timescaledb

timescaledb : Enables scalable inserts and complex queries for time-series data

Overview

Packages

Source

pig build pkg timescaledb;		# build rpm/deb

Install

Make sure PGDG and PIGSTY repo available:

pig repo add pgsql -u   # add both repo and update cache

Install this extension with pig:

pig install timescaledb;		# install via package name, for the active PG version

pig install timescaledb -v 18;   # install for PG 18
pig install timescaledb -v 17;   # install for PG 17
pig install timescaledb -v 16;   # install for PG 16

Config this extension to shared_preload_libraries:

shared_preload_libraries = 'timescaledb';

Create this extension with:

CREATE EXTENSION timescaledb;

Usage

Sources:

• TimescaleDB v2.29.1 README
• TimescaleDB 2.29.0 release
• TimescaleDB 2.29.1 security and bug-fix release
• TimescaleDB v2.29.1 control file
• CREATE TABLE API
• create_hypertable() API
• Continuous aggregate API
• add_columnstore_policy() API
• TimescaleDB GUCs

timescaledb is a PostgreSQL extension for time-series and event analytics. The current docs emphasize CREATE TABLE ... WITH (tsdb.hypertable), continuous aggregates, automation jobs, and moving chunks into the columnstore.

Hypertables

CREATE EXTENSION timescaledb;

CREATE TABLE ts_test (
  ts timestamptz NOT NULL,
  id bigint,
  v integer
) WITH (
  tsdb.hypertable,
  tsdb.orderby = 'ts DESC'
);

To convert an existing PostgreSQL table, use the generalized hypertable API:

CREATE TABLE ts_existing (
  ts timestamptz NOT NULL,
  id bigint,
  v integer
);
SELECT create_hypertable('ts_existing', by_range('ts'));

• CREATE TABLE ... WITH (tsdb.hypertable) has been documented since TimescaleDB 2.20.0 and is the best-practice path for new hypertables.
• For TimescaleDB 2.23.0 and later, the first TIMESTAMP or TIMESTAMPTZ column is selected automatically as the partition column unless more than one candidate makes the choice ambiguous.
• create_hypertable() still works for converting existing tables.

Continuous aggregates and jobs

CREATE MATERIALIZED VIEW ts_hourly
WITH (timescaledb.continuous) AS
SELECT time_bucket('1 hour', ts) AS bucket,
       count(*) AS cnt,
       avg(v)   AS avg_v
FROM ts_test
GROUP BY bucket;

SELECT add_continuous_aggregate_policy(
  'ts_hourly',
  start_offset => INTERVAL '3 hours',
  end_offset => INTERVAL '1 hour',
  schedule_interval => INTERVAL '1 hour'
);

SELECT add_job('user_defined_action', '1h');

• Continuous aggregates require time_bucket(...) on the hypertable’s time dimension.
• The continuous aggregate WITH clause supports timescaledb.materialized_only; the current API default is TRUE, so real-time aggregation is not enabled unless configured otherwise.
• TimescaleDB 2.28.0 lets manual refresh_continuous_aggregate() calls run incrementally in batches. Use buckets_per_batch, max_batches_per_execution, and refresh_newest_first to break large manual refreshes into smaller work units.
• TimescaleDB 2.28.0 also allows adding a new generated aggregate column to an existing continuous aggregate with ALTER MATERIALIZED VIEW ... ADD COLUMN ... GENERATED ALWAYS AS (...) STORED; existing rows are NULL until refreshed.

Columnstore

CREATE TABLE crypto_ticks (
  "time" timestamptz,
  symbol text,
  price double precision,
  day_volume numeric
) WITH (
  tsdb.hypertable,
  tsdb.segmentby = 'symbol',
  tsdb.orderby = 'time DESC'
);

CALL add_columnstore_policy('crypto_ticks', after => INTERVAL '60 days');

• CREATE TABLE ... WITH (tsdb.hypertable) enables columnstore by default unless tsdb.columnstore = false.
• add_columnstore_policy() replaces the older add_compression_policy() API and requires either after or created_before, not both.
• Bloom filters are enabled by default for new columnstore chunks. Existing chunks need recompression before they have bloom indexes.

Relevant GUCs

SET timescaledb.enable_direct_compress_insert = on;
SET timescaledb.enable_cagg_rewrites = on;
SET timescaledb.enable_columnar_scan_filter_pushdown = on;

timescaledb.enable_direct_compress_insert and timescaledb.enable_direct_compress_copy enable tech-preview direct compression during ingestion. TimescaleDB 2.27.0 adds timescaledb.enable_cagg_rewrites and timescaledb.cagg_rewrites_debug_info, and documents timescaledb.enable_columnar_scan_filter_pushdown as enabled by default.

Version 2.29.1 and Caveats

• TimescaleDB 2.29 supports PostgreSQL 16, 17, and 18. PostgreSQL 15 support ended with the 2.28 line, so upgrade PostgreSQL before moving a PG15 database to 2.29.
• Version 2.29.0 adds compact_chunk() and a compaction policy for merging small columnstore batches, plus optimized DML chunk exclusion and small-LIMIT columnstore scans. Review the release notes before enabling compaction policies on existing workloads.
• The 2.29 line adds alter_job(..., config_merge => ...), direct-compression and unordered-recompression controls, and concurrent refresh policies for hierarchical continuous aggregates.
• Use 2.29.1 rather than 2.29.0. It fixes missing permission checks, malformed compressed-data handling, several crash paths, and validation of compact_chunk batch limits; upstream recommends upgrading all 2.29.0 installations.
• The control file marks timescaledb trusted and non-relocatable. The server library still has to be preloaded and PostgreSQL restarted according to the packaged deployment configuration.