Isolation Levels

• YSQL

Yugabyte supports three isolation levels in the transactional layer - Serializable, Snapshot and Read Committed. PostgreSQL (and the SQL standard) have four isolation levels - Serializable, Repeatable read, Read Committed and Read uncommitted. The mapping between the PostgreSQL isolation levels in YSQL, along with which transaction anomalies can occur at each isolation level are shown below.

^$ Read Committed Isolation is supported only if the tserver gflag yb_enable_read_committed_isolation is set to true. By default this gflag is false and in this case the Read Committed isolation level of Yugabyte's transactional layer falls back to the stricter Snapshot Isolation (in which case READ COMMITTED and READ UNCOMMITTED of YSQL also in turn use Snapshot Isolation). Read Committed support is currently in Beta.

As seen from the table above, the most strict isolation level is Serializable, which requires that any concurrent execution of a set of Serializable transactions is guaranteed to produce the same effect as running them in some serial (one transaction at a time) order. The other levels are defined by which anomalies must not occur as a result of interactions between concurrent transactions. Due to the definition of Serializable isolation, none of these anomalies are possible at that level. For reference, the various transaction anomalies are described briefly below:

• Dirty read: A transaction reads data written by a concurrent uncommitted transaction.

• Nonrepeatable read: A transaction re-reads data it has previously read and finds that data has been modified by another transaction (that committed since the initial read).

• Phantom read: A transaction re-executes a query returning a set of rows that satisfy a search condition and finds that the set of rows satisfying the condition has changed due to another recently-committed transaction.

• Serialization anomaly: The result of successfully committing a group of transactions is inconsistent with all possible orderings of running those transactions one at a time.

Let us now look at how Serializable, Snapshot and Read Committed isolation works in YSQL.

Serializable Isolation

The Serializable isolation level provides the strictest transaction isolation. This level emulates serial transaction execution for all committed transactions; as if transactions had been executed one after another, serially, rather than concurrently. Serializable isolation can detect read-write conflicts in addition to write-write conflicts. This is accomplished by writing provisional records for read operations as well.

Let's use a bank overdraft protection example to illustrate this case. The hypothetical case is that there is a bank which allows depositors to withdraw money up to the total of what they have in all accounts. The bank will later automatically transfer funds as needed to close the day with a positive balance in each account. Within a single transaction they check that the total of all accounts exceeds the amount requested.

Let's say someone tries to withdraw $900 from two of their accounts simultaneously, each with $500 balances. At the REPEATABLE READ transaction isolation level, that could work; but if the SERIALIZABLE transaction isolation level is used, a read/write conflict will be detected and one of the transactions will be rejected.

The example can be set up with these statements:

create table account
  (
    name text not null,
    type text not null,
    balance money not null default '0.00'::money,
    primary key (name, type)
  );
insert into account values
  ('kevin','saving', 500),
  ('kevin','checking', 500);

begin isolation level serializable;

select type, balance from account
  where name = 'kevin';

   type   | balance
----------+---------
 saving   | $500.00
 checking | $500.00
(2 rows)
    

begin isolation level serializable;

select type, balance from account
  where name = 'kevin';

   type   | balance
----------+---------
 saving   | $500.00
 checking | $500.00
(2 rows)
    

update account
  set balance = balance - 900::money
  where name = 'kevin' and type = 'saving';
    

update account
  set balance = balance - 900::money
  where name = 'kevin' and type = 'checking';
ERROR:  40001: Operation failed.
Try again.: Transaction aborted: XXXX

commit;

select type, balance from account
  where name = 'kevin';

   type   | balance
----------+----------
 checking |  $500.00
 saving   | -$400.00
(2 rows)
    

Snapshot Isolation

The Snapshot isolation level only sees data committed before the transaction began (or in other words, it works on a "snapshot" of the table). Transactions running under Snapshot isolation do not see either uncommitted data or changes committed during transaction execution by other concurrently running transactions. Note that the query does see the effects of previous updates executed within its own transaction, even though they are not yet committed. This is a stronger guarantee than is required by the SQL standard for the REPEATABLE READ isolation level.

Snapshot isolation detects only write-write conflicts, it does not detect read-write conflicts. In other words:

• INSERT, UPDATE, and DELETE commands behave the same as SELECT in terms of searching for target rows. They will only find target rows that were committed as of the transaction start time.
• If such a target row might have already been updated (or deleted or locked) by another concurrent transaction by the time it is found. This scenario is called a transaction conflict, where the current transaction conflicts with the transaction that made (or is attempting to make) an update. In such cases, one of the two transactions get aborted, depending on priority.

Let's run through the scenario below to understand how transactions behave under the snapshot isolation level (which PostgreSQL's Repeatable Read maps to).

First, create an example table with sample data.

CREATE TABLE IF NOT EXISTS example (k INT PRIMARY KEY);
TRUNCATE TABLE example;

Next, connect to the cluster using two independent ysqlsh instances called session #1 and session #2 below.

BEGIN TRANSACTION;
    

INSERT INTO example VALUES (1);

SELECT * FROM example;
 k
---
 1
(1 row)
    

INSERT INTO example VALUES (2);

SELECT * FROM example;
 k
---
 2
(1 row)
    

SELECT * FROM example;
 k
---
 1
(1 row)
    

COMMIT; 

SELECT * FROM example;
 k
---
 1
 2
(2 rows)
    

Read Committed Isolation

This is same as Snapshot Isolation except that every statement in the transaction will see all data that has been committed before it is issued (note that this implicitly also means that the statement will see a consistent snapshot). In other words, each statement works on a new "snapshot" of the database that includes everything that is committed before the statement is issued. Conflict detection is the same as in Snapshot Isolation.

CREATE TABLE test (k int PRIMARY KEY, v int);
INSERT INTO test VALUES (1, 2);
    

BEGIN TRANSACTION ISOLATION LEVEL READ COMMITTED;
SELECT * FROM test;

 k | v
---+---
 1 | 2
(1 row)

INSERT INTO test VALUES (2, 3);

SELECT * FROM test;
COMMIT;

 k | v
---+---
 1 | 2
(1 row)

BEGIN TRANSACTION ISOLATION LEVEL READ COMMITTED;
SELECT * FROM test;

 k | v
---+---
 1 | 2
 2 | 3
(2 rows)

INSERT INTO test VALUES (3, 4);

SELECT * FROM test;
 k | v
---+---
 1 | 2
 2 | 3
 3 | 4
(3 rows)