[
https://issues.apache.org/jira/browse/FLINK-18112?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Yuan Mei updated FLINK-18112:
-----------------------------
Description:
This is the Jira ticket for Milestone One of [FLIP-135 Approximate Task-Local
Recovery|[https://cwiki.apache.org/confluence/display/FLINK/FLIP-135+Approximate+Task-Local+Recovery].]
In short, in Approximate Task-Local Recovery, if a task fails, only the failed
task restarts without affecting the rest of the job. To ease discussion, we
divide the problem of approximate task-local recovery into three parts with
each part only focusing on addressing a set of problems. This Jira ticket
focuses on address the first milestone.
Milestone One: sink recovery. Here a sink task stands for no consumers reading
data from it. In this scenario, if a sink vertex fails, the sink is restarted
from the last successfully completed checkpoint and data loss is expected. If a
non-sink vertex fails, a regional failover strategy takes place. In milestone
one, we focus on issues related to task failure handling and upstream
reconnection.
Milestone one includes two parts of change:
*Part 1*: Network Part: how the failed task able to link to the upstream
Result(Sub)Partitions, and continue processing data
*Part 2*: Scheduling part, a new failover strategy to restart the sink only
when the sink fails.
was:
Build a prototype of single task failure recovery to address and answer the
following questions:
*Step 1*: Scheduling part, restart a single node without restarting the
upstream or downstream nodes.
*Step 2*: Checkpointing part, as my understanding of how regional failover
works, this part might not need modification.
*Step 3*: Network part
- how the recovered node able to link to the upstream ResultPartitions, and
continue getting data
- how the downstream node able to link to the recovered node, and continue
getting node
- how different netty transit mode affects the results
- what if the failed node buffered data pool is full
*Step 4*: Failover process verification
> Approximate Task-Local Recovery -- Milestone One
> ------------------------------------------------
>
> Key: FLINK-18112
> URL: https://issues.apache.org/jira/browse/FLINK-18112
> Project: Flink
> Issue Type: New Feature
> Components: Runtime / Checkpointing, Runtime / Coordination, Runtime
> / Network
> Affects Versions: 1.12.0
> Reporter: Yuan Mei
> Assignee: Yuan Mei
> Priority: Major
> Fix For: 1.12.0
>
>
> This is the Jira ticket for Milestone One of [FLIP-135 Approximate Task-Local
> Recovery|[https://cwiki.apache.org/confluence/display/FLINK/FLIP-135+Approximate+Task-Local+Recovery].]
> In short, in Approximate Task-Local Recovery, if a task fails, only the
> failed task restarts without affecting the rest of the job. To ease
> discussion, we divide the problem of approximate task-local recovery into
> three parts with each part only focusing on addressing a set of problems.
> This Jira ticket focuses on address the first milestone.
> Milestone One: sink recovery. Here a sink task stands for no consumers
> reading data from it. In this scenario, if a sink vertex fails, the sink is
> restarted from the last successfully completed checkpoint and data loss is
> expected. If a non-sink vertex fails, a regional failover strategy takes
> place. In milestone one, we focus on issues related to task failure handling
> and upstream reconnection.
>
> Milestone one includes two parts of change:
> *Part 1*: Network Part: how the failed task able to link to the upstream
> Result(Sub)Partitions, and continue processing data
> *Part 2*: Scheduling part, a new failover strategy to restart the sink only
> when the sink fails.
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
