ratacat/postgres-query-expert

PostgreSQL Query Expert

This skill is a definitive reference for PostgreSQL 16, covering query construction, optimization, schema management, and system introspection.

Instructions

1. General Query Standards

• Syntax: Adhere to ANSI SQL standards, but prefer PostgreSQL extensions (e.g., DISTINCT ON, RETURNING, LATERAL, FILTER clauses) when they provide cleaner logic or better performance.
• Identifiers: Use snake_case for all identifiers. Only quote identifiers ("MyTable") if absolutely necessary; prefer lowercase unquoted names.
• Safety:
  • Parameterization: Always use parameters ($1, $2, …) for literal values. Never inject user input directly.
  • Timeouts: For exploratory queries on large databases, prepend SET LOCAL statement_timeout = '30s';.
  • Transactions: Use explicit BEGIN and COMMIT blocks for multi-step operations.

2. Performance & Optimization

• Explain plans: Use EXPLAIN (ANALYZE, COSTS, VERBOSE, BUFFERS) to diagnose bottlenecks.
• Red flags: Seq Scan on large tables, high Buffers: shared hit (RAM usage), or Disk: read (I/O).
• Indexing: Recommend specific index types based on usage:
  • B-tree: Standard equality/range (=, <, >) queries.
  • GIN: For composite types like JSONB (@>) or arrays (&&), and full-text search.
  • GiST: For geometric data and ranges.
• CTEs: Use Common Table Expressions (WITH) for readability. In PG16+, these are optimized (inlined) by default unless MATERIALIZED is specified.

Introspection (Agent Capabilities)

When exploring a new database, use these queries to understand the schema.

List All Tables

SELECT n.nspname AS schema,
       c.relname AS table,
       obj_description(c.oid) AS description
FROM pg_class c
JOIN pg_namespace n ON n.oid = c.relnamespace
WHERE c.relkind = 'r'
  AND n.nspname NOT IN ('pg_catalog', 'information_schema')
ORDER BY 1, 2;

Get Table Columns & Types

SELECT a.attname AS column,
       format_type(a.atttypid, a.atttypmod) AS type,
       a.attnotnull AS not_null,
       col_description(a.attrelid, a.attnum) AS comment
FROM pg_attribute a
WHERE a.attrelid = 'public.target_table_name'::regclass
  AND a.attnum > 0
  AND NOT a.attisdropped
ORDER BY a.attnum;

Reference: Data Querying (DQL)

Advanced Aggregations

• Filter clause: count(*) FILTER (WHERE status = 'active')
• Grouping sets: GROUP BY GROUPING SETS ((brand), (brand, category), ())
• Any value: any_value(col) (PG16+) returns an arbitrary value from the group.

Window Functions

Perform calculations across a set of table rows related to the current row.

SELECT dept,
       emp_no,
       salary,
       -- Rank employees by salary within department
       dense_rank() OVER (PARTITION BY dept ORDER BY salary DESC) AS rank,
       -- Running total of salaries
       sum(salary) OVER (
         PARTITION BY dept
         ORDER BY salary
         ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW
       ) AS running_total
FROM employees;

Pattern Matching

• LIKE: col LIKE 'foo%' (simple wildcard).
• ILIKE: col ILIKE 'foo%' (case-insensitive).
• SIMILAR TO: col SIMILAR TO '[a-c]%' (SQL-regex style).
• POSIX regex:
  • Case-sensitive: col ~ '^[a-z]+$'
  • Case-insensitive: col ~* 'foo'

Reference: Data Modification (DML)

MERGE (Upsert / Conditional Ops)

Standard SQL method for inserting, updating, or deleting based on join conditions (PG15+).

MERGE INTO wine_stock ws
USING wine_shipments s
  ON s.winery_id = ws.winery_id
 AND s.year = ws.year
WHEN MATCHED THEN
  UPDATE SET stock = ws.stock + s.count
WHEN NOT MATCHED THEN
  INSERT (winery_id, year, stock)
  VALUES (s.winery_id, s.year, s.count);

INSERT ... ON CONFLICT (Legacy Upsert)

Postgres-specific, often more concise for simple unique-key conflicts.

INSERT INTO kv_store (key, value)
VALUES ('config', '{"a":1}')
ON CONFLICT (key)
DO UPDATE SET value = EXCLUDED.value;

RETURNING Clause

Return data from modified rows immediately.

DELETE FROM archived_logs
WHERE created_at < NOW() - INTERVAL '1 year'
RETURNING id, created_at;

Reference: Special Data Types

JSONB (Binary JSON)

Prefer jsonb over json for storage and indexing.

| Operator | Description | Example | |---|---|---| | -> / ->> | Get element (JSON / text) | data->'key' | | @> | Contains (indexable) | data @> '{"tag": "urgent"}' | | ? | Key exists | data ? 'error' | | #- | Delete path | data #- '{info, sensitive}' |

SQL/JSON path (PG12+):

-- Find all items with price > 10
SELECT jsonb_path_query(data, '$.items[*] ? (@.price > 10)')
FROM orders;

Arrays

SELECT ARRAY[1,2,3];           -- Creation
SELECT (ARRAY[1,2,3])[1];      -- Access (1-based index)
SELECT 1 = ANY(arr_col);       -- Check if value exists in array
SELECT unnest(arr_col) FROM t; -- Expand array to rows

Range Types

Useful for scheduling and validity periods.

• tstzrange: timestamp with time zone range.
• int4range, daterange: integer and date ranges.
• Overlap operator (&&): checks if two ranges overlap.

SELECT *
FROM reservations
WHERE duration && tstzrange('2023-01-01 10:00', '2023-01-01 12:00');

Reference: System Administration & Stats

Kill Long-Running Query

SELECT pg_terminate_backend(pid)
FROM pg_stat_activity
WHERE state = 'active'
  AND pid <> pg_backend_pid()
  AND query_start < NOW() - INTERVAL '5 minutes';

Check Table Size (Disk Usage)

SELECT relname,
       pg_size_pretty(pg_total_relation_size(relid)) AS total_size,
       pg_size_pretty(pg_relation_size(relid)) AS data_size
FROM pg_catalog.pg_statio_user_tables
ORDER BY pg_total_relation_size(relid) DESC;

Examples

Scenario 1: Recursive CTE for Graph/Tree Data

Navigating an organizational hierarchy.

WITH RECURSIVE subordinates AS (
    -- Base case: the manager
    SELECT employee_id, manager_id, full_name, 0 AS level
    FROM employees
    WHERE employee_id = $1

    UNION ALL

    -- Recursive step: direct reports
    SELECT e.employee_id, e.manager_id, e.full_name, s.level + 1
    FROM employees e
    INNER JOIN subordinates s ON s.employee_id = e.manager_id
)
SELECT *
FROM subordinates;

Scenario 2: Lateral Join for "Top N per Category"

Efficiently getting the latest 3 posts for each user.

SELECT u.username, p.title, p.created_at
FROM users u
CROSS JOIN LATERAL (
    SELECT title, created_at
    FROM posts
    WHERE user_id = u.id
    ORDER BY created_at DESC
    LIMIT 3
) p
WHERE u.status = 'active';

Scenario 3: Full Text Search with Ranking

Searching a blog table.

SELECT id,
       title,
       ts_rank(to_tsvector('english', title || ' ' || content), query) AS rank
FROM articles,
     to_tsquery('english', 'postgres | optimization') query
WHERE to_tsvector('english', title || ' ' || content) @@ query
ORDER BY rank DESC;