.redshift.amazonaws.com).
port	string	yes	—	TCP port as a string (typical: "5439").
database	string or null	no	—	Database to scan.
user	string	yes	—	Login.
password	string (Secret)	yes	—	Password.
schemas	list of string or null	no	null	Allowlist of schema names. When omitted, every non-system schema is ingested. Literal name list, not a regex filter — different from the schemas_filter available on postgresql and snowflake.
connection_timeout	integer or null	no	10	Connection timeout in seconds.

Source: RedshiftPlugin in odd_collector/domain/plugin.py; reference YAML at config_examples/redshift.yaml.

plugins:
  - type: redshift
    name: warehouse_redshift
    host: my-cluster.abc123.us-east-1.redshift.amazonaws.com
    port: "5439"
    database: warehouse
    user: odd_reader
    password: !ENV ${REDSHIFT_PASSWORD}
    schemas: ["public", "analytics"]
    connection_timeout: 10

CockroachDB (type: cockroachdb)

Pulls schemas, tables, columns, and foreign-key relationships. Inherits from the PostgreSQL plugin — same field shape plus the same schemas_filter regex behavior; ERD edges are emitted for cross-schema foreign keys exactly as on PostgreSQL.

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
host	string	yes	—	CockroachDB SQL endpoint.
port	integer	no	5432	TCP port. CockroachDB's typical SQL port is 26257; the model default is the PostgreSQL port inherited from PostgreSQLPlugin.
database	string	yes	—	Database to scan.
user	string	yes	—	Login.
password	string (Secret)	no	empty	Password.
schemas_filter.include	list of regex	no	[".*"]	Schemas to include.
schemas_filter.exclude	list of regex	no	[]	Schemas to drop after include matches.

Source: CockroachDBPlugin in odd_collector/domain/plugin.py (extends PostgreSQLPlugin); reference YAML at config_examples/cocroachdb.yaml. The upstream filename has a cocroach typo — the type literal cockroachdb is correct and is what you write in collector_config.yaml.

plugins:
  - type: cockroachdb
    name: orders_crdb
    host: crdb.internal
    port: 26257
    database: orders
    user: odd_reader
    password: !ENV ${CRDB_PASSWORD}

Vertica (type: vertica)

Pulls schemas, tables, views, and columns from a Vertica analytic database.

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
host	string	yes	—	Vertica host.
port	string	yes	—	TCP port (typical: "5433").
database	string or null	no	—	Database to scan.
user	string	yes	—	Login.
password	string	yes	—	Password.

Source: VerticaPlugin in odd_collector/domain/plugin.py; reference YAML at config_examples/vertica.yaml.

plugins:
  - type: vertica
    name: vertica_main
    host: vertica.internal
    port: "5433"
    database: warehouse
    user: odd_reader
    password: !ENV ${VERTICA_PASSWORD}

SingleStore (type: singlestore)

Pulls schemas, tables, views, and columns from a SingleStore (formerly MemSQL) cluster. Wire-compatible with MySQL.

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
host	string	yes	—	SingleStore host.
port	string	yes	—	TCP port.
database	string or null	no	—	Database to scan.
user	string	yes	—	Login.
password	string	yes	—	Password.
ssl_disabled	boolean or null	no	false	Disables TLS to the server — typically only for local development.

Source: SingleStorePlugin in odd_collector/domain/plugin.py; reference YAML at config_examples/singlestore.yaml.

plugins:
  - type: singlestore
    name: singlestore_main
    host: singlestore.internal
    port: "3306"
    database: warehouse
    user: odd_reader
    password: !ENV ${SINGLESTORE_PASSWORD}

Oracle Database (type: oracle)

Pulls schemas (one per Oracle user), tables, views, and columns from an Oracle Database.

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
host	string	yes	—	Oracle server hostname.
port	string	yes	—	TCP port (typical: "1521").
user	string	yes	—	Oracle login (becomes the schema name in Oracle's data model).
service	string	yes	—	Oracle service name (e.g., XEPDB1). Use the service name, not the SID.
password	string (Secret)	yes	—	Password.
thick_mode	boolean or null	no	false	When true, switches the underlying Oracle client to thick mode (requires the Oracle Instant Client to be installed in the container). Default thin mode is pure Python and works without Instant Client.

Source: OraclePlugin in odd_collector/domain/plugin.py; reference YAML at config_examples/oracle.yaml.

plugins:
  - type: oracle
    name: oracle_main
    host: oracle.internal
    port: "1521"
    user: odd_reader
    service: XEPDB1
    password: !ENV ${ORACLE_PASSWORD}
    thick_mode: false

Generic ODBC source (type: odbc)

Pulls schemas, tables, and columns from any source reachable through an ODBC driver registered on the collector container. Useful for sources without a dedicated adapter.

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
host	string	yes	—	Source hostname.
port	string	yes	—	TCP port.
database	string	yes	—	Database to scan.
user	string	yes	—	Login.
password	string (Secret) or null	no	—	Password.
driver	string	no	"{ODBC Driver 17s for SQL Server}"	ODBC driver name as registered in odbcinst.ini on the container. The upstream default contains a typo (17s should be 17) — always set this field explicitly to the driver string for your environment (e.g., {ODBC Driver 17 for SQL Server} or your platform's equivalent). See Known limitations.

Source: OdbcPlugin in odd_collector/domain/plugin.py; reference YAML at config_examples/odbc.yaml.

plugins:
  - type: odbc
    name: legacy_odbc
    host: odbc.internal
    port: "1433"
    database: legacy
    user: odd_reader
    password: !ENV ${ODBC_PASSWORD}
    driver: "{ODBC Driver 17 for SQL Server}"

SQLite (type: sqlite)

Reads a SQLite database file from a local path on the collector container. In-memory SQLite databases are not supported (each connection sees its own private DB).

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
data_source	string (file path)	yes	—	Absolute path to the .db file inside the container. The file must exist at startup; the model uses Pydantic's FilePath validator.

Source: SQLitePlugin in odd_collector/domain/plugin.py; reference YAML at config_examples/sqlite.yaml.

plugins:
  - type: sqlite
    name: local_sqlite
    data_source: /data/local/file.db

Wide-column, document, and key-value stores

MongoDB (type: mongodb)

Catalogs MongoDB databases, collections, and inferred field types.

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
host	string	yes	—	MongoDB host or seed-list host.
port	string	yes	—	TCP port (typical: "27017").
database	string or null	no	—	Database to scan.
user	string	yes	—	Login.
password	string	yes	—	Password.
protocol	string	yes	—	Connection scheme passed to the MongoDB driver — mongodb for direct host/port connections, mongodb+srv for SRV-resolved seed lists (typical for MongoDB Atlas).

Source: MongoDBPlugin in odd_collector/domain/plugin.py; reference YAML at config_examples/mongodb.yaml.

plugins:
  - type: mongodb
    name: mongo_orders
    host: mongo.internal
    port: "27017"
    database: orders
    user: odd_reader
    password: !ENV ${MONGO_PASSWORD}
    protocol: mongodb

Apache Cassandra (type: cassandra)

Catalogs keyspaces, tables, and columns from a Cassandra cluster.

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
host	string	yes	—	Cassandra contact host.
port	string	yes	—	TCP port (typical: "9042").
database	string or null	no	—	Keyspace name; one plugin scans one keyspace when supplied.
user	string	yes	—	Login.
password	string	yes	—	Password.
contact_points	list of string	no	[]	Additional contact-host endpoints for the cluster's gossip layer. Empty list means the driver uses host only.

Source: CassandraPlugin in odd_collector/domain/plugin.py; reference YAML at config_examples/cassandra.yaml.

plugins:
  - type: cassandra
    name: cassandra_events
    host: cassandra-1.internal
    port: "9042"
    database: events
    user: odd_reader
    password: !ENV ${CASSANDRA_PASSWORD}
    contact_points: ["cassandra-2.internal", "cassandra-3.internal"]

ScyllaDB (type: scylladb)

Catalogs keyspaces, tables, and columns from a ScyllaDB cluster. Same field shape as Cassandra — Scylla is wire-compatible with the Cassandra driver.

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
host	string	yes	—	Scylla contact host.
port	string	yes	—	TCP port (typical: "9042").
database	string or null	no	—	Keyspace name.
user	string	yes	—	Login.
password	string	yes	—	Password.
contact_points	list of string	no	[]	Additional contact-host endpoints.

Source: ScyllaDBPlugin in odd_collector/domain/plugin.py; reference YAML at config_examples/scylladb.yaml.

plugins:
  - type: scylladb
    name: scylla_events
    host: scylla-1.internal
    port: "9042"
    database: events
    user: odd_reader
    password: !ENV ${SCYLLA_PASSWORD}
    contact_points: ["scylla-2.internal", "scylla-3.internal"]

Tarantool (type: tarantool)

Catalogs spaces and indexes from a Tarantool instance. Uses the standard DatabasePlugin shape with no Tarantool-specific fields.

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
host	string	yes	—	Tarantool host.
port	string	yes	—	TCP port (typical: "3301").
database	string or null	no	—	Database / space-collection identifier.
user	string	yes	—	Login.
password	string	yes	—	Password.

Source: TarantoolPlugin in odd_collector/domain/plugin.py; reference YAML at config_examples/tarantool.yaml.

plugins:
  - type: tarantool
    name: tarantool_main
    host: tarantool.internal
    port: "3301"
    user: odd_reader
    password: !ENV ${TARANTOOL_PASSWORD}

Couchbase (type: couchbase)

Catalogs Couchbase buckets and infers document field types by sampling. Couchbase is schemaless, so the adapter samples N documents per collection to derive a structural view.

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
host	string	yes	—	Couchbase connection string (e.g., couchbase://node1.internal,node2.internal).
bucket	string	yes	—	Bucket name; one plugin scans one bucket.
user	string	yes	—	Login.
password	string (Secret)	yes	—	Password.
sample_size	integer or null	no	0	Number of documents to sample per collection for schema inference. 0 disables sampling and uses the metadata-only view.
num_sample_values	integer or null	no	10	When sampling is on, number of value examples to retain per inferred field.

Source: CouchbasePlugin in odd_collector/domain/plugin.py; reference YAML at config_examples/couchbase.yaml.

plugins:
  - type: couchbase
    name: couchbase_orders
    host: couchbase://couchbase.internal
    bucket: orders
    user: odd_reader
    password: !ENV ${COUCHBASE_PASSWORD}
    sample_size: 100
    num_sample_values: 10

Neo4j (type: neo4j)

Catalogs Neo4j databases, node labels, and relationship types. Uses the standard DatabasePlugin shape; the typical Bolt port is 7687.

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
host	string	yes	—	Neo4j host.
port	string	yes	—	Bolt port (typical: "7687").
database	string or null	no	—	Database name (Neo4j 4.x+ multi-database).
user	string	yes	—	Login.
password	string	yes	—	Password.

Source: Neo4jPlugin in odd_collector/domain/plugin.py; reference YAML at config_examples/neo4j.yaml.

plugins:
  - type: neo4j
    name: neo4j_graph
    host: neo4j.internal
    port: "7687"
    database: neo4j
    user: neo4j
    password: !ENV ${NEO4J_PASSWORD}

Search engines

Elasticsearch (type: elasticsearch)

Catalogs Elasticsearch indices and field mappings.

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
host	string	yes	—	Elasticsearch host (typically including scheme — e.g., https://es.internal).
port	integer	yes	—	TCP port (typical: 9200).
username	string	yes	—	Login.
password	string (Secret)	yes	—	Password.
verify_certs	boolean or null	no	null	Whether to verify TLS certificates on the Elasticsearch endpoint. null defers to the Elasticsearch client default (verify when scheme is https).
ca_certs	string or null	no	null	Optional path to a CA bundle file inside the container.

Source: ElasticsearchPlugin in odd_collector/domain/plugin.py; reference YAML at config_examples/elasticsearch.yaml.

plugins:
  - type: elasticsearch
    name: es_logs
    host: https://es.internal
    port: 9200
    username: elastic
    password: !ENV ${ES_PASSWORD}
    verify_certs: true
    ca_certs: /etc/ssl/certs/ca-bundle.crt

OpenSearch (type: opensearch)

Catalogs OpenSearch indices and field mappings.

No config_examples/opensearch.yaml file exists upstream. The fields below are read directly from OpensearchPlugin in plugin.py; the YAML below is hand-crafted from that model.

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
host	string	yes	—	OpenSearch host (include scheme when using HTTPS).
port	integer or null	no	443	TCP port. The model defaults to 443 (the typical AWS OpenSearch Service port); set to 9200 for self-hosted.
http_compress	boolean or null	no	true	Whether to gzip request bodies.
use_ssl	boolean or null	no	true	Toggle TLS on the connection.
username	string or null	yes	—	Login. The model is Optional[str] with no explicit default — provide a value (or null) at config time.
password	string (Secret) or null	yes	—	Password. Same Pydantic shape as username — provide a value or null.
verify_certs	boolean or null	no	null	Whether to verify TLS certificates. null defers to the OpenSearch client default.
ca_certs	string or null	no	null	Optional path to a CA bundle file inside the container.

Source: OpensearchPlugin in odd_collector/domain/plugin.py.

plugins:
  - type: opensearch
    name: opensearch_logs
    host: https://search-mydomain.us-east-1.es.amazonaws.com
    port: 443
    use_ssl: true
    username: opensearch_admin
    password: !ENV ${OPENSEARCH_PASSWORD}
    verify_certs: true

Analytics engines and warehouses

Databricks Unity Catalog (type: databricks)

Catalogs Databricks Unity Catalog catalogs, schemas, tables, and columns via the Databricks workspace REST API.

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
workspace	string	yes	—	Databricks workspace URL (e.g., https://adb-1234567890.0.azuredatabricks.net).
token	string (Secret)	yes	—	Databricks personal access token (PAT) or service-principal token authorized for Unity Catalog.
catalogs	list of string or null	no	null	Allowlist of Unity Catalog catalogs. When omitted, every catalog the token can see is ingested. Literal name list — not a regex.

Source: DatabricksPlugin in odd_collector/domain/plugin.py; reference YAML at config_examples/databricks_unity_catalog.yaml. The upstream filename is databricks_unity_catalog.yaml while the type literal is the shorter databricks — write type: databricks in collector_config.yaml.

plugins:
  - type: databricks
    name: databricks_main
    workspace: https://adb-1234567890.0.azuredatabricks.net
    token: !ENV ${DATABRICKS_TOKEN}
    catalogs: ["main", "analytics"]

DuckDB (type: duckdb)

Reads one or more DuckDB database files from local paths on the collector container; can scan multiple files or whole directories of .db files in a single plugin.

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
paths	list of string (file paths)	yes	—	List of paths to .db files or directories containing .db files. Each path is opened independently.
host	string or null	no	"localhost"	Logical hostname used when generating ODDRNs for the catalog entries.

Source: DuckDBPlugin in odd_collector/domain/plugin.py; reference YAML at config_examples/duckdb.yaml.

plugins:
  - type: duckdb
    name: duckdb_local
    paths:
      - /data/analytics/warehouse.db
      - /data/analytics/extras/
    host: analytics-runner

Presto (type: presto)

Catalogs schemas, tables, and columns from a Presto coordinator.

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
host	string	yes	—	Presto coordinator host.
port	integer	yes	—	Coordinator HTTP port (typical: 8080 or 8081).
user	string	yes	—	User identity (Presto authenticates by user header by default).
principal_id	string or null	yes	—	Optional principal identifier for LDAP-configured clusters. The model is Optional[str] with no default — pass null (or empty string, as the upstream example does) when not using LDAP.
password	string or null	yes	—	LDAP password. Same Pydantic shape as principal_id — pass null / empty string on non-LDAP clusters.

Source: PrestoPlugin in odd_collector/domain/plugin.py; reference YAML at config_examples/presto.yaml.

plugins:
  - type: presto
    name: presto_main
    host: presto.internal
    port: 8081
    user: odd_reader
    principal_id: null
    password: null

Trino (type: trino)

Catalogs schemas, tables, and columns from a Trino coordinator. Wire-compatible with Presto (same client family).

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
host	string	yes	—	Trino coordinator host.
port	integer	yes	—	Coordinator HTTP port (typical: 8080 / 8081).
user	string	yes	—	User identity.
password	string or null	yes	—	LDAP password. Pass null / empty string on non-LDAP clusters.

Source: TrinoPlugin in odd_collector/domain/plugin.py; reference YAML at config_examples/trino.yaml.

plugins:
  - type: trino
    name: trino_main
    host: trino.internal
    port: 8081
    user: odd_reader
    password: null

Apache Druid (type: druid)

Catalogs Druid datasources via the broker API.

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
host	string	yes	—	Druid broker host.
port	integer	yes	—	Broker HTTP port (typical: 8082).

Source: DruidPlugin in odd_collector/domain/plugin.py; reference YAML at config_examples/druid.yaml.

plugins:
  - type: druid
    name: druid_main
    host: druid-broker.internal
    port: 8082

Apache Hive (type: hive)

Catalogs Hive databases, tables, and columns via HiveServer2. Configuration is grouped under a nested connection_params object — Hive's auth surface is varied enough that the adapter exposes the full HS2 connection knob set.

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
count_statistics	boolean	no	false	Whether to collect row-count statistics (SELECT COUNT(*)) per table. Off by default — these queries can be expensive on large warehouses.
connection_params	object	yes	—	Nested HS2 connection block — see fields below.
connection_params.host	string	yes	—	HiveServer2 host.
connection_params.port	integer or null	no	null	HS2 port. Defaults to 10000 when scheme is unset, 1000 when scheme: http / https (per the upstream HS2 client convention).
connection_params.database	string	yes	—	Hive database to scan.
connection_params.scheme	string or null	no	null	HS2 transport — "http" or "https" for HTTP transport; null for binary transport.
connection_params.auth	string or null	no	null	Auth mode — one of "BASIC", "NOSASL", "KERBEROS", "NONE". Defaults to NONE when omitted.
connection_params.username	string or null	no	null	Username. Used with auth: LDAP or auth: CUSTOM.
connection_params.password	string or null	no	null	Password. Used with auth: LDAP or auth: CUSTOM.
connection_params.kerberos_service_name	string or null	no	null	Used with auth: KERBEROS only.
connection_params.configuration	object or null	no	null	Free-form dict of Hive session configuration overrides.
connection_params.check_hostname	string or null	no	null	TLS hostname check toggle as a string "true" / "false".
connection_params.ssl_cert	string or null	no	null	Path to a CA / client certificate file inside the container.

Source: HivePlugin and HiveConnectionParams in odd_collector/domain/plugin.py; reference YAML at config_examples/hive.yaml.

plugins:
  - type: hive
    name: hive_main
    count_statistics: false
    connection_params:
      host: hive.internal
      port: 10000
      database: default
      auth: NONE

Business intelligence and dashboards

Tableau (type: tableau)

Catalogs Tableau site content (workspaces, projects, dashboards, sheets) via the Tableau REST API.

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
server	string	yes	—	Tableau Server / Tableau Cloud URL.
site	string or null	yes	—	Tableau site name (empty string for the default site). The model is Optional[str] with no default — pass an explicit value.
user	string or null	yes	—	Username. Pass null if authenticating via token_name / token_value.
password	string (Secret) or null	yes	—	Password. Pass null if authenticating via PAT.
token_name	string or null	yes	—	Personal access token name (for 2FA / SSO accounts that can't use password auth).
token_value	string (Secret) or null	yes	—	Personal access token value.
pagination_size	integer	no	10	Page size for the REST API. Larger values reduce request count but increase per-request latency; tune for very large sites.

Source: TableauPlugin in odd_collector/domain/plugin.py; reference YAML at config_examples/tableau.yaml.

plugins:
  - type: tableau
    name: tableau_main
    server: https://tableau.internal
    site: analytics
    user: null
    password: null
    token_name: odd-reader-pat
    token_value: !ENV ${TABLEAU_TOKEN}
    pagination_size: 50

Apache Superset (type: superset)

Catalogs Superset datasets, dashboards, and charts via the Superset REST API.

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
server	string	yes	—	Superset base URL (include trailing slash).
username	string	yes	—	Superset login.
password	string (Secret)	yes	—	Password.

Source: SupersetPlugin in odd_collector/domain/plugin.py; reference YAML at config_examples/superset.yaml.

plugins:
  - type: superset
    name: superset_main
    server: https://superset.internal/
    username: admin
    password: !ENV ${SUPERSET_PASSWORD}

Metabase (type: metabase)

Catalogs Metabase dashboards, questions, and the underlying datasets they reference.

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
host	string	yes	—	Metabase host.
port	string	yes	—	TCP port.
login	string	yes	—	Metabase login email.
password	string (Secret)	yes	—	Password.

Source: MetabasePlugin in odd_collector/domain/plugin.py; reference YAML at config_examples/metabase.yaml.

plugins:
  - type: metabase
    name: metabase_main
    host: metabase.internal
    port: "3000"
    login: [email protected]
    password: !ENV ${METABASE_PASSWORD}

Redash (type: redash)

Catalogs Redash queries and dashboards via the Redash API.

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
server	string	yes	—	Redash server base URL.
api_key	string	yes	—	Redash API key (account-scoped — gives the adapter access to whatever the account can see).

Source: RedashPlugin in odd_collector/domain/plugin.py; reference YAML at config_examples/redash.yaml.

plugins:
  - type: redash
    name: redash_main
    server: https://redash.internal
    api_key: !ENV ${REDASH_API_KEY}

Mode Analytics (type: mode)

Catalogs Mode reports and the underlying datasets they reference.

No config_examples/mode.yaml file exists upstream. The fields below are read directly from ModePlugin in plugin.py; the YAML below is hand-crafted from that model.

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
host	string	yes	—	Mode workspace host (e.g., https://app.mode.com).
account	string	yes	—	Mode account / workspace identifier.
data_source	string	yes	—	Mode data-source identifier the adapter should report against.
token	string (Secret) or null	yes	—	API token. The model is Optional[SecretStr] with no default — pass a value or null if relying on password auth.
password	string (Secret) or null	yes	—	Password (legacy auth). Pass null if using token auth.

Source: ModePlugin in odd_collector/domain/plugin.py.

plugins:
  - type: mode
    name: mode_main
    host: https://app.mode.com
    account: my_workspace
    data_source: my_warehouse
    token: !ENV ${MODE_TOKEN}
    password: null

Cube.js (type: cubejs)

Catalogs Cube.js cubes and members; uses the cube's underlying SQL data source to resolve lineage from cube measures back to the source columns.

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
host	string	yes	—	Cube.js server base URL.
dev_mode	boolean	no	false	When true, the adapter relaxes auth — token may be null. In production (dev_mode: false), token is required and the adapter raises ValueError on startup if it isn't set.
token	string (Secret) or null	conditional	null	Cube.js auth token — required unless dev_mode: true.
predefined_datasource	object	yes	—	Sub-object describing the SQL data source backing the cubes — used by the adapter's SQL parser to generate lineage-edge ODDRNs. Only postgres and clickhouse are recognised types (see PostgresDatasource / ClickHouseDatasource in predefined_data_source.py).
predefined_datasource.type	string	yes	—	"postgres" or "clickhouse".
predefined_datasource.host	string or null	no	null	Source host — used as the lineage ODDRN host.
predefined_datasource.database	string or null	no	null	Source database.

Source: CubeJSPlugin in odd_collector/domain/plugin.py; reference YAML at config_examples/cubejs.yaml.

plugins:
  - type: cubejs
    name: cubejs_main
    host: http://cube.internal:4000
    dev_mode: false
    token: !ENV ${CUBEJS_TOKEN}
    predefined_datasource:
      type: postgres
      host: pg.internal
      database: warehouse

Catalog, ingestion, and federation

CKAN (type: ckan)

Catalogs CKAN packages and resources from the CKAN action API.

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
host	string	yes	—	CKAN host.
port	string	yes	—	TCP port.
ckan_endpoint	string	no	empty	Optional path prefix between the host and the CKAN action API (e.g., "/additional/endpoint"). When the API is mounted at the root, leave empty.
token	string (Secret) or null	no	null	CKAN auth token. Some action endpoints require authorization.

Source: CKANPlugin in odd_collector/domain/plugin.py; reference YAML at config_examples/ckan.yaml.

plugins:
  - type: ckan
    name: ckan_main
    host: ckan.internal
    port: "80"
    ckan_endpoint: ""
    token: !ENV ${CKAN_TOKEN}

Airbyte (type: airbyte)

Catalogs Airbyte connectors, sources, destinations, and the lineage edges between them.

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
host	string	yes	—	Airbyte API host.
port	string	yes	—	Airbyte API port (typical: "8000").
user	string or null	yes	—	Airbyte username. The model is Optional[str] with no default — pass a value or null for unauthenticated deployments.
password	string or null	yes	—	Airbyte password. Same Pydantic shape as user.
platform_host_url	string	yes	—	The ODD Platform URL the adapter advertises in generated ODDRNs for downstream destinations. This is a per-plugin field on AirbytePlugin that overlaps with the collector-level platform_host_url at the top of collector_config.yaml — both must be set when using Airbyte.
store_raw_tables	boolean	no	true	Whether to ingest Airbyte's _airbyte_raw_* staging tables. Set to false to keep them out of the catalog.

Source: AirbytePlugin in odd_collector/domain/plugin.py; reference YAML at config_examples/airbyte.yaml.

plugins:
  - type: airbyte
    name: airbyte_main
    host: airbyte.internal
    port: "8000"
    platform_host_url: http://odd-platform.internal:8080
    user: airbyte
    password: !ENV ${AIRBYTE_PASSWORD}
    store_raw_tables: false

Fivetran (type: fivetran)

Catalogs a single Fivetran connector and its destination via the Fivetran REST API.

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
base_url	string	no	"https://api.fivetran.com"	Fivetran API base URL. Override only for Fivetran's regional API endpoints.
api_key	string	yes	—	Fivetran API key.
api_secret	string (Secret)	yes	—	Fivetran API secret.
connector_id	string	yes	—	Fivetran connector identifier — one plugin = one connector. Add a second plugin entry per additional connector.
destination_id	string	yes	—	Fivetran destination identifier corresponding to the connector.

Source: FivetranPlugin in odd_collector/domain/plugin.py; reference YAML at config_examples/fivetran.yaml.

plugins:
  - type: fivetran
    name: fivetran_orders
    api_key: !ENV ${FIVETRAN_API_KEY}
    api_secret: !ENV ${FIVETRAN_API_SECRET}
    connector_id: orders_connector
    destination_id: warehouse_destination

dbt Cloud catalog import (type: dbt)

Pulls dbt model lineage and metadata via a pre-uploaded catalog.json on a host the adapter can reach. This is the pull dbt adapter — distinct from odd-dbt, the push-strategy adapter that emits live test results from dbt runs.

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
host	string	yes	—	Logical host used for ODDRN generation — typically the dbt Cloud / dbt Core deployment host.
odd_catalog_url	string	yes	—	URL the adapter fetches the catalog.json from.

Source: DbtPlugin in odd_collector/domain/plugin.py; reference YAML at config_examples/dbt.yaml.

plugins:
  - type: dbt
    name: dbt_catalog
    host: dbt.internal
    odd_catalog_url: https://dbt.internal/catalog.json

Federated ODD Platform (type: odd_adapter)

Pulls metadata from another ODD Platform instance — federate a child platform's catalog into a parent platform.

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
host	string	yes	—	URL of the source ODD service that implements the odd_adapter Ingress API.
data_source_oddrn	string	yes	—	The ODDRN to advertise as the federated data source root (e.g., //my_adapter/host/source-platform.internal).

Source: OddAdapterPlugin in odd_collector/domain/plugin.py; reference YAML at config_examples/odd_adapter.yaml.

plugins:
  - type: odd_adapter
    name: federated_eu
    host: http://odd-platform-eu.internal:8000
    data_source_oddrn: //my_adapter/host/odd-platform-eu.internal:8000

Machine learning platforms

MLflow (type: mlflow)

Catalogs MLflow experiments, runs, and registered models from the MLflow tracking and model-registry APIs.

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
dev_mode	boolean	no	false	Adapter-side dev mode toggle.
tracking_uri	string	yes	—	MLflow tracking server URI.
registry_uri	string	yes	—	MLflow model-registry URI (often the same as tracking_uri).
filter_experiments	list of string or null	no	null	Allowlist of experiment names. When omitted, every experiment is ingested. Literal name list — not a regex.

Source: MlflowPlugin in odd_collector/domain/plugin.py; reference YAML at config_examples/mlflow.yaml.

plugins:
  - type: mlflow
    name: mlflow_main
    dev_mode: false
    tracking_uri: https://mlflow.internal
    registry_uri: https://mlflow.internal
    filter_experiments: ["churn_v2", "fraud_detection"]

Feast feature store (type: feast)

Catalogs Feast feature views and entities by reading the Feast repo definition from a path on the collector container.

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
host	string	yes	—	Logical host used for ODDRN generation.
repo_path	string	yes	—	Path to a checked-out Feast feature-repo inside the container.

Source: FeastPlugin in odd_collector/domain/plugin.py; reference YAML at config_examples/feast.yaml.

plugins:
  - type: feast
    name: feast_features
    host: feast.internal
    repo_path: /opt/feast/feature_repo

Kubeflow Pipelines (type: kubeflow)

Catalogs Kubeflow pipelines, runs, and the lineage edges between them.

Field	Type	Required	Default	Description
name	string	yes	—	Operator-chosen unique plugin name.
host	string	yes	—	Kubeflow Pipelines host (typically the KFP UI URL).
namespace	string	yes	—	Kubernetes namespace Kubeflow runs in — not the same as ODD's namespace metadata field. The Kubeflow plugin redeclares namespace as required at the plugin level, which shadows BasePlugin's optional metadata field.
session_cookie0	string or null	yes	—	First half of the KFP session cookie (Istio AuthService split-cookie pattern). The model is Optional[str] with no default — provide a value or null.
session_cookie1	string or null	yes	—	Second half of the KFP session cookie.

Source: KubeflowPlugin in odd_collector/domain/plugin.py; reference YAML at config_examples/kubeflow.yaml.

plugins:
  - type: kubeflow
    name: kubeflow_main
    host: https://kfp.internal
    namespace: kubeflow-user
    session_cookie0: !ENV ${KFP_COOKIE_0}
    session_cookie1: !ENV ${KFP_COOKIE_1}

Per-adapter feature matrix

Cross-cutting capabilities and where they apply across the 41-adapter set:

Feature	Where it applies	What it does
Regex ingestion filters (Filter)	postgresql.schemas_filter, snowflake.schemas_filter, cockroachdb.schemas_filter (inherited from PostgreSQLPlugin)	Regex include / exclude lists scope which schemas / objects the adapter sees. Defaults to "include everything" when omitted. Source: Filter in odd_collector_sdk/domain/filter.py.
Literal-name allowlist filters	redshift.schemas (schemas), databricks.catalogs (Unity Catalog catalogs), mlflow.filter_experiments (experiment names)	Plain list of names the adapter restricts ingestion to. Not a regex — literal exact match. When omitted, the adapter ingests everything visible to the credentials.
ERD relationships (foreign keys)	postgresql, snowflake, cockroachdb (via PostgreSQL inheritance)	The adapter emits ENTITY_RELATIONSHIP entities for tables connected by foreign keys, including cross-schema. The platform renders these as ERD edges on dataset detail pages. No other adapter currently extracts foreign-key relationships.
TLS toggles on the source connection	clickhouse.secure + clickhouse.verify, mysql.ssl_disabled, singlestore.ssl_disabled, elasticsearch.verify_certs + elasticsearch.ca_certs, opensearch.use_ssl + opensearch.verify_certs + opensearch.ca_certs	Per-adapter knobs for TLS toggling and certificate validation. Defaults are tuned per adapter (see each section above) — only override for self-signed certs on local clusters or for unencrypted local development.
Token-based auth (PAT / API token alternative to password)	tableau.token_name + tableau.token_value, databricks.token, redash.api_key, fivetran.api_key + fivetran.api_secret, cubejs.token, mlflow (via tracking_uri auth), ckan.token, mode.token, kubeflow.session_cookie0 + kubeflow.session_cookie1	Replaces username/password auth; required when the source enforces SSO / 2FA / MFA on user accounts.
Schema inference via document sampling	couchbase.sample_size + couchbase.num_sample_values	The adapter samples N documents per collection to derive a structural view of fields and value types. Defaults to no sampling (sample_size: 0) — set explicitly to enable.
Sub-object connection block (advanced auth surface)	hive.connection_params (full HS2 connection knob set), cubejs.predefined_datasource (postgres / clickhouse only — used to resolve cube-to-source lineage)	Some adapters expose a nested object instead of flat fields when the auth or lineage surface needs more knobs than a flat schema supports.
Multiple file paths in one plugin	duckdb.paths	DuckDB accepts a list of .db files or directories of .db files in one plugin — every file is opened independently. Other file-source adapters (sqlite) take a single path.
Special operating modes	oracle.thick_mode (Oracle Instant Client vs. pure-Python), cubejs.dev_mode (relax token requirement), mlflow.dev_mode	Adapter-level toggles that alter runtime behaviour or auth strictness; safe defaults are off.

Other adapters either do not expose filters (the SDK ones don't carry a Filter field) or do not emit relationships. For the filter mechanism's user-facing explanation (include / exclude semantics, when filters apply, default behaviour without filters), see Ingestion filters. The full cross-adapter capability matrix — which adapter exposes which filter, which emits which relationship type — lives on the odd-collectors monorepo README; check that table when planning a new deployment.

Known limitations

• README drift on the source repo: as flagged above, the upstream README's adapter table omits four adapters (databricks, couchbase, opensearch, oracle) that exist in PLUGIN_FACTORY. This is a docs gap on the collector repo, not a missing capability — those four adapters work; they're just under-advertised.

• Foreign-key extraction is PostgreSQL/Snowflake only today. ClickHouse, MySQL, MSSQL, and others extract schemas and columns but not foreign-key relationships.

• No per-plugin pulling_interval: every plugin in the file shares default_pulling_interval. Splitting workloads with different cadences requires running multiple collector containers, each with its own config.

• M1 / Apple Silicon build issues: pyodbc, confluent-kafka, and grpcio need extra environment variables to build natively. See the generic collector README → M1 building issue.

• odbc.driver upstream typo: OdbcPlugin.driver defaults to "{ODBC Driver 17s for SQL Server}" (with an extra s) — s should not be in the driver string. Always set driver: explicitly in the plugin config to the registered driver name on your container (e.g., {ODBC Driver 17 for SQL Server} or your platform's equivalent). Without an explicit value, the adapter's connection attempt fails because no ODBC driver matches the typoed string. The reference YAML at config_examples/odbc.yaml uses the correct value, so copy from there rather than relying on the model default.

• Missing upstream config examples for mode and opensearch: both adapters are present in PLUGIN_FACTORY and shipped, but odd-collectors/odd-collector/config_examples/ does not contain a mode.yaml or opensearch.yaml. The per-adapter sections above include hand-crafted YAML examples derived from the Pydantic models for both.

• config_examples/cocroachdb.yaml filename typo: the file containing the CockroachDB reference YAML is cocroachdb.yaml (missing the k). The type literal (cockroachdb) is correct — the file's contents work as-is; only the filename is misspelled.

Where to next

• odd-collector-aws — when your source is an AWS managed service.

• odd-collector-azure / odd-collector-gcp — for Azure / GCP.

• odd-collector-profiler — when you want statistical profiles on a Postgres / Azure SQL source.

• Collector secrets backend — to source any field from AWS SSM instead of inline YAML.

• Build and run ODD Collectors — full SDK config reference and from-source build / run instructions.