Using Branches

Branches and database revisions allow you to work with your data at any commit in your database's commit graph. This is useful for isolating development on different branches, analyzing historical data, tracking data lineage, and much more.

Unlike other relational databases, Doltgres has multiple heads, one for each branch in the database. A head can be a branch, a tag, or a working set. Multiple clients can connect to each branch, and will see other writes to the same branch following the normal SQL transactional isolation semantics (REPEATABLE_READ). In effect, each branch functions as its own isolated database instance, with changes only visible to other clients connected to the same branch.

A Doltgres database server with multiple heads

A database server has a default branch, which is the checked-out branch at the time the server was started, and can be changed for new connections with a system variable. Using database revision specifiers, clients can choose a specific branch, tag, or commit to pin their queries to.

Specify a database revision in the connection string

The exact connection string you need to use will vary depending on your client.

To connect to the default branch, use a connection string with the name of the database only.

postgresql://localhost/mydb

To connect to a different branch, specify that branch name with a slash after the database name:

postgresql://localhost/mydb/feature-branch

To connect to a specific revision of the database, use a commit hash or tag instead of a branch name. The database will be read-only in this case.

postgresql://localhost/mydb/ia1ibijq8hq1llr7u85uivsi5lh3310ppostgresql://localhost/mydb/v1.0

You can also use the same ancestry syntax as Git to reference specific parent commits in a connection string, or anywhere else you would use a database revision specifier. For example,postgresql://localhost/mydb/feature-branch~2 will connect you to a read-only database for the grandparent commit of the feature-branch branch.

This also works with the standard psql command line client:

PGPASSWORD=password psql -h 127.0.0.1 "dbname=mydb/feature-branch" -U postgres

Switch heads by connecting to a different database

Different clients have different syntax for changing the connected database. In psql, use the \c command:

\c mydb/feature-branch

To switch to a read-only database at a commit hash or tag:

\c `mydb/ia1ibijq8hq1llr7u85uivsi5lh3310p`
\c `mydb/v1.0`

Use fully-qualified references with database revisions

You can also use fully-qualified names with database revisions in your queries. For example, the following query references a specific branch of a database by using a fully-qualified name that includes a revision specification:

insert into `mydatabase/feature-branch`.accounts (id) values (1);

You can use the same syntax for specific commits:

SELECT * from `mydatabase/ia1ibijq8hq1llr7u85uivsi5lh3310p`.accounts;

and for tags:

SELECT * from `mydatabase/v1.0`.accounts;

Switch branches with the DOLT_CHECKOUT() function

SELECT DOLT_CHECKOUT('feature-branch'); switches the session to thefeature-branch branch. You can also switch to a new branch, like so:

SELECT DOLT_CHECKOUT('-b', 'new-branch');

You can switch to a new branch with a starting commit as well:

SELECT DOLT_CHECKOUT('-b', 'new-branch-at-commit', 'ia1ibijq8hq1llr7u85uivsi5lh3310p')

Branches and transactions

The set of branches and their HEAD commits are established at transaction start time. Changes made to the set of branches or their HEAD commits in other transactions will not be visible to this session until a new transaction begins.

Recovering a deleted branch

The data on a branch is versioned, but the metadata of the branch head itself is not, so if you delete a branch or reset it to point at an older commit, you can't revert or undo that change the same way you can with your data. Instead, you can usethe dolt_reflog() table function to see the history of commits your branch has referenced and either recreate the branch from the last referenced commit withthe dolt_branch() function or reset the branch to a previous commit with the dolt_reset() function. Seethe dolt_reflog() table function for an example of recreating a deleted branch and more information on how the Doltgres reflog works and what limitations it has.

Restrictions on committing to multiple branches in a single transaction

The server will permit you to modify more than one branch in a single transaction, but will not permit such transactions to be committed. For example, the following sequence of statements will be rejected:

start transaction;
insert into `mydb/branch1`.t1 values (100);
insert into `mydb/branch2`.t1 values (200);
commit; -- ERROR: can only commit changes to one branch at a time

This restriction is true regardless of how you perform the modifications. This sequence also fails.

start transaction;
select dolt_checkout('branch1');
insert into t1 values (100);
select dolt_checkout('branch2');
insert into t1 values (200);
commit; -- ERROR: can only commit changes to one branch at a time

This restriction will be lifted in a future release of the database. For now you must ROLLBACK any transaction that modifies more than one branch.

Notes on unqualified database names

If you use a database name that isn't qualified by a branch or other revision specifier, it still resolves to a particular branch. The rules for this are subtle.

• If dolt_checkout() was called to switch the checked-out branch previously in this session, an unqualified database name will resolve to that branch. dolt_checkout() has the side-effect of changing what branch an unqualified database name resolves to for the remainder of a session.

• Otherwise, an unqualified database name resolves to the default branch, typically main.

An example:

\c mydb/branch1;
insert into t1 values (1); -- modifying the `branch1` branch
\c mydb/branch2;
insert into t1 values (2); -- modifying the `branch2` branch
use mydb;
insert into t1 values (3); -- modifying the `main` branch

In the last line, mydb resolves to mydb/main because no branch was checked out withdolt_checkout() in this session.

Using dolt_checkout() changes this behavior:

select dolt_checkout('branch1');
insert into t1 values (1); -- modifying the `branch1` branch
select dolt_checkout('branch2');
insert into t1 values (2); -- modifying the `branch2` branch
use mydb;
insert into t1 values (3); -- modifying the `branch2` branch

In the last line of this example, mydb resolves to mydb/branch2, because that was the branch last checked out with dolt_checkout().

Note that these name resolution rules apply to all statements that use an unqualified database name, not just USE. For example, insert into mydb.t1 values (4) will also modify the last checked-out branch, or the default branch, depending on session history.