Transactions

CockroachDB supports bundling multiple SQL statements into a single all-or-nothing transaction. Each transaction guarantees ACID semantics spanning arbitrary tables and rows, even when data is distributed. If a transaction succeeds, all mutations are applied together with virtual simultaneity. If any part of a transaction fails, the entire transaction is aborted, and the database is left unchanged. CockroachDB guarantees that while a transaction is pending, it is isolated from other concurrent transactions with serializable isolation.

For a detailed discussion of CockroachDB transaction semantics, see How CockroachDB Does Distributed Atomic Transactions and Serializable, Lockless, Distributed: Isolation in CockroachDB. The explanation of the transaction model described in this blog post is slightly out of date. See the Transaction Retries section for more details.

SQL statements

The following SQL statements control transactions.

Statement Description
BEGIN Initiate a transaction, as well as control its priority.
SET TRANSACTION Control a transaction's priority.
COMMIT Commit a regular transaction, or clear the connection after committing a transaction using the advanced retry protocol.
ROLLBACK Abort a transaction and roll the database back to its state before the transaction began.
SHOW Display the current transaction settings.
SAVEPOINT Used for nested transactions; also used to implement advanced client-side transaction retries.
RELEASE SAVEPOINT Commit a nested transaction; also used for retryable transactions.
ROLLBACK TO SAVEPOINT Roll back a nested transaction; also used to handle retryable transaction errors.
Note:

If you are using a framework or library that does not have advanced retry logic built in, you should implement an application-level retry loop with exponential backoff. See Client-side intervention.

Syntax

In CockroachDB, a transaction is set up by surrounding SQL statements with the BEGIN and COMMIT statements.

To use advanced client-side transaction retries, you should also include the SAVEPOINT, ROLLBACK TO SAVEPOINT and RELEASE SAVEPOINT statements.

icon/buttons/copy
> BEGIN;

> SAVEPOINT cockroach_restart;

<transaction statements>

> RELEASE SAVEPOINT cockroach_restart;

> COMMIT;

At any time before it's committed, you can abort the transaction by executing the ROLLBACK statement.

Clients using transactions must also include logic to handle retries.

Error handling

To handle errors in transactions, you should check for the following types of server-side errors:

Type Description
Retry Errors Errors with the code 40001 or string retry transaction, which indicate that a transaction failed because it could not be placed in a serializable ordering of transactions by CockroachDB. This is often due to contention: conflicts with another concurrent or recent transaction accessing the same data. In such cases, the transaction needs to be retried by the client as described in client-side intervention. For a reference listing all of the retry error codes emitted by CockroachDB, see the Transaction Retry Error Reference.
Ambiguous Errors Errors with the code 40003 which indicate that the state of the transaction is ambiguous, i.e., you cannot assume it either committed or failed. How you handle these errors depends on how you want to resolve the ambiguity. For information about how to handle ambiguous errors, see here.
SQL Errors All other errors, which indicate that a statement in the transaction failed. For example, violating the UNIQUE constraint generates a 23505 error. After encountering these errors, you can either issue a COMMIT or ROLLBACK to abort the transaction and revert the database to its state before the transaction began.

If you want to attempt the same set of statements again, you must begin a completely new transaction.

Transaction retries

Transactions may require retries if they experience deadlock or read/write contention with other concurrent transactions which cannot be resolved without allowing potential serializable anomalies.

To mitigate read-write contention and reduce the need for transaction retries, use the following techniques:

  • Perform reads using AS OF SYSTEM TIME.
  • Use SELECT FOR UPDATE to order transactions by controlling concurrent access to one or more rows of a table. This reduces retries in scenarios where a transaction performs a read and then updates the same row it just read.

There are two cases in which transaction retries can occur:

Automatic retries

CockroachDB automatically retries individual statements (implicit transactions) and transactions sent from the client as a single batch, as long as the size of the results being produced for the client, including protocol overhead, is less than 16KiB by default. Once that buffer overflows, CockroachDB starts streaming results back to the client, at which point automatic retries cannot be performed any more. As long as the results of a single statement or batch of statements are known to stay clear of this limit, the client does not need to worry about transaction retries.

You can change the results buffer size for all new sessions using the sql.defaults.results_buffer.size cluster setting, or for a specific session using the results_buffer_size session variable. Decreasing the buffer size can increase the number of transaction retry errors a client receives, whereas increasing the buffer size can increase the delay until the client receives the first result row.

Individual statements

Individual statements are treated as implicit transactions, and so they fall under the rules described above. If the results are small enough, they will be automatically retried. In particular, INSERT/UPDATE/DELETE statements without a RETURNING clause are guaranteed to have minuscule result sizes. For example, the following statement would be automatically retried by CockroachDB:

> DELETE FROM customers WHERE id = 1;

Batched statements

Transactions can be sent from the client as a single batch. Batching implies that CockroachDB receives multiple statements without being asked to return results in between them; instead, CockroachDB returns results after executing all of the statements, except when the accumulated results overflow the buffer mentioned above, in which case they are returned sooner and automatic retries can no longer be performed.

Batching is generally controlled by your driver or client's behavior. Technically, it can be achieved in two ways, both supporting automatic retries:

  1. When the client/driver is using the PostgreSQL Extended Query protocol, a batch is made up of all queries sent in between two Sync messages. Many drivers support such batches through explicit batching constructs.

  2. When the client/driver is using the PostgreSQL Simple Query protocol, a batch is made up of semicolon-separated strings sent as a unit to CockroachDB. For example, in Go, this code would send a single batch (which would be automatically retried):

    db.Exec(
      "BEGIN;
    
      DELETE FROM customers WHERE id = 1;
    
      DELETE orders WHERE customer = 1;
    
      COMMIT;"
    )
    

Within a batch of statements, CockroachDB infers that the statements are not conditional on the results of previous statements, so it can retry all of them. Of course, if the transaction relies on conditional logic (e.g., statement 2 is executed only for some results of statement 1), then the transaction cannot be all sent to CockroachDB as a single batch. In these common cases, CockroachDB cannot retry, say, statement 2 in isolation. Since results for statement 1 have already been delivered to the client by the time statement 2 is forcing the transaction to retry, the client needs to be involved in retrying the whole transaction and so you should write your transactions to use client-side intervention.

Bounded staleness reads

In the event bounded staleness reads are used along with either the with_min_timestamp function or the with_max_staleness function and the nearest_only parameter is set to true, the query will throw an error if it can't be served by a nearby replica.

Client-side intervention

Your application should include client-side retry handling when the statements are sent individually, such as:

icon/buttons/copy
> BEGIN;

> UPDATE products SET inventory = 0 WHERE sku = '8675309';

> INSERT INTO orders (customer, status) VALUES (1, 'new');

> COMMIT;

To indicate that a transaction must be retried, CockroachDB signals an error with the SQLSTATE error code 40001 (serialization error) and an error message that begins with the string "restart transaction". For a complete list of transaction retry error codes, see Transaction retry error reference.

To handle these types of errors, you have the following options:

Client-side intervention example

The Python-like pseudocode below shows how to implement an application-level retry loop; it does not require your driver or ORM to implement advanced retry handling logic, so it can be used from any programming language or environment. In particular, your retry loop must:

  • Raise an error if the max_retries limit is reached
  • Retry on 40001 error codes
  • COMMIT at the end of the try block
  • Implement exponential backoff logic as shown below for best performance
while true:
    n++
    if n == max_retries:
        throw Error("did not succeed within N retries")
    try:
        # add logic here to run all your statements
        conn.exec('COMMIT')
        break
    catch error:
        if error.code != "40001":
            throw error
        else:
            # This is a retry error, so we roll back the current transaction
            # and sleep for a bit before retrying. The sleep time increases
            # for each failed transaction.  Adapted from
            # https://colintemple.com/2017/03/java-exponential-backoff/
            conn.exec('ROLLBACK');
            sleep_ms = int(((2**n) * 100) + rand( 100 - 1 ) + 1)
            sleep(sleep_ms) # Assumes your sleep() takes milliseconds

Testing transaction retry logic

New in v22.1: To test your transaction retry logic, use the inject_retry_errors_enabled session variable. When inject_retry_errors_enabled is set to true, any statement (with the exception of SET statements) executed in the session inside of an explicit transaction will return a transaction retry error with the message restart transaction: TransactionRetryWithProtoRefreshError: injected by `inject_retry_errors_enabled` session variable.

If the client retries the transaction using the special cockroach_restart SAVEPOINT name, after the 3rd retry, the transaction will proceed as normal. Otherwise, the errors will continue until the client issues a SET inject_retry_errors_enabled=false statement.

For example, suppose you've written a wrapper function with some retry logic that you want to use to execute statements across a psycopg2 connection:

def run_transaction(conn, op, max_retries=3):    
    """
    Execute the operation *op(conn)* retrying serialization failures.

    If the database returns an error asking to retry the transaction, retry it
    *max_retries* times before giving up (and propagate it).
    """
    # leaving this block the transaction will commit or rollback
    # (if leaving with an exception)
    with conn:
        for retry in range(1, max_retries + 1):
            try:
                op(conn)
                # If we reach this point, we were able to commit, so we break
                # from the retry loop.
                return

            except SerializationFailure as e:
                # This is a retry error, so we roll back the current
                # transaction and sleep for a bit before retrying. The
                # sleep time increases for each failed transaction.
                logging.debug("got error: %s", e)
                conn.rollback()
                sleep_ms = (2 ** retry) * 0.1 * (random.random() + 0.5)
                logging.debug("Sleeping %s seconds", sleep_ms)
                time.sleep(sleep_ms)

            except psycopg2.Error as e:
                logging.debug("got error: %s", e)
                raise e

        raise ValueError(f"Transaction did not succeed after {max_retries} retries")

run_transaction takes a SQL-executing function op(conn) and attempts to run the function, retrying on serialization failures (exposed in pscyopg2 as the SerializationFailure exception class) with exponential backoff, until reaching a maximum number of tries.

You can add a quick test to this function using the inject_retry_errors_enabled session variable.

def run_transaction(conn, op=None, max_retries=3):
    """
    Execute the operation *op(conn)* retrying serialization failure.
    If no op is specified, the function runs a test, using the
    inject_retry_errors_enabled session variable to inject errors.

    If the database returns an error asking to retry the transaction, retry it
    *max_retries* times before giving up (and propagate it).
    """
    # leaving this block the transaction will commit or rollback
    # (if leaving with an exception)
    with conn:
        for retry in range(1, max_retries + 1):
            try:
                if not op:
                    with conn.cursor() as cur:
                        if retry == 1:
                            cur.execute("SET inject_retry_errors_enabled = 'true'")
                        if retry == max_retries:
                            cur.execute("SET inject_retry_errors_enabled = 'false'")
                        cur.execute("SELECT now()")
                        logging.debug("status message: %s", cur.statusmessage)
                else:
                    op(conn)
                # If we reach this point, we were able to commit, so we break
                # from the retry loop.
                return

            except SerializationFailure as e:
                # This is a retry error, so we roll back the current
                # transaction and sleep for a bit before retrying. The
                # sleep time increases for each failed transaction.
                logging.debug("got error: %s", e)
                conn.rollback()
                sleep_ms = (2 ** retry) * 0.1 * (random.random() + 0.5)
                logging.debug("Sleeping %s seconds", sleep_ms)
                time.sleep(sleep_ms)

            except psycopg2.Error as e:
                logging.debug("got error: %s", e)
                raise e

        raise ValueError(f"Transaction did not succeed after {max_retries} retries")

Calling run_transaction without an op input sets inject_retry_errors_enabled as true until the final retry attempt, before which the inject_retry_errors_enabled is set back to false. For all attempts except the last one, CockroachDB will inject a retryable serialization error for the client to handle. If the client cannot handle the error properly, the retry logic isn't working properly.

Transaction contention

Transactions in CockroachDB lock data resources that are written during their execution. When a pending write from one transaction conflicts with a write of a concurrent transaction, the concurrent transaction must wait for the earlier transaction to complete before proceeding. When a dependency cycle is detected between transactions, the transaction with the higher priority aborts the dependent transaction to avoid deadlock, which must be retried.

For more details about transaction contention and best practices for avoiding contention, see Transaction Contention.

Nested transactions

CockroachDB supports the nesting of transactions using savepoints. These nested transactions are also known as sub-transactions. Nested transactions can be rolled back without discarding the state of the entire surrounding transaction.

This can be useful in applications that abstract database access using an application development framework or ORM. Different components of the application can operate on different sub-transactions without having to know about each others' internal operations, while trusting that the database will maintain isolation between sub-transactions and preserve data integrity.

Just as COMMIT and ROLLBACK are used to commit and discard entire transactions, respectively, RELEASE SAVEPOINT and ROLLBACK TO SAVEPOINT are used to commit and discard nested transactions. This relationship is shown in the following table:

Statement Effect
COMMIT Commit an entire transaction.
ROLLBACK Discard an entire transaction.
RELEASE SAVEPOINT Commit (really, forget) the named nested transaction.
ROLLBACK TO SAVEPOINT Discard the changes in the named nested transaction.

For more information, including examples showing how to use savepoints to create nested transactions, see the savepoints documentation.

Transaction priorities

Every transaction in CockroachDB is assigned an initial priority. By default, the transaction priority is NORMAL.

Set transaction priority

For transactions that should be given higher or lower preference in high-contention scenarios, you can set the priority in the BEGIN statement:

> BEGIN PRIORITY <LOW | NORMAL | HIGH>;

You can also set the priority immediately after a transaction is started:

> SET TRANSACTION PRIORITY <LOW | NORMAL | HIGH>;

To set the default transaction priority for all transactions in a session, use the default_transaction_priority session variable. For example:

> SET default_transaction_priority = 'high';
Note:

When two transactions contend for the same resources indirectly, they may create a dependency cycle leading to a deadlock situation, where both transactions are waiting on the other to finish. In these cases, CockroachDB allows the transaction with higher priority to abort the other, which must then retry. On retry, the transaction inherits the higher priority. This means that each retry makes a transaction more likely to succeed in the event it again experiences deadlock.

View transaction priority

transaction_priority is a read-only session variable.

To view the current priority of a transaction, use SHOW transaction_priority or SHOW TRANSACTION PRIORITY:

> SHOW transaction_priority;
  transaction_priority
------------------------
  high
> SHOW TRANSACTION PRIORITY;
  transaction_priority
------------------------
  high

Isolation levels

CockroachDB executes all transactions at the strongest ANSI transaction isolation level: SERIALIZABLE. All other ANSI transaction isolation levels (e.g., SNAPSHOT, READ UNCOMMITTED, READ COMMITTED, and REPEATABLE READ) are automatically upgraded to SERIALIZABLE. Weaker isolation levels have historically been used to maximize transaction throughput. However, ACIDRain: Concurrency-Related Attacks on Database-Backed Web Applications has demonstrated that the use of weak isolation levels results in substantial vulnerability to concurrency-based attacks.

For a detailed discussion of isolation in CockroachDB transactions, see Serializable, Lockless, Distributed: Isolation in CockroachDB.

Serializable isolation

With SERIALIZABLE isolation, a transaction behaves as though it has the entire database all to itself for the duration of its execution. This means that no concurrent writers can affect the transaction unless they commit before it starts, and no concurrent readers can be affected by the transaction until it has successfully committed. This is the strongest level of isolation provided by CockroachDB and it's the default.

SERIALIZABLE isolation permits no anomalies. To prevent write skew anomalies, SERIALIZABLE isolation may require transaction restarts. For a demonstration of SERIALIZABLE preventing write skew, see Serializable Transactions.

Comparison to ANSI SQL isolation levels

CockroachDB uses slightly different isolation levels than ANSI SQL isolation levels.

Aliases

SNAPSHOT, READ UNCOMMITTED, READ COMMITTED, and REPEATABLE READ are aliases for SERIALIZABLE.

Comparison

The CockroachDB SERIALIZABLE level is stronger than the ANSI SQL READ UNCOMMITTED, READ COMMITTED, and REPEATABLE READ levels and equivalent to the ANSI SQL SERIALIZABLE level.

For more information about the relationship between these levels, see A Critique of ANSI SQL Isolation Levels.

Limit the number of rows written or read in a transaction

You can limit the number of rows written or read in a transaction at the cluster or session level. This allows you configure CockroachDB to log or reject statements that could destabilize a cluster or violate application best practices.

Use the cluster settings sql.defaults.transaction_rows_written_log, sql.defaults.transaction_rows_written_err, sql.defaults.transaction_rows_read_log, and sql.defaults.transaction_rows_read_err and session settings transaction_rows_written_log, transaction_rows_written_err, transaction_rows_read_log, and transaction_rows_read_err to limit the number of rows written or read in a transaction. When the log limit is reached, the transaction is logged to the SQL_PERF channel. When the err limit is reached, the transaction is rejected. The limits are enforced after each statement of a transaction has been fully executed.

The "write" limits apply to INSERT, INSERT INTO SELECT FROM, INSERT ON CONFLICT, UPSERT, UPDATE, and DELETE SQL statements. The "read" limits apply to the SELECT statement in addition to the statements subject to the "write" limits. The limits do not apply to CREATE TABLE AS, SELECT, IMPORT, TRUNCATE, DROP, ALTER TABLE, BACKUP, RESTORE, or CREATE STATISTICS statements.

Note:

Enabling transaction_rows_read_err disables a performance optimization for mutation statements in implicit transactions where CockroachDB can auto-commit without additional network round trips.

See also

YesYes NoNo