Point-in-time recovery

SDS Point-in-time recovery provides a finer data backup and recovery mechanism than snapshots. After turning on this feature, you can restore the table status to any specified point in time. Point-in-time recovery uses Stream mechanism to open Point-in-time recovery table. Checkpoint for queue will be periodically created in background to record offset in queue at a certain point in time. You can specify a timestamp less than minimum checkpoint, and restore the table back to the state of this timestamp.

Item configurations

topicName: Queue name, used for opening Stream on this topic
snapshotPeriod: Snapshot cycle
ttl: Expiration time for Snapshot
createdTimestamp: Create timestamp
enablePointInTimeRecovery: On/off Switch
timeToArchive: The time snapshot is archived to fds

Use process

Create topic in Talos
When building table or updating table schema, input topic name to add and open Stream, and add and turn on Point-in-time recovery at the same time
Produce data in a table
Specify timestamp ts to restore table back to the state at ts
View existing snapshots of PointInTimeRecovery through pitrSnapshots in the result of listSnapshots() and find the closest snapshot that is less than ts. If snapshot has been archived to fds, contact us for recovery, or you can restore it through restoreSnapshot
Find the largest checkPoint that is less than the snapshot time via getFloorStreamCheckpoint()
Based on checkpoint, put the modified log of queue back to the table of snapshot recovery. In accordance with the timestamp in message, only put back modified log not larger than ts

Interface

listSnapshots(tableName) can be used to view existing snapshots of PointInTimeRecovery through pitrSnapshots of the returned value
getFloorStreamCheckpoint(ts) queries maximum stream checkpoint less than or equal to a timestamp
restoreSnapshot(tableName, snapshotName, destTableName, snapshotType) restores table. destTableName is the name of the restored table. For snapshotType, select PITR

Issues to be noted

Minimum snapshot cycle is 1 day, while maximum number is 5 (tentatively, and adjustable)
Time for snapshot to be archived to fds takes at least 1 month

Example

Example table</a>. The following is example code for modifying table

$tableName = "php-note";
$tableSpec = $adminClient->describeTable($tableName);

$tableSpec->metadata->pitr = new PointInTimeRecovery(array(
    'enablePointInTimeRecovery' => true,
    'topicName' => 'test-pirt-topic',
    'ttl' => 7776000,
    'snapshotPeriod' => 2592000,
    'timeToArchive' => 7776000
));

$adminClient->alterTable($tableName, $tableSpec);

Point-In-Time Recovery