Skip to main content

Changelog

All notable changes to LakeTS are documented in this file.

Format follows Keep a Changelog, versioning follows Semantic Versioning.

[Unreleased]​

[0.1.3] - 2026-06-09​

Fixed​

  • Timezone-stable chunk names. _ensure_partitions rendered chunk_name with to_char(range_start, …) in the session timezone, while range_start is an absolute timestamptz. Running the partition manager under different session timezones (e.g. a job in UTC vs a client in UTC+2) produced names that no longer matched range_start, so the partition manager hit duplicate key value violates unique constraint "idx_chunk_metadata_chunk_name" (the ON CONFLICT (chronotable_id, range_start) couldn't absorb a chunk_name collision). Names are now rendered from the instant in UTC (v_start AT TIME ZONE 'UTC'), keeping chunk_name 1:1 with range_start regardless of session timezone.

Upgrade by reinstalling dist/lakets.sql. Any chunk rows created before this fix under a non-UTC session keep their old names; if the partition manager still collides, drop the stale rows for already-dropped chunks (their partitions are gone) and let it recreate them.

[0.1.2] - 2026-06-08​

Fixed​

  • add_tiered_retention_policy now actually tiers. The flag/tier phase only recognized tiering policies, so a combined tiered_retention policy was never flagged (Found 0 table(s) with tiering policies) — only its drop horizon worked. Fixed across the flag path: the tiering job and _get_chunks_to_tier now match policy_type IN ('tiering', 'tiered_retention') and read the horizon from after or tier_after. add_tiered_retention_policy also installs the write-tracking trigger, backfills last_write_lsn on existing chunks, and sets tiering_enabled = TRUE — without these, tier_chunk deferred every chunk.

Upgrade by reinstalling dist/lakets.sql and redeploying the maintenance jobs (the fix spans both SQL and tiering_job.py).

[0.1.1] - 2026-06-08​

Fixed​

  • Sync/tiering no longer require USAGE on the Lakebase-managed wal2delta schema. enable_sync() previously aborted with permission denied for schema wal2delta when the connected role couldn't see the CDF subsystem schema. The CDF probe now tolerates insufficient_privilege: enable_sync() warns and continues (the shadow and trigger are created as usual), and the tiering durability gate fails closed (it evicts nothing it cannot verify). See Troubleshooting for the grants needed if you want CDF-gated tiering to evict.

Upgrade by reinstalling dist/lakets.sql (the change is function-level — no schema migration required).

[0.1.0] - 2026-06-05​

Initial public release. LakeTS turns Databricks Lakebase (PostgreSQL 17+) into a time-series database — pure PL/pgSQL, no custom extensions.

Added​

  • ChronoTables — automatic time-based partitioning via create_chronotable(), plus multi-metric (tag + field) tables via create_metric_table().
  • Time-series functions — time_bucket, time_bucket_gapfill, first/last aggregates, locf, interpolate, delta, rate, and histogram.
  • RollUp engine — incremental aggregates with watermark-driven refresh, per-bucket invalidation tracking, chunk-skip pruning, and DAG-ordered cascade refresh.
  • Last Value Cache — sub-10ms latest-state lookups via enable_lvc().
  • Lifecycle — policy-driven tiering that flags chunks once durable in Unity Catalog, and retention that drops Lakebase partitions only after CDF confirms durability (with a force override).
  • Unity Catalog sync — enable_sync() mirrors ChronoTables and RollUps to Unity Catalog through Lakebase CDF via an unpartitioned shadow table.
  • Alerts — SQL-native alert_check() and alert_deadman().
  • Bulk ingest — ingest_batch() for JSONB arrays and ingest_prometheus().
  • Monitoring — Prometheus-compatible metrics, chunk_health(), and a Databricks AI/BI dashboard.
  • Databricks jobs — serverless maintenance for partitioning, RollUp refresh, tiering, and retention.