Flink checkpointing behavior

classic Classic list List threaded Threaded
2 messages Options
Reply | Threaded
Open this post in threaded view
|

Flink checkpointing behavior

amran dean
Hello,
Exact semantics for checkpointing/task recovery are still a little confusing to me after parsing docs: so a few questions.

- What does Flink consider a task failure? Is it any exception that the job does not handle?

- Do the failure recovery strategies mentioned in https://ci.apache.org/projects/flink/flink-docs-stable/dev/task_failure_recovery.html refer to restarting from the most recent checkpoint?
E.g for fixed-delay recoveries, a fixed number of restarts from a specific checkpoint are attempted.

- The docs mention the following command to resume from a checkpoint. In the checkpoint metadata path I have configured, I only see a series of directories named by hashes:
 
- 24c8d7a38dd90ca8bd5f04c36d1442ba
    - shared
    - taskowned
- 5d202a0ba04cdc1b917892c1e35d00dc
    - shared
    - taskowned
How do I know which is the most recent checkpoint?

Really appreciate any help. Thank you.
Reply | Threaded
Open this post in threaded view
|

Re: Flink checkpointing behavior

vino yang
Hi Amran,

See my inline answers.

Best,
Vino

amran dean <[hidden email]> 于2019年10月30日周三 上午2:59写道:
Hello,
Exact semantics for checkpointing/task recovery are still a little confusing to me after parsing docs: so a few questions.

- What does Flink consider a task failure? Is it any exception that the job does not handle?

Flink believes that the task failure is: any factor makes the task itself unable to continue to run. 

- Do the failure recovery strategies mentioned in https://ci.apache.org/projects/flink/flink-docs-stable/dev/task_failure_recovery.html refer to restarting from the most recent checkpoint?
E.g for fixed-delay recoveries, a fixed number of restarts from a specific checkpoint are attempted.

For an automatic restart, Flink will try to find the nearest checkpoint. 

- The docs mention the following command to resume from a checkpoint. In the checkpoint metadata path I have configured, I only see a series of directories named by hashes:
 
- 24c8d7a38dd90ca8bd5f04c36d1442ba
    - shared
    - taskowned
- 5d202a0ba04cdc1b917892c1e35d00dc
    - shared
    - taskowned
How do I know which is the most recent checkpoint?

In the checkpoint directory corresponding to the jobID, you should see some folder names, like "chk-xxx", so specify this path. More details please see here[1].


Really appreciate any help. Thank you.