Mengidentifikasi dan Mengatasi Persaingan Kunci

Imagine a busy road. When multiple cars try to use the same lane at the same time, traffic slows down or stops. In PostgreSQL, this 'traffic jam' is called lock contention.

It happens when one transaction holds a lock on a resource (like a row or table) that another transaction needs. The second transaction then has to wait for the first one to release its lock.

Lock contention isn't just an inconvenience; it can severely impact your database's performance and application responsiveness. Here's how:

• Increased Query Latency: Queries take longer to complete.
• Reduced Throughput: The database processes fewer transactions per second.
• Application Timeouts: Frontend applications might time out waiting for a database response.
• Resource Waste: Waiting sessions consume server resources without making progress.

PostgreSQL provides built-in tools to help us spot contention. The pg_locks system view is your first stop. It shows all active locks held or awaited by backend processes.

Key columns to watch are pid (process ID), locktype, relation (the object being locked), mode (the type of lock), and especially granted.

A granted = false value in pg_locks indicates a session that is currently waiting for a lock. Let's see how to query for these waiting sessions:

Once you've identified a waiting session using pg_locks, the next step is to find out who is holding the lock and preventing it from proceeding. This is where pg_stat_activity comes in.

This view gives you details about all active sessions, including their current query, state, and when they started.

By combining information from pg_locks and pg_stat_activity, we can pinpoint blocking queries. Here's a simplified query to find active queries that might be causing contention:

Understanding the root causes helps in prevention:

• Long-Running Transactions: Transactions that hold locks for extended periods.
• Missing Indexes: Forgetting an index can lead to full table scans, acquiring more locks than necessary.
• DDL Operations: Commands like ALTER TABLE often require exclusive table locks.
• 'Hot Rows' / 'Hot Pages': Frequent updates or deletions on the same few rows or data blocks.

One of the most effective strategies is to keep your database transactions as short and efficient as possible. This means:

• Commit Frequently: Don't hold locks longer than needed.
• Batch Operations: Break down large operations into smaller, manageable chunks.
• Optimize Queries: Ensure SQL queries within transactions are highly optimized and use appropriate indexes.

Sometimes, waiting indefinitely isn't an option. PostgreSQL offers ways to manage this:

• SET lock_timeout: Prevents queries from waiting forever. The query will error out if it can't acquire a lock within the specified time.
• FOR UPDATE SKIP LOCKED: For specific use cases (like processing a queue), this clause allows a query to skip rows that are currently locked by other transactions, rather than waiting.

Great job! You've learned how to identify and begin resolving lock contention in PostgreSQL. We covered:

• What lock contention is and its performance impact.
• Using pg_locks and pg_stat_activity to find waiting sessions and their blockers.
• Common causes like long transactions and missing indexes.
• Key resolution strategies: shortening transactions, optimizing queries, and using lock_timeout or SKIP LOCKED.

Next, we'll dive deeper into advanced row-level locking strategies to further optimize concurrent writes!