Home

Understanding PostgreSQL locks

November 12, 2017 · Igor Šarčević

Locking is an important topic in any kind of database. Without properly handling locks, an application might not only be slow, it might also be wrong and behave in some insane ways. Therefore, learning proper locking techniques is essential for good performance and correctness of our applications.

Blind Monks Examining an Elephant

Let’s explore the types of locks available in PostgreSQL, when they are used, and how to explore open locks in the system.

Setting up a sandbox environment

I strongly believe that in order to learn something, you need to try it out with your own hands. Following that thought, here is an easy way to set up a sandbox environment for PostgreSQL testing.

Start by creating a test database a connect to it via psql:

$ createdb test-db-001

$ psql -d test-db-001
psql (10.0)
Type "help" for help.

test-db-001=#

In the database, create a users table with several records:

test-db-001=# CREATE TABLE users (id int, username text);
CREATE TABLE

test-db-001=# INSERT INTO users (id, username)
test-db-001=# VALUES (1, 'igor'), (2, 'bob'), (3, 'john'), (4, 'susan');
INSERT 0 4

test-db-001=# SELECT * FROM users;
 id | username
----+----------
  1 | igor
  2 | bob
  3 | john
  4 | susan
(4 rows)

I find a user table the best table for doing exercises, as it a relation that you will encounter in most most web applications.

Quit the connection to the database:

test-db-001=# \q

Test environment for learning about PostgreSQL locks

We will start two parallel connections to the database, one for starting and stopping transactions, and the other one for observing and listing locks that are created.

For the sake of simplicity, we will call these two connection Alice and Bob. Bob will create transactions and locks, while Alice will be our administrator that observes the state of the database.

Open two parallel terminal windows. In the first terminal, start a psql session a change the name of the prompt to bob:

$ psql -d test-db-001
psql (10.0)
Type "help" for help.

test-db-001=# \set PROMPT1 '(bob) # '
(bob) #

In the second terminal window, open a new psql session and name it alice:

$ psql -d test-db-001
psql (10.0)
Type "help" for help.

test-db-001=# \set PROMPT1 '(alice) # '
(alice) #

Everything is set up. We are ready to explore.

Exploring the pg_locks view

The pg_locks view provides access to information about the locks held by open transactions within the database server.

Alice, our administrator, will start by describing the pg_locks view and listing the available columns.

(alice) # \d pg_locks;
                   View "pg_catalog.pg_locks"
       Column       |   Type   | Collation | Nullable | Default
--------------------+----------+-----------+----------+---------
 locktype           | text     |           |          |
 database           | oid      |           |          |
 relation           | oid      |           |          |
 page               | integer  |           |          |
 tuple              | smallint |           |          |
 virtualxid         | text     |           |          |
 transactionid      | xid      |           |          |
 classid            | oid      |           |          |
 objid              | oid      |           |          |
 objsubid           | smallint |           |          |
 virtualtransaction | text     |           |          |
 pid                | integer  |           |          |
 mode               | text     |           |          |
 granted            | boolean  |           |          |
 fastpath           | boolean  |           |          |

Let’s look into open locks in the database with the select command:

(alice) # SELECT locktype, relation, mode, pid FROM pg_locks;

  locktype  | relation |      mode       |  pid
------------+----------+-----------------+-------
 relation   |    11577 | AccessShareLock | 16524
 virtualxid |          | ExclusiveLock   | 16524

(2 rows)

The 11577 relation is not descriptive enough. We will use ::regclass to cast the number to the name of the relation:

(alice) # SELECT locktype, relation::regclass, mode, pid FROM pg_locks;

  locktype  | relation |      mode       |  pid
------------+----------+-----------------+-------
 relation   | pg_locks | AccessShareLock | 16524
 virtualxid |          | ExclusiveLock   | 16524

(2 rows)

This is clearer. The open lock in the database is the result of the select statement that lists the locks. Let’s filter that lock out of the resulting list:

(alice) # SELECT locktype, relation::regclass, mode, pid
(alice) # FROM pg_locks WHERE pid != pg_backend_pid();

 locktype | relation | mode | pid
----------+----------+------+-----
(0 rows)

(alice) #

We used the pg_backend_pid() to filter out all locks that are created from Alice’s session. With that change, we can see that there are no open locks in the database.

Exploring lock modes

Bob, our user, steps in to the game. He will open some transactions to help us to learn about the various lock modes.

Bob will now open a transaction with BEGIN, and select all users. However, he will not close the transaction. This will help Alice to look into the lock that is implicitly created with Bob’s SELECT statement.

(bob) # BEGIN;
(bob) # SELECT * FROM users;

Alice now sweeps in and lists the locks:

(alice) # select locktype, relation::regclass, mode
(alice) # FROM pg_locks WHERE pid != pg_backend_pid();

  locktype  | relation |      mode
------------+----------+-----------------
 relation   | users    | AccessShareLock
 virtualxid |          | ExclusiveLock

(2 rows)

The SELECT statement creates a AccessShareLock. This is type of lock that is generally created by queries that read a table but do not modify it.

The AccessShareLock conflicts with the AccessExclusiveLock. That means that if another transactions puts a AccessExclusiveLock lock on the table, select statements will not work.

Bob will now end the open transaction, releasing the lock, and he will try to acquire a AccessExclusiveLock for the user table. Adding a new column to the table does just that, it locks up the table with exclusive access.

(bob) # END;

(bob) # BEGIN;
(bob) # ALTER TABLE users ADD age int;
(alice) # SELECT locktype, relation::regclass, mode, pid
(alice) # FROM pg_locks WHERE pid != pg_backend_pid();

   locktype    | relation |        mode         |  pid
---------------+----------+---------------------+-------
 virtualxid    |          | ExclusiveLock       | 16537
 transactionid |          | ExclusiveLock       | 16537
 relation      | users    | AccessExclusiveLock | 16537
(3 rows)

(alice) #

What happens if Alice tries to list the content of the users table at this moment?

(alice) # SELECT * FROM users;

The command never finishes. It waits for Bob’s AccessExclusiveLock to be released. We can hit CTRL+C to cancel the select statement:

(alice) # SELECT * FROM users;
^CCancel request sent
ERROR:  canceling statement due to user request

Let’s also stop Bob’s ALTER TABLE statement:

(bob) # ROLLBACK;

Lock Modes in PostgreSQL

We have learned about two locks modes so far. The AccessShareLock that is created for read queries like the select statements, and AccessExclusiveLock that is created for operations that modify the whole table.

There are several more lock modes in PostgreSQL.

ACCESS SHARE — Acquired by queries that only read from a table but do not modify it. Typically, this is a select query.

ROW SHARE — Acquired by the SELECT FOR UPDATE and SELECT FOR SHARE queries.

ROW EXCLUSIVE — Acquired by queries that modify the data in a table. Typically, update, delete, and insert queries.

SHARE UPDATE EXCLUSIVE — Acquired by vacuum, concurrent indexes, statistics, and some variants of the alter table commands. This mode protects a table against concurrent schema changes and vacuum runs.

SHARE — Acquired by create index that is not executed in concurrent mode. This mode protects a table against concurrent data changes.

SHARE ROW EXCLUSIVE — This mode protects a table against concurrent data changes, and is self-exclusive so that only one session can hold it at a time. Acquired by create collation, create trigger, and many forms of alter table.

EXCLUSIVE — This mode allows only concurrent ACCESS SHARE locks, i.e., only reads from the table can proceed in parallel with a transaction holding this lock mode.

ACCESS EXCLUSIVE — This mode guarantees that the holder is the only transaction accessing the table in any way. Acquired by DROP TABLE, ALTER TABLE, VACUUM FULl commands.

Explicit locking

In the previous sections, we have learned that all of the typical SQL commands acquire some sort of lock implicitly. We can also acquire locks explicitly with the WITH LOCK statement.

Let’s observe Bob as he acquires an explicit lock.

(bob) # begin;
(bob) # LOCK TABLE users;

When Alice lists open locks we will see that an AccessExclusiveLock lock was acquired for the users table. This is the default for the LOCK TABLE statement.

(alice) # SELECT locktype, relation::regclass, mode, pid
(alice) # FROM pg_locks WHERE pid != pg_backend_pid();

   locktype    | relation |        mode         |  pid
---------------+----------+---------------------+-------
 virtualxid    |          | ExclusiveLock       | 22546
 relation      | users    | AccessExclusiveLock | 22546
 transactionid |          | ExclusiveLock       | 22546
(3 rows)

Bob can acquire any lock mode with the WITH LOCK statement. For example, to acquire with a Share Update Exclusive mode he would enter:

(bob) # LOCK TABLE users IN SHARE UPDATE EXCLUSIVE MODE;

The locks are granted until the transaction ends.

Did you like this article? Or, do you maybe have a helpful hint to share? Please leave it in the comment section bellow.



Follow me on Twitter and LinkedIn.
Copyright © 2014-2024 Igor Šarčević. All rights reserved.