Jet Fault Tolerance

Giriş

Açıklaması şöyle

How Jet Fault Tolerance Works

• Regular backups (snapshots)

• Restart computation from last snapshot if topology changes

• Single node failures/additions - snapshot restored from replicated memory storage

Preconditions:

• Replayable source (e.g Kafka, Hazelcast Event Journal, JMS)

• Deterministic (no mutable lookup tables, no randomness)

• Idempotent sink (Upsert or distributed transactions: JMS, JDBC, Kafka)

Snapshot

Açıklaması şöyle. Yani Job state IMap üzerinde saklanır. Job çökerse en son yedek (snapshot) alınan yerden devam edebilir. Bu yöntem aynı zamanda Job'ların ölçeklendirilmesi için de kullanılır

In Hazelcast Jet 0.5, we introduced fault tolerance for streaming computations. Hazelcast Jet periodically takes snapshots of the state of a running job and stores these snapshots in Hazelcast IMaps. In case of a failure, the job is restarted from the last successful snapshot.