Metaxy + PostgreSQL¶
Experimental
This functionality is experimental.
Metadata managed by Metaxy can be stored in PostgreSQLMetadataStore.
It uses PostgreSQL.
This metadata store backend is limited in comparison to others, because PostgreSQL doesn't support map-like data types, and Metaxy's versioning engine can't run in the database.
The local Polars versioning engine is used instead.
This results in the following limitations for MetadataStore.resolve_update:
- Increased I/O: entire upstream metadata has to be fetched to memory
- Increased Memory footprint: expect high memory usage, especially when having many upstream features
Metaxy's Versioning Struct Columns¶
PostgreSQL doesn't have native map-like or struct types, so it's recommended to store Metaxy's versioning columns as JSONB.
As a convenience feature, PostgreSQLMetadataStore will automatically json-encodes pl.Struct columns when writing metadata and parse them to pl.Struct when reading.
This behavior can be disabled with auto_cast_struct_for_jsonb configuration parameter. This setting only affects user-defined columns, while Metaxy's versioning columns are always encoded/parsed.
API Reference¶
metaxy.ext.postgresql
¶
PostgreSQL metadata store extension.
metaxy.ext.postgresql.PostgreSQLMetadataStore
¶
PostgreSQLMetadataStore(
connection_string: str | None = None,
*,
connection_params: dict[str, Any] | None = None,
fallback_stores: list[MetadataStore] | None = None,
auto_cast_struct_for_jsonb: bool = True,
**kwargs: Any,
)
Bases: IbisMetadataStore
Experimental
This functionality is experimental.
PostgreSQL metadata store with storage/compute separation.
Uses PostgreSQL for storage and Polars for versioning. Filters push down to SQL WHERE clauses, then data materializes to Polars.
Example
Parameters:
-
connection_string(str | None, default:None) –PostgreSQL connection URI (e.g., "postgresql://user:pass@host:5432/db")
-
connection_params(dict[str, Any] | None, default:None) –Dict with keys: host, port, database, user, password (alternative to connection_string)
-
fallback_stores(list[MetadataStore] | None, default:None) –List of fallback stores for chaining
-
auto_cast_struct_for_jsonb(bool, default:True) –If True, JSON-encode all Struct columns to strings on write (Metaxy system Struct columns are always converted). The actual SQL column type (e.g., JSON, JSONB, or TEXT) is determined by the table schema, not this flag.
-
**kwargs(Any, default:{}) –Additional arguments passed to IbisMetadataStore
Source code in src/metaxy/ext/postgresql/metadata_store.py
def __init__(
self,
connection_string: str | None = None,
*,
connection_params: dict[str, Any] | None = None,
fallback_stores: list["MetadataStore"] | None = None,
auto_cast_struct_for_jsonb: bool = True,
**kwargs: Any,
):
"""Initialize PostgreSQL metadata store.
Args:
connection_string: PostgreSQL connection URI
(e.g., "postgresql://user:pass@host:5432/db")
connection_params: Dict with keys: host, port, database, user, password
(alternative to connection_string)
fallback_stores: List of fallback stores for chaining
auto_cast_struct_for_jsonb: If True, JSON-encode all Struct columns to strings on write
(Metaxy system Struct columns are always converted). The actual SQL column type
(e.g., JSON, JSONB, or TEXT) is determined by the table schema, not this flag.
**kwargs: Additional arguments passed to IbisMetadataStore
"""
if connection_string is None and connection_params is None:
raise ValueError("Must provide either connection_string or connection_params for PostgreSQL")
self.auto_cast_struct_for_jsonb = auto_cast_struct_for_jsonb
super().__init__(
connection_string=connection_string,
backend="postgres",
connection_params=connection_params,
fallback_stores=fallback_stores,
**kwargs,
)
Attributes¶
metaxy.ext.postgresql.PostgreSQLMetadataStore.materialization_id
property
¶
materialization_id: str | None
The external orchestration ID for this store instance.
If set, all metadata writes include this ID in the metaxy_materialization_id column,
allowing filtering of rows written during a specific materialization run.
metaxy.ext.postgresql.PostgreSQLMetadataStore.name
property
¶
name: str | None
The configured name of this store, if any.
metaxy.ext.postgresql.PostgreSQLMetadataStore.qualified_class_name
property
¶
qualified_class_name: str
The fully qualified class name (module.classname).
metaxy.ext.postgresql.PostgreSQLMetadataStore.conn
property
¶
Get Ibis backend connection.
Returns:
-
SQLBackend–Active Ibis backend connection
Raises:
-
StoreNotOpenError–If store is not open
metaxy.ext.postgresql.PostgreSQLMetadataStore.sqlalchemy_url
property
¶
sqlalchemy_url: str
Get SQLAlchemy-compatible connection URL for tools like Alembic.
Returns the connection string if available. If the store was initialized with backend + connection_params instead of a connection string, raises an error since constructing a proper URL is backend-specific.
Returns:
-
str–SQLAlchemy-compatible URL string
Raises:
-
ValueError–If connection_string is not available
Functions¶
metaxy.ext.postgresql.PostgreSQLMetadataStore.config_model
classmethod
¶
config_model() -> type[PostgreSQLMetadataStoreConfig]
metaxy.ext.postgresql.PostgreSQLMetadataStore.native_implementation
¶
native_implementation() -> Implementation
metaxy.ext.postgresql.PostgreSQLMetadataStore.transform_before_write
¶
transform_before_write(
df: Frame, feature_key: FeatureKey, table_name: str
) -> Frame
Convert Struct columns to JSON strings using Polars.
Uses Polars' struct.json_encode() to serialize structs to JSON strings. Stored SQL type depends on target table schema (JSON/JSONB vs TEXT).
Source code in src/metaxy/ext/postgresql/metadata_store.py
def transform_before_write(self, df: Frame, feature_key: FeatureKey, table_name: str) -> Frame:
"""Convert Struct columns to JSON strings using Polars.
Uses Polars' struct.json_encode() to serialize structs to JSON strings.
Stored SQL type depends on target table schema (JSON/JSONB vs TEXT).
"""
pl_df = collect_to_polars(df)
return nw.from_native(self._encode_struct_columns(pl_df))
metaxy.ext.postgresql.PostgreSQLMetadataStore.transform_after_read
¶
transform_after_read(
table: Table, feature_key: FeatureKey
) -> Table
Cast JSONB columns to String so Polars can parse them back to Structs.
PostgreSQL JSONB columns appear as JSON type in Ibis. We cast to String so that when materialized to Polars, we can parse them back to Structs.
Source code in src/metaxy/ext/postgresql/metadata_store.py
def transform_after_read(self, table: "ibis.Table", feature_key: FeatureKey) -> "ibis.Table":
"""Cast JSONB columns to String so Polars can parse them back to Structs.
PostgreSQL JSONB columns appear as JSON type in Ibis. We cast to String
so that when materialized to Polars, we can parse them back to Structs.
"""
schema = table.schema()
json_columns = self._get_json_columns_for_struct(schema)
if json_columns:
mutations = {col_name: table[col_name].cast("string") for col_name in json_columns}
return table.mutate(**mutations)
return table
Configuration¶
Experimental
This functionality is experimental.
Configuration for PostgreSQLMetadataStore.
Inherits connection_string, connection_params, table_prefix, auto_create_tables from IbisMetadataStoreConfig.
Show JSON schema:
{
"$defs": {
"HashAlgorithm": {
"description": "Supported hash algorithms for field provenance calculation.\n\nThese algorithms are chosen for:\n\n- Speed (non-cryptographic hashes preferred)\n\n- Cross-database availability\n\n- Good collision resistance for field provenance calculation",
"enum": [
"xxhash64",
"xxhash32",
"wyhash",
"sha256",
"md5",
"farmhash"
],
"title": "HashAlgorithm",
"type": "string"
}
},
"additionalProperties": false,
"description": "Configuration for PostgreSQLMetadataStore.\n\nInherits connection_string, connection_params, table_prefix, auto_create_tables from IbisMetadataStoreConfig.",
"properties": {
"fallback_stores": {
"description": "List of fallback store names to search when features are not found in the current store.",
"items": {
"type": "string"
},
"title": "Fallback Stores",
"type": "array"
},
"hash_algorithm": {
"anyOf": [
{
"$ref": "#/$defs/HashAlgorithm"
},
{
"type": "null"
}
],
"default": null,
"description": "Hash algorithm for versioning. If None, uses store's default."
},
"versioning_engine": {
"default": "auto",
"description": "Which versioning engine to use: 'auto' (prefer native), 'native', or 'polars'.",
"enum": [
"auto",
"native",
"polars"
],
"title": "Versioning Engine",
"type": "string"
},
"connection_string": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"description": "Ibis connection string (e.g., 'clickhouse://host:9000/db').",
"title": "Connection String"
},
"backend": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"description": "Ibis backend name (e.g., 'clickhouse', 'postgres', 'duckdb').",
"mkdocs_metaxy_hide": true,
"title": "Backend"
},
"connection_params": {
"anyOf": [
{
"additionalProperties": true,
"type": "object"
},
{
"type": "null"
}
],
"default": null,
"description": "Backend-specific connection parameters.",
"title": "Connection Params"
},
"table_prefix": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"description": "Optional prefix for all table names.",
"title": "Table Prefix"
},
"auto_create_tables": {
"anyOf": [
{
"type": "boolean"
},
{
"type": "null"
}
],
"default": null,
"description": "If True, create tables on open. For development/testing only.",
"title": "Auto Create Tables"
},
"auto_cast_struct_for_jsonb": {
"default": true,
"description": "Whether to encode/decode Struct columns to/from JSON on writes/reads. Metaxy system columns are always converted.",
"title": "Auto Cast Struct For Jsonb",
"type": "boolean"
}
},
"title": "PostgreSQLMetadataStoreConfig",
"type": "object"
}
fallback_stores
pydantic-field
¶
List of fallback store names to search when features are not found in the current store.
hash_algorithm
pydantic-field
¶
hash_algorithm: HashAlgorithm | None = None
Hash algorithm for versioning. If None, uses store's default.
versioning_engine
pydantic-field
¶
versioning_engine: Literal["auto", "native", "polars"] = (
"auto"
)
Which versioning engine to use: 'auto' (prefer native), 'native', or 'polars'.
connection_string
pydantic-field
¶
connection_string: str | None = None
Ibis connection string (e.g., 'clickhouse://host:9000/db').
connection_params
pydantic-field
¶
Backend-specific connection parameters.
auto_create_tables
pydantic-field
¶
auto_create_tables: bool | None = None
If True, create tables on open. For development/testing only.