[SAMZA-905] Host Affinity - State restore doesn't work if the previous shutdown was uncontrolled (continuous offset)

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 0.10.1
• Assignee: Jake Maes
• Reporter: Jake Maes
• Created: 21/Mar/16 23:14
• Updated: 04/Apr/16 20:36
• Resolved: 04/Apr/16 20:36

Description

The host affinity feature writes an OFFSET file during shutdown which records the changelog offset for the persisted state store. Because the file is written at shutdown, it depends on a clean shutdown. Otherwise the file will not be there when the task restarts and the state will not be restored.

This ticket is to add a continuous checkpointing feature so the OFFSET file is written at regular intervals and the most recent update can be used to restore state.

Note: this feature assumes all changelog entries are idempotent as warned in https://issues.apache.org/jira/secure/attachment/12696372/DESIGN-SAMZA-516-2.pdf