Configuration fields¶

Every field of TablePartitionConfig and of the models it is built from, in one place. Types are the accepted inputs; defaults are what you get when the field is omitted.

TablePartitionConfig¶

Flat fields, accepted instead of scheme / lifecycle for a time-partitioned RANGE root:

Derived, read-only: qualified_name, partition_type, partition_strategy, partition_column, partition_columns, key_arity, granularity, time_boundaries, subpartition, create_ahead_count, retention_count, levels, is_time_based, is_progression_root, has_progression_level, manages_foreign_leaves.

Scheme levels¶

Common to every level:

RangePartitioning¶

HashPartitioning¶

ListPartitioning¶

ListGroup(name, values): a name fragment and the values it owns, written as string literals.

Boundaries¶

TimeBoundaries¶

NumericBoundaries¶

IntegerSequence¶

LifecyclePolicy¶

Creation rules: CreateAhead(count ≥ 1); CreateUntil(position) — a datetime, an int, or their string forms; CreateNextIf(when) — a predicate over the newest partition.

Retention rules and predicates: KeepNewest(count ≥ 1), KeepFor(age: timedelta), KeepBehind(distance ≥ 1), ExpireIf(when), SizeAbove(bytes ≥ 1), RowsAbove(rows ≥ 0), WindowAgeAbove(age), Unreferenced(), SqlPredicate(sql) (must contain {partition}), Callback(fn, facts=frozenset(), label="callback"), AllOf(members), AnyOf(members), Not(member).

Drop rules: DropAfter(grace: timedelta = 0, when: predicate | None = None); DropNever().

Leaves¶

LocalLeaves¶

ForeignLeaves¶

Repository and metadata provider¶

PostgresPartitionRepository(engine, *, ddl_timezone="UTC", ddl_timeout_seconds=30, marker_prefix=None, drop_allow_unmanaged=False, drop_lock_timeout_ms=3000, drop_max_retries=3, drop_retry_delay=0.5, drop_max_backoff=300).

PostgresMetadataProvider(engine, *, marker_prefix=None, boundary_codec=None, ddl_timezone=None).

PartitionToolkit.from_engine(engine, *, hooks=None, locks=None, marker_prefix=None, ddl_timezone="UTC", ddl_timeout_seconds=30, boundary_codec=None, lock_prefix="partitioner", lock_min_interval_seconds=0, drop_allow_unmanaged=False, drop_lock_timeout_ms=3000, drop_max_retries=3, drop_retry_delay=0.5, drop_max_backoff=300) builds all of them around one engine and returns them as repo / metadata / locks / service / maintainer, with each shared setting given once. Construct PartitionToolkit(...) directly to hold parts of your own.

marker_prefix must be the same on the repository and the provider — the first writes the ownership marker, the second finds it — and PartitionLifecycleService refuses a pair that disagrees. The provider's boundary_codec and ddl_timezone are used by is_partition_closed alone, which also takes boundaries=config.time_boundaries and reads both from there instead.

PostgresAdvisoryLockManager(engine, prefix="partitioner", acquire_min_interval_seconds=0); RedisDistributedLockManager(redis_client, prefix="partitioner:lock", ttl_seconds=300, acquire_min_interval_seconds=0).

Serialization¶

config.model_dump(mode="json", by_alias=True) round-trips through TablePartitionConfig.model_validate. config.fingerprint is a digest of that form, which is what a plan records in config_fingerprint and what apply() compares against; anything excluded from serialization is invisible to it, so two configurations differing only in a Callback's function share a fingerprint. Boundaries dump with a kind (time, integer, sequence), levels with method (range, list, hash), policies and predicates with kind, leaves with kind (local, foreign). Custom calculators, custom codecs and Callback functions are left out.