PostgreSQL Performance & Query Optimization · レッスン

JSONBのGINインデックスと式インデックスの比較

クエリの形状に応じて、jsonb_path_opsのGINインデックスと対象を絞った式インデックスを使い分けます。

「JSONBのGINインデックスと式インデックスの比較」はCoddyKit上の無料PostgreSQL Performance & Query Optimizationレッスンです。これはレッスン2/4です。下記で完全なレッスンを無料で読むことができます。その後、ブラウザ内の組み込みコードエディタと24時間対応のAIチューターでハンズオン演習できます。これはPostgreSQL Performance & Query Optimization学習パスの一部であり、ウェブとCoddyKitアプリ全体で進捗が同期されます。 PostgreSQL Performance & Query Optimizationコースには全4レッスンが含まれています。

このレッスンの一部はまだ翻訳されておらず、英語で表示されています。

Two Ways to Index JSONB

When you store data in a jsonb column, an unindexed query forces PostgreSQL to read and parse every row. There are two very different tools to fix this:

GIN index — a general inverted index over the whole document, great for flexible containment and key/value lookups.
Expression (B-tree) index — a targeted index on one extracted scalar, great for a specific known query shape.

This lesson is about choosing the right one for your query patterns.

The Sample Table

Imagine an events table where each row carries a flexible JSON payload. We will index its data column.

Notice the payload mixes a few common keys (type, user_id) with arbitrary extras.

CREATE TABLE events (
  id     bigserial PRIMARY KEY,
  data   jsonb NOT NULL
);

INSERT INTO events (data) VALUES
  ('{"type": "login",  "user_id": 42, "ip": "10.0.0.1"}'),
  ('{"type": "logout", "user_id": 42}'),
  ('{"type": "login",  "user_id": 99, "mfa": true}');

The Default GIN: jsonb_ops

A plain GIN index uses the default jsonb_ops operator class. It indexes every key AND every value as separate entries.

This supports the widest set of operators: containment @>, key existence ?, ?|, and ?&.

The cost: it is larger on disk and slower to build/update because it stores far more entries per row.

CREATE INDEX idx_events_data_gin
  ON events USING gin (data);

-- Supports key existence AND containment:
-- WHERE data ? 'mfa'
-- WHERE data @> '{"type":"login"}'

The Leaner GIN: jsonb_path_ops

If you only ever use the containment operator @> (and the JSONPath operators @? / @@), prefer the jsonb_path_ops operator class.

It hashes whole key→value paths into single entries.
Result: noticeably smaller index and faster containment lookups.
Trade-off: it does NOT support the key-existence operators ?, ?|, ?&.

CREATE INDEX idx_events_data_pathops
  ON events USING gin (data jsonb_path_ops);

-- Great for:
SELECT id FROM events
WHERE data @> '{"type": "login"}';

How Containment Uses the GIN Index

The @> operator asks "does the left document contain the right one?" Both GIN operator classes accelerate it.

Containment is structural: it matches nested keys and values, not just top-level ones. This is why a single GIN index can serve many different filter combinations.

-- Match by one key:
SELECT * FROM events WHERE data @> '{"user_id": 42}';

-- Match by two keys at once (same index):
SELECT * FROM events
WHERE data @> '{"type": "login", "user_id": 42}';

-- Match a nested shape:
SELECT * FROM events WHERE data @> '{"flags": {"beta": true}}';

When GIN Falls Short: Range & Sort

GIN is built for equality-style containment. It canNOT help with:

Range comparisons on an extracted value (>, <, BETWEEN).
Ordering by a JSON field (ORDER BY ... LIMIT).
Prefix / pattern matching on a text value.

For these shapes you want a B-tree, and on JSONB that means an expression index.

-- GIN can't accelerate this range filter on an inner number:
SELECT * FROM events
WHERE (data ->> 'user_id')::int > 50
ORDER BY (data ->> 'user_id')::int
LIMIT 10;

Building an Expression Index

An expression index stores the result of an expression, not the raw column. You extract one scalar from the JSON and index that as a normal B-tree.

Two operators matter here:

-> returns jsonb.
->> returns text — usually what you cast and index.

-- B-tree on user_id extracted as an integer:
CREATE INDEX idx_events_user_id
  ON events (((data ->> 'user_id')::int));

-- Now ranges, sorts and equality all use it:
SELECT * FROM events
WHERE (data ->> 'user_id')::int BETWEEN 40 AND 99
ORDER BY (data ->> 'user_id')::int;

Match the Index Expression Exactly

The planner only uses an expression index when the query expression matches the indexed expression token for token, including the cast.

If you index (data ->> 'user_id')::int but query (data ->> 'user_id') as plain text, the index is ignored.

Keep the extraction + cast identical everywhere.

-- Indexed expression:
--   ((data ->> 'user_id')::int)

-- USES the index:
WHERE (data ->> 'user_id')::int = 42

-- IGNORES the index (text vs int mismatch):
WHERE (data ->> 'user_id') = '42'

Reading EXPLAIN to Confirm

Never guess which index wins — ask the planner. Use EXPLAIN (ANALYZE, BUFFERS) and look at the node type:

Bitmap Heap Scan + Bitmap Index Scan on ...gin → your GIN index is serving containment.
Index Scan / Index Only Scan on the expression index → your B-tree is serving the range/sort.
Seq Scan → nothing matched; revisit the expression or operator.

EXPLAIN (ANALYZE, BUFFERS)
SELECT * FROM events
WHERE data @> '{"type": "login"}';

EXPLAIN (ANALYZE, BUFFERS)
SELECT * FROM events
WHERE (data ->> 'user_id')::int = 42;

Partial Expression Indexes

If queries only ever target a subset of rows, add a WHERE clause to the index. A partial expression index is smaller and cheaper to maintain because it only stores the rows you actually search.

Here we index user_id only for login events — perfect when that is the only query shape that needs it.

CREATE INDEX idx_events_login_user
  ON events (((data ->> 'user_id')::int))
  WHERE data @> '{"type": "login"}';

Choosing: A Quick Decision Guide

Pick by the shape of your queries, not by habit:

Flexible filters on many different keys, or key-existence (?) → GIN jsonb_ops.
Only containment @> / JSONPath, want it lean and fast → GIN jsonb_path_ops.
One known field with ranges, sorting, or equality on a scalar → expression B-tree index.
That field queried on a narrow slice of rows → partial expression index.

It is common and correct to keep BOTH a GIN and one or two expression indexes on the same column.

Quick Check

Test your understanding of the GIN vs expression decision.

Recap

You learned to choose JSONB indexes by query shape:

GIN jsonb_ops — widest operator support including key existence ?; largest.
GIN jsonb_path_ops — leaner and faster, containment @> and JSONPath only.
Expression B-tree — one extracted, casted scalar for ranges, sorts, and equality; the query expression must match the index expression exactly.
Partial expression index — same idea, scoped to a row subset for a smaller footprint.

Always confirm with EXPLAIN (ANALYZE, BUFFERS), and don't hesitate to keep a GIN and one or two expression indexes side by side.

よくある質問

「JSONBのGINインデックスと式インデックスの比較」レッスンは無料ですか？

はい。「JSONBのGINインデックスと式インデックスの比較」の完全なテキストはこのウェブで無料で読めます。インタラクティブに演習し（組み込みコードエディタと24時間対応のAIチューター）、PostgreSQL Performance & Query Optimizationコースの残りをアンロックするには、CoddyKit PROにアップグレードしてください。 PostgreSQL Performance & Query Optimizationコースには全4レッスンが含まれています。

「JSONBのGINインデックスと式インデックスの比較」で何を学びますか？

クエリの形状に応じて、jsonb_path_opsのGINインデックスと対象を絞った式インデックスを使い分けます。ブラウザで直接実行するハンズオンコードでPostgreSQL Performance & Query Optimizationを演習し、24時間対応のAIチューターがレッスンを進める中での質問に答えます。

PostgreSQL Performance & Query Optimizationを始めるのに経験は必要ですか？

事前経験は必要ありません。CoddyKitのPostgreSQL Performance & Query Optimizationは初級者から上級者向けに構成されているため、ここから始めるか最初から始めて、自分のペースで進むことができます。これはレッスン2/4です。

「JSONBのGINインデックスと式インデックスの比較」レッスンにはどのくらい時間がかかりますか？

ほとんどのCoddyKitレッスンは約5～10分かかります。各レッスンはコンパクトでインタラクティブなので、着実に進歩し、ウェブとアプリ全体で正確に前回の場所から再開できます。

このPostgreSQL Performance & Query Optimizationレッスンでコードを書いて実行できますか？

はい。すべてのPostgreSQL Performance & Query Optimizationレッスンに組み込みコードエディタが含まれているため、ブラウザでリアルコードを書いて実行し、即座のAIフィードバックを取得できます。ローカル設定は不要です。

このコースのすべてのレッスン

← PostgreSQL Performance & Query Optimizationに戻る