[FLINK-12047] State Processor API (previously named Savepoint Connector) to read / write / process savepoints

• Type: New Feature
• Status: Open
• Priority: Not a Priority
• Resolution: Unresolved
• Component/s: API / State Processor
• Labels: auto-deprioritized-major auto-deprioritized-minor auto-unassigned
• Assignee: Unassigned
• Reporter: Tzu-Li (Gordon) Tai
• Created: 28/Mar/19 09:19
• Updated: 04/Apr/22 22:38

Description

This JIRA tracks the ongoing efforts and discussions about a means to read / write / process state in savepoints.

There are already two known existing works (that was mentioned already in the mailing lists) related to this:
1. Bravo [1]
2. https://github.com/sjwiesman/flink/tree/savepoint-connector

Essentially, the two tools both provide a connector to read or write a Flink savepoint, and allows to utilize Flink's processing APIs for querying / processing the state in the savepoint.

We should try to converge the efforts on this, and have a savepoint connector like this in Flink.
With this connector, the high-level benefits users should be able to achieve with it are:
1. Create savepoints using existing data from other systems (i.e. bootstrapping a Flink job's state with data in an external database).
2. Derive new state using existing state
3. Query state in savepoints, for example for debugging purposes
4. Migrate schema of state in savepoints offline, compared to the current more limited approach of online migration on state access.
5. Change max parallelism of jobs, or any other kind of fixed configuration, such as operator uids.

[1] https://github.com/king/bravo

Sub-Tasks

• Provide an easy way to modify max parallelism using the State Processor API: Open
• Extend State Processor API to read ListCheckpointed operator state: Open