Create, Restore, and Delete Flink Checkpoints & Savepoints - Realtime Compute for Apache Flink

Checkpoint vs. savepoint

Concept	Who manages the lifecycle	Key behaviors	When to use
System checkpoint (Checkpoint in Apache Flink)	Flink system	Created and deleted automatically. Configure Checkpointing Interval, Checkpointing Timeout, and Min Interval Between Checkpoints. By default, one checkpoint is retained for recovery. Cannot be created or deleted manually.	Routine failure recovery during normal job operation.
Savepoint (Savepoint in Apache Flink)	User	Created manually or on a schedule. Deleted manually or by cleanup rules. Portable: can be shared between jobs with compatible state schemas.	Before upgrading job code, migrating to a new job, performing A/B testing, or any scenario requiring an explicit, user-controlled snapshot.

Operations in this topic

The following table lists the operations covered in this topic. Find the scenario that matches your goal to jump directly to the relevant section.

Operation	If you want to...
Manually create a savepoint	Capture job state at a specific point while the job is running, so you can restore from it later. Important A manually created savepoint can only be deleted manually—automatic cleanup rules do not apply to it.
Configure scheduled automatic creation of savepoints	Have the system create savepoints at regular intervals while the job runs, without manual intervention.
View the state generation overview	Review checkpoint and savepoint history, including status, completion time, and size.
Restore a job from a specified savepoint	Start a stopped job from a specific savepoint, including a savepoint from another job.
Configure automatic cleanup rules for savepoints	Limit OSS storage costs by automatically deleting old savepoints based on count or age.
Manually delete a specific savepoint	Delete a savepoint that automatic cleanup rules do not cover (for example, a manually created savepoint).
Set state-related parameters	Configure the state backend, state TTL for SQL jobs, or local disk storage for temporary savepoint data.

Manually create a savepoint

You can only create a savepoint while the job is in the RUNNING state.

Important

If you select Stop With Savepoint when canceling a deployment, less data needs to be reprocessed on restart, but cancellation takes longer.

Go to the savepoint creation page.
1. Log on to the Realtime Compute for Apache Flink console.
2. In the Actions column of the target workspace, click Console.
3. On the O&M > Deployments page, click the name of the target job.
In the upper-right corner of the page, click Create Savepoint.
Select a savepoint format and enter a description.

Click Create.

Configure scheduled automatic creation of savepoints

Go to the State page.
1. Log on to the Realtime Compute for Apache Flink console.
2. In the Actions column of the target workspace, click Console.
3. On the O&M > Deployments page, click the name of the target job.
4. Click State.
On the Configure tab, enable Savepoint automatically generated at regular intervals and set the schedule. Click Quick Setting to choose a preset interval, or enter a cron expression manually. To enter an expression manually, click each field and modify its value.

The minimum interval is 10 minutes. If you specify a shorter interval, the system uses 10 minutes.

Cron expression reference

The cron expression has five fields: minute hour day-of-month month day-of-week.

Field	Required	Value range	Supported special characters
Minute	Yes	[0, 59]	`*` `,` `-` `/`
Hour	Yes	[0, 23]	`*` `,` `-` `/`
Day	Yes	[1, 31]	`*` `,` `-` `/`
Month	Yes	[1, 12] or [JAN, DEC]	`*` `,` `-` `/`
Week	Yes	[1, 7] or [MON, SUN]	`*` `,` `-` `/`

In the [1, 7] format, 1 represents Monday and 7 represents Sunday.

Special characters:

Character	Meaning	Example
`*`	All possible values	`*` in Month means every month
`,`	Specific values	`5,20` in Minute triggers at minutes 5 and 20
`-`	Range	`5-20` in Minute triggers every minute from minute 5 to minute 20
`/`	Increment	`0/15` in Minute triggers every 15 minutes starting at minute 0; `3/20` triggers every 20 minutes starting at minute 3

Examples:

Expression	Schedule
`15 10 * * *`	10:15 AM every day
`0 12 * * *`	12:00 PM every day
`0 10,14,16 * * *`	10:00 AM, 2:00 PM, and 4:00 PM every day
`0/30 9-17 * * *`	Every 30 minutes from 9:00 AM to 5:00 PM every day
`* 14 * * *`	Every minute from 2:00 PM to 2:59 PM every day
`0-5 14 * * *`	Every minute from 2:00 PM to 2:05 PM every day
`0/5 14 * * *`	Every 5 minutes from 2:00 PM to 2:55 PM every day
`0/5 14,18 * * *`	Every 5 minutes from 2:00–2:55 PM and 6:00–6:55 PM every day
`0 12 * * WED`	12:00 PM every Wednesday
`15 10 15 * *`	10:15 AM on the 15th of every month
`10,44 14 * 3 WED`	2:10 PM and 2:44 PM every Wednesday in March

Important

Field order: minute hour day-of-month month day-of-week
day-of-month and day-of-week use OR logic: 0 12 1 * 5 triggers at 12:00 PM on the first day of each month or on every Friday—not only when the first day of the month falls on a Friday.
The `?` character is not supported. Use * to ignore a field.

View the state generation overview

Go to the State page.
1. Log on to the Realtime Compute for Apache Flink console.
2. In the Actions column of the target workspace, click Console.
3. On the O&M > Deployments page, click the name of the target job.
4. Click State.
On the Overview tab, review the following information.

Category	Information shown
System checkpoint	Last successful and failed checkpoint: ID, completion time, duration, and checkpoint size
Savepoint	Last successful and failed savepoint: ID, completion time, duration, and failure time

Restore a job from a specified savepoint

You can restore a job from any savepoint, including a savepoint created by a different job. To share a savepoint across jobs, the state schemas of both jobs must be compatible—for example, during A/B testing.

The job must be in the CANCELLED state before you can restore it from a savepoint. If the job is RUNNING, stop it first.

If you want to restart the job without restoring historical state, see Start a job for a stateless restart.

Go to the State page.
1. Log on to the Realtime Compute for Apache Flink console.
2. In the Actions column of the target workspace, click Console.
3. On the O&M > Deployments page, click the name of the target job.
4. Click State.
On the History tab, find the target savepoint in the Savepoints list.
In the Actions column for the target savepoint, select More > Start job from this savepoint.
In the Start Job dialog box, configure the startup settings. For details, see Start a job.

Configure automatic cleanup rules for savepoints

Automatic cleanup rules apply only to savepoints created by scheduled or automatic triggers. Manually created savepoints are never deleted by cleanup rules—delete them manually when no longer needed.

If a deployment is suspended for an extended period, savepoints are deleted based on the cleanup rules when the deployment next starts.

Go to the State page.
1. Log on to the Realtime Compute for Apache Flink console.
2. In the Actions column of the target workspace, click Console.
3. On the O&M > Deployments page, click the name of the target job.
4. Click State.
On the Configure tab, enable Scheduled automatic cleaning of savepoint and set the cleanup rules. Configure Number of historical savepoints saved and The maximum time to save historical savepoints. A savepoint is deleted when either condition is met.
Click Save Changes.

Manually delete a specific savepoint

Delete outdated savepoints periodically to avoid unnecessary Object Storage Service (OSS) charges. Manually created savepoints are not covered by automatic cleanup rules, so they must be deleted manually.

Go to the State page.
1. Log on to the Realtime Compute for Apache Flink console.
2. In the Actions column of the target workspace, click Console.
3. On the O&M > Deployments page, click the name of the target job.
4. Click State.
On the History tab, find the target savepoint in the Savepoints list.
In the Actions column for the target savepoint, select More > Delete this savepoint.
Click OK.

Set state-related parameters

Go to the O&M page of the target job.
1. Log on to the Realtime Compute for Apache Flink console.
2. In the Actions column of the target workspace, click Console.
3. On the O&M > Deployments page, click the target job.
On the Configuration tab, in the Parameters section, click Edit.
In Other Configuration, add the following parameters and click Save.
```
state.backend.incremental: true
table.exec.state.ttl: 129600000
```
For the full list of state-related parameters, see GeminiStateBackend configurations.

What's next

For common questions about checkpoints and savepoints, see Checkpoint FAQ.
For the impact of SQL changes on state compatibility, see SQL modifications and compatibility.