timescaledb（timescaledb 开发）

TimescaleDB 2.24.0 Time-Series Database

Overview

TimescaleDB 2.24.0 introduces transformational features: lightning-fast recompression (100x faster updates), Direct Compress integration with continuous aggregates, UUIDv7 support in aggregates, and bloom filter sparse index changes. This skill ensures you leverage these capabilities correctly.

Core principle: Time-series data has unique access patterns. Design for append-heavy, time-range queries from the start.

Announce at start: "I'm applying timescaledb to ensure TimescaleDB 2.24.0 best practices."

When This Skill Applies

This skill is MANDATORY when ANY of these patterns are touched:

Or when files contain:

-- These patterns trigger this skill
create_hypertable
continuous aggregate
compress_chunk
add_compression_policy

TimescaleDB 2.24.0 Features

1. Lightning-Fast Recompression

TimescaleDB 2.24.0 introduces recompress:= true for dramatically faster updates to compressed data:

-- OLD (2.23 and earlier): Decompress entire chunk, update, recompress
-- Could take minutes for large chunks

-- NEW (2.24.0): Update compressed data directly
UPDATE sensor_data
SET value = corrected_value
WHERE time BETWEEN '2026-01-01' AND '2026-01-02';
-- 100x faster for compressed chunks

-- Enable recompression mode (automatic in 2.24.0)
-- Updates to compressed chunks now:
-- 1. Identify affected segments
-- 2. Decompress only those segments
-- 3. Apply updates
-- 4. Recompress immediately

-- Verify recompression is happening
SELECT * FROM timescaledb_information.job_stats
WHERE job_id IN (
  SELECT job_id FROM timescaledb_information.jobs
  WHERE proc_name = 'policy_recompression'
);

When this matters:

• Late-arriving data corrections
• Backfill operations
• Data quality fixes
• Retroactive updates

2. Direct Compress with Continuous Aggregates

Continuous aggregates can now compress directly without materialized hypertable overhead:

-- Create continuous aggregate with direct compression
CREATE MATERIALIZED VIEW hourly_metrics
WITH (timescaledb.continuous, timescaledb.compress = true) AS
SELECT
  time_bucket('1 hour', time) AS bucket,
  device_id,
  avg(temperature) AS avg_temp,
  min(temperature) AS min_temp,
  max(temperature) AS max_temp,
  count(*) AS sample_count
FROM sensor_readings
GROUP BY bucket, device_id
WITH NO DATA;

-- Add compression policy directly to continuous aggregate
SELECT add_compression_policy('hourly_metrics', INTERVAL '7 days');

-- Refresh policy
SELECT add_continuous_aggregate_policy('hourly_metrics',
  start_offset => INTERVAL '1 month',
  end_offset => INTERVAL '1 hour',
  schedule_interval => INTERVAL '1 hour'
);

Benefits:

• No intermediate materialized hypertable
• Automatic compression of aggregate data
• Reduced storage for historical aggregates
• Simpler management

3. UUIDv7 in Continuous Aggregates

TimescaleDB 2.24.0 supports PostgreSQL 18's native UUIDv7 in continuous aggregates:

-- Hypertable with UUIDv7 primary key (PostgreSQL 18)
CREATE TABLE events (
  id uuid DEFAULT uuidv7(),
  time timestamptz NOT NULL,
  event_type text NOT NULL,
  payload jsonb,
  PRIMARY KEY (id, time)
);

SELECT create_hypertable('events', 'time');

-- Continuous aggregate can now reference UUIDv7 columns
CREATE MATERIALIZED VIEW event_counts
WITH (timescaledb.continuous) AS
SELECT
  time_bucket('1 hour', time) AS bucket,
  event_type,
  count(*) AS event_count,
  count(DISTINCT id) AS unique_events  -- UUIDv7 works here now
FROM events
GROUP BY bucket, event_type
WITH NO DATA;

4. Bloom Filter Sparse Index Changes

TimescaleDB 2.24.0 modifies bloom filter behavior for sparse indexes:

-- Bloom filters for sparse data patterns
-- Useful for columns with many NULLs or low cardinality

CREATE TABLE logs (
  time timestamptz NOT NULL,
  level text,
  message text,
  error_code text,  -- Often NULL, sparse
  trace_id uuid     -- Often NULL, sparse
);

SELECT create_hypertable('logs', 'time');

-- Configure compression with bloom filter for sparse columns
ALTER TABLE logs SET (
  timescaledb.compress,
  timescaledb.compress_segmentby = 'level',
  timescaledb.compress_orderby = 'time DESC',
  -- Bloom filter helps find rare non-NULL values
  timescaledb.compress_bloomfilter = 'error_code, trace_id'
);

-- Query efficiency: Bloom filter skips segments without matches
SELECT * FROM logs
WHERE error_code = 'E500'
  AND time > now() - INTERVAL '1 day';
-- Scans only segments where bloom filter indicates possible match

When to use bloom filters:

• Sparse columns (many NULLs)
• Rare value queries (finding errors in logs)
• High-cardinality exact match queries
• NOT useful for range queries

Hypertable Design

Creating Hypertables

-- Standard time-series table
CREATE TABLE metrics (
  time timestamptz NOT NULL,
  device_id uuid NOT NULL,
  metric_name text NOT NULL,
  value double precision,
  metadata jsonb DEFAULT '{}'
);

-- Convert to hypertable
SELECT create_hypertable('metrics', 'time',
  chunk_time_interval => INTERVAL '1 day',  -- Chunk size
  create_default_indexes => true
);

-- With space partitioning (for high-cardinality dimensions)
SELECT create_hypertable('metrics', 'time',
  partitioning_column => 'device_id',
  number_partitions => 4,
  chunk_time_interval => INTERVAL '1 day'
);

Chunk Interval Selection

-- Adjust chunk interval
SELECT set_chunk_time_interval('metrics', INTERVAL '6 hours');

-- View current chunks
SELECT show_chunks('metrics', older_than => INTERVAL '1 day');

Primary Key Design

-- CORRECT: Time column in primary key for efficient chunk pruning
CREATE TABLE events (
  id uuid DEFAULT uuidv7(),
  time timestamptz NOT NULL,
  event_type text NOT NULL,
  PRIMARY KEY (id, time)  -- time included
);

-- WRONG: Time not in primary key (inefficient queries)
CREATE TABLE events_bad (
  id uuid PRIMARY KEY DEFAULT uuidv7(),
  time timestamptz NOT NULL  -- Not in PK
);

Compression Strategy

Enabling Compression

-- Configure compression
ALTER TABLE metrics SET (
  timescaledb.compress,
  timescaledb.compress_segmentby = 'device_id',      -- Group by this
  timescaledb.compress_orderby = 'time DESC',         -- Sort order
  timescaledb.compress_chunk_time_interval = '1 day'  -- Recompress interval
);

-- Manual compression
SELECT compress_chunk(c)
FROM show_chunks('metrics', older_than => INTERVAL '7 days') c;

-- Automatic compression policy
SELECT add_compression_policy('metrics', INTERVAL '7 days');

Segment By Selection

-- GOOD: Segment by commonly filtered dimension
-- Queries filter on device_id get excellent performance
ALTER TABLE metrics SET (
  timescaledb.compress_segmentby = 'device_id'
);

-- GOOD: Multiple segment columns for flexible queries
ALTER TABLE metrics SET (
  timescaledb.compress_segmentby = 'device_id, metric_name'
);

-- BAD: High cardinality segment (too many segments)
-- Don't segment by user_id if you have millions of users
ALTER TABLE events SET (
  timescaledb.compress_segmentby = 'user_id'  -- Too many segments!
);

-- BETTER for high cardinality: Include in orderby instead
ALTER TABLE events SET (
  timescaledb.compress_segmentby = 'event_type',
  timescaledb.compress_orderby = 'user_id, time DESC'
);

Order By Selection

-- Time descending for "most recent" queries
ALTER TABLE metrics SET (
  timescaledb.compress_orderby = 'time DESC'
);

-- Composite order for specific query patterns
ALTER TABLE logs SET (
  timescaledb.compress_orderby = 'level, time DESC'
);
-- Benefits: WHERE level = 'error' ORDER BY time DESC

-- Include frequently filtered columns
ALTER TABLE events SET (
  timescaledb.compress_orderby = 'device_id, time DESC'
);

Continuous Aggregates

Creating Aggregates

-- Basic continuous aggregate
CREATE MATERIALIZED VIEW hourly_stats
WITH (timescaledb.continuous) AS
SELECT
  time_bucket('1 hour', time) AS bucket,
  device_id,
  avg(value) AS avg_value,
  min(value) AS min_value,
  max(value) AS max_value,
  count(*) AS sample_count
FROM metrics
GROUP BY bucket, device_id
WITH NO DATA;

-- Hierarchical aggregates (aggregate of aggregate)
CREATE MATERIALIZED VIEW daily_stats
WITH (timescaledb.continuous) AS
SELECT
  time_bucket('1 day', bucket) AS bucket,
  device_id,
  avg(avg_value) AS avg_value,
  min(min_value) AS min_value,
  max(max_value) AS max_value,
  sum(sample_count) AS sample_count
FROM hourly_stats
GROUP BY 1, device_id
WITH NO DATA;

Refresh Policies

-- Add refresh policy
SELECT add_continuous_aggregate_policy('hourly_stats',
  start_offset => INTERVAL '3 days',   -- Refresh this far back
  end_offset => INTERVAL '1 hour',      -- Don't refresh latest (incomplete)
  schedule_interval => INTERVAL '1 hour'
);

-- Real-time aggregates (include unrefreshed data)
ALTER MATERIALIZED VIEW hourly_stats SET (
  timescaledb.materialized_only = false  -- Include real-time data
);

-- Force refresh
CALL refresh_continuous_aggregate('hourly_stats',
  '2026-01-01'::timestamptz,
  '2026-01-02'::timestamptz
);

With Compression (2.24.0)

-- Continuous aggregate with built-in compression
CREATE MATERIALIZED VIEW hourly_metrics
WITH (
  timescaledb.continuous,
  timescaledb.compress = true  -- New in 2.24.0
) AS
SELECT
  time_bucket('1 hour', time) AS bucket,
  device_id,
  avg(temperature) AS avg_temp,
  percentile_agg(temperature) AS temp_pct  -- For percentiles later
FROM sensor_readings
GROUP BY bucket, device_id
WITH NO DATA;

-- Add compression policy for the aggregate
SELECT add_compression_policy('hourly_metrics', INTERVAL '30 days');

-- Combined with refresh policy
SELECT add_continuous_aggregate_policy('hourly_metrics',
  start_offset => INTERVAL '7 days',
  end_offset => INTERVAL '1 hour',
  schedule_interval => INTERVAL '1 hour'
);

Retention Policies

Data Lifecycle

-- Drop old raw data (keep aggregates)
SELECT add_retention_policy('metrics', INTERVAL '90 days');

-- View retention policies
SELECT * FROM timescaledb_information.jobs
WHERE proc_name = 'policy_retention';

-- Remove retention policy
SELECT remove_retention_policy('metrics');

Tiered Storage Pattern

-- Pattern: Raw → Hourly → Daily → Archive

-- 1. Raw data: Keep 7 days uncompressed
-- 2. Raw data: Keep 30 days compressed
-- 3. Raw data: Drop after 90 days

-- 4. Hourly aggregates: Keep 1 year
-- 5. Daily aggregates: Keep forever

-- Implementation:
-- Raw data policies
SELECT add_compression_policy('metrics', INTERVAL '7 days');
SELECT add_retention_policy('metrics', INTERVAL '90 days');

-- Hourly aggregate policies
SELECT add_compression_policy('hourly_stats', INTERVAL '30 days');
SELECT add_retention_policy('hourly_stats', INTERVAL '1 year');

-- Daily stats: No retention (keep forever)
SELECT add_compression_policy('daily_stats', INTERVAL '90 days');

Query Patterns

Time Range Queries

-- Recent data (uses index)
SELECT * FROM metrics
WHERE time > now() - INTERVAL '1 hour'
  AND device_id = $1
ORDER BY time DESC
LIMIT 100;

-- Time range with aggregation
SELECT
  time_bucket('5 minutes', time) AS bucket,
  avg(value) AS avg_value
FROM metrics
WHERE time BETWEEN $1 AND $2
  AND device_id = $3
GROUP BY bucket
ORDER BY bucket;

-- Last value per device
SELECT DISTINCT ON (device_id)
  device_id,
  time,
  value
FROM metrics
WHERE time > now() - INTERVAL '1 day'
ORDER BY device_id, time DESC;

Using Continuous Aggregates

-- Query aggregate instead of raw data
SELECT * FROM hourly_stats
WHERE bucket > now() - INTERVAL '7 days'
  AND device_id = $1
ORDER BY bucket DESC;

-- Real-time aggregate (includes unrefreshed data)
SELECT * FROM hourly_stats
WHERE bucket > now() - INTERVAL '1 hour';
-- Automatically combines materialized + real-time data

Percentiles and Statistics

-- Use percentile_agg for continuous aggregates
CREATE MATERIALIZED VIEW metrics_percentiles
WITH (timescaledb.continuous) AS
SELECT
  time_bucket('1 hour', time) AS bucket,
  device_id,
  percentile_agg(value) AS value_pct,  -- Aggregate percentile state
  stats_agg(value) AS value_stats       -- Statistical aggregates
FROM metrics
GROUP BY bucket, device_id;

-- Query percentiles from aggregate
SELECT
  bucket,
  device_id,
  approx_percentile(0.50, value_pct) AS median,
  approx_percentile(0.95, value_pct) AS p95,
  approx_percentile(0.99, value_pct) AS p99,
  average(value_stats) AS avg,
  stddev(value_stats) AS stddev
FROM metrics_percentiles
WHERE bucket > now() - INTERVAL '24 hours';

Index Strategy

Default Indexes

-- create_hypertable creates this by default:
-- CREATE INDEX ON metrics (time DESC);

-- Add composite indexes for common queries
CREATE INDEX idx_metrics_device_time ON metrics (device_id, time DESC);

-- Partial indexes for specific patterns
CREATE INDEX idx_metrics_errors ON metrics (time DESC)
WHERE value > threshold;

Compressed Chunk Considerations

-- Indexes are not used on compressed chunks
-- Query planner uses:
-- 1. Chunk exclusion (time range)
-- 2. Segment filtering (compress_segmentby columns)
-- 3. Orderby optimization (compress_orderby columns)

-- Design compression settings for query patterns, not indexes
ALTER TABLE metrics SET (
  timescaledb.compress_segmentby = 'device_id',  -- Filter column
  timescaledb.compress_orderby = 'time DESC'      -- Sort column
);

Migration Patterns

Converting Regular Table to Hypertable

-- 1. Ensure time column exists and is NOT NULL
ALTER TABLE legacy_metrics ALTER COLUMN time SET NOT NULL;

-- 2. Convert to hypertable
SELECT create_hypertable('legacy_metrics', 'time',
  migrate_data => true,
  chunk_time_interval => INTERVAL '1 day'
);

-- 3. Add compression
ALTER TABLE legacy_metrics SET (
  timescaledb.compress,
  timescaledb.compress_segmentby = 'device_id',
  timescaledb.compress_orderby = 'time DESC'
);

-- 4. Add policies
SELECT add_compression_policy('legacy_metrics', INTERVAL '7 days');
SELECT add_retention_policy('legacy_metrics', INTERVAL '90 days');

Adding TimescaleDB to Existing Database

-- 1. Install extension
CREATE EXTENSION IF NOT EXISTS timescaledb;

-- 2. Verify version
SELECT extversion FROM pg_extension WHERE extname = 'timescaledb';
-- Should show 2.24.0

-- 3. Check PostgreSQL compatibility
SELECT timescaledb_information.version();

TimescaleDB Artifact

When implementing time-series features, post this artifact:

<!-- TIMESCALEDB_IMPLEMENTATION:START -->
## TimescaleDB Implementation Summary

### Hypertables

| Table | Chunk Interval | Space Partitions | Compression |
|-------|----------------|------------------|-------------|
| metrics | 1 day | device_id (4) | Yes |
| events | 6 hours | None | Yes |
| logs | 1 hour | level (2) | Yes |

### Compression Settings

| Table | Segment By | Order By | Bloom Filter |
|-------|------------|----------|--------------|
| metrics | device_id | time DESC | None |
| logs | level | time DESC | error_code, trace_id |

### Continuous Aggregates

| Aggregate | Source | Interval | Compression |
|-----------|--------|----------|-------------|
| hourly_metrics | metrics | 1 hour | Yes (30d) |
| daily_metrics | hourly_metrics | 1 day | Yes (90d) |

### Policies

| Table/Aggregate | Compression | Retention | Refresh |
|-----------------|-------------|-----------|---------|
| metrics | 7 days | 90 days | N/A |
| hourly_metrics | 30 days | 1 year | 1 hour |
| daily_metrics | 90 days | Never | 1 day |

### TimescaleDB 2.24.0 Features Used

- [ ] Lightning-fast recompression
- [ ] Direct Compress with continuous aggregates
- [ ] UUIDv7 in continuous aggregates
- [ ] Bloom filter sparse indexes

**TimescaleDB Version:** 2.24.0
**Verified At:** [timestamp]
<!-- TIMESCALEDB_IMPLEMENTATION:END -->

Checklist

Before completing TimescaleDB implementation:

• Hypertable created with appropriate chunk interval
• Compression configured with correct segmentby/orderby
• Compression policy added
• Retention policy added (if applicable)
• Continuous aggregates created for common queries
• Refresh policies configured
• Indexes appropriate for uncompressed chunks
• Query patterns tested with EXPLAIN ANALYZE
• 2.24.0 features leveraged where beneficial
• Artifact posted to issue

Integration

This skill integrates with:

• database-architecture - Hypertables follow general schema patterns
• postgres-rls - RLS works with hypertables (use caution with compression)
• postgis - Spatial time-series data

References

• TimescaleDB 2.24.0 Release Notes
• TimescaleDB Documentation
• Compression Best Practices
• Continuous Aggregates Guide