SNAPSHOT keyword

Prepares the database for a full backup or a filesystem (disk) snapshot.

note

Snapshot statements are not supported on Windows OS.

Syntax#

Flow chart showing the syntax of the SNAPSHOT keyword

Snapshot process#

Snapshot recovery mechanism requires a snapshot instance ID to be specified using the cairo.snapshot.instance.id configuration key:

server.conf
cairo.snapshot.instance.id=your_id

A snapshot instance ID may be an arbitrary string value, such as string representation of a UUID.

Database snapshots may be used in combination with filesystem snapshots or together with file copying (full backup). Collecting a snapshot involves the following steps:

  1. Run SNAPSHOT PREPARE statement to acquire reader locks for all database tables, create table metadata file copies in the snapshot directory, and flush the committed data to disk.
  2. Start a filesystem snapshot or copy the root directory to the backup location on the disk. Refer to the next section to learn how to create a filesystem snapshot on the most common cloud providers.
  3. Run SNAPSHOT COMPLETE statement to release the reader locks and delete the metadata file copies.

For some cloud vendors, filesystem snapshot creation operation is asynchronous, i.e. the point-in-time snapshot is created immediately, as soon as the operation starts, but the end snapshot artifact may become available later. In such case, the SNAPSHOT COMPLETE statement (step 3) may be run without waiting for the end artifact, but once the snapshot creation has started.

In case you prefer full backups over filesystem snapshots, you should keep in mind that the database will retain older partition and column file files on disk until SNAPSHOT COMPLETE. This means that you may run out of disk space if your disk doesn't have enough free space at the time you call SNAPSHOT PREPARE.

caution

No DDL statements, such as CREATE TABLE, should be run in parallel with the above steps. Otherwise, the snapshot may contain corrupted metadata making it unusable.

Filesystem snapshot#

The most common ways to perform cloud-native filesystem snapshots are described in the following resources, which rely on similar steps but have minor differences in terminology and services:

  • AWS - creating EBS snapshots
  • Azure - creating snapshots of a virtual hard disk
  • GCP - working with persistent disk snapshots

Snapshot recovery#

In case of a full backup, you should also delete the old root directory and copy the files from your backup to the same location or, alternatively, you can point the database at the new root directory.

To start the database on a filesystem snapshot, you should make sure to configure a different snapshot instance ID.

When the database starts, it checks the current instance ID and the ID stored in the snapshot directory, if present. On IDs mismatch, the database runs a snapshot recovery procedure restoring the metadata files from the snapshot. When this happens, you should see something like the following in the server logs:

2022-03-07T08:24:12.348004Z I i.q.g.DatabaseSnapshotAgent starting snapshot recovery [currentId=`id2`, previousId=`id1`]
...
2022-03-07T08:24:12.349922Z I i.q.g.DatabaseSnapshotAgent snapshot recovery finished [metaFilesCount=1, txnFilesCount=1, cvFilesCount=1]

Snapshot recovery can be disabled using the cairo.snapshot.recovery.enabled configuration key:

server.conf
cairo.snapshot.recovery.enabled=false

Examples#

SNAPSHOT PREPARE;
-- Start a filesystem snapshot.
SNAPSHOT COMPLETE;
SNAPSHOT PREPARE;
-- Copy the root directory:
-- $ cp -r /root/dir/path /backup/dir/path
SNAPSHOT COMPLETE;

โญ Something missing? Page not helpful? Please suggest an edit on GitHub.