[
https://issues.apache.org/jira/browse/GOBBLIN-1509?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Kip Kohn updated GOBBLIN-1509:
------------------------------
Summary: Ensure flows transition to FAILED and not stuck in COMPILED upon
DagManager::addDag error (was: Ensure flows not stuck in COMPILED upon
DagManager::addDag error)
> Ensure flows transition to FAILED and not stuck in COMPILED upon
> DagManager::addDag error
> -----------------------------------------------------------------------------------------
>
> Key: GOBBLIN-1509
> URL: https://issues.apache.org/jira/browse/GOBBLIN-1509
> Project: Apache Gobblin
> Issue Type: Bug
> Reporter: Kip Kohn
> Priority: Major
>
> Presently, addDag failure leaves the flow marooned in the COMPILED state, as
> the warranted FLOW_FAILED event is never sent. Particularly insidious is
> that scheduled flows with their execution stuck in COMPILED miss their next
> execution, unless `flow.allowConcurrentExecutions` is set. Thus the
> scheduled flow is stuck in its entirety, not merely a single execution.
> One observed cause of addDag failure is when the DagStateStore is backed by a
> replicated DB (e.g. MySqlDagStateStore) that just switched leaders. Cached
> connections in the pool may suddently point to a read-only follower unable to
> DagStateStore::writeCheckpoint.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
