[
https://issues.apache.org/jira/browse/GOBBLIN-1921?focusedWorklogId=882978&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-882978
]
ASF GitHub Bot logged work on GOBBLIN-1921:
-------------------------------------------
Author: ASF GitHub Bot
Created on: 02/Oct/23 19:17
Start Date: 02/Oct/23 19:17
Worklog Time Spent: 10m
Work Description: phet commented on code in PR #3790:
URL: https://github.com/apache/gobblin/pull/3790#discussion_r1343044330
##########
gobblin-service/src/main/java/org/apache/gobblin/service/modules/orchestration/FlowTriggerHandler.java:
##########
@@ -278,6 +280,8 @@ public static JobDataMap updatePropsInJobDataMap(JobDataMap
jobDataMap,
// excess flows to be triggered by the reminder functionality.
prevJobProps.setProperty(ConfigurationKeys.SCHEDULER_PRESERVED_CONSENSUS_EVENT_TIME_MILLIS_KEY,
String.valueOf(leasedToAnotherStatus.getEventTimeMillis()));
+ // Use this boolean to indicate whether this is a reminder event
+ prevJobProps.setProperty(ConfigurationKeys.FLOW_IS_REMINDER_EVENT_KEY,
String.valueOf(true));
Review Comment:
please document in comment why defaulting to `true` is preferable. In
isolation, I'd naively imagine `false` to be what we want, but if the dynamic
code path to get here tells us to presume `true`, just highlight the reasoning
Issue Time Tracking
-------------------
Worklog Id: (was: 882978)
Time Spent: 0.5h (was: 20m)
> Properly handle reminder events
> -------------------------------
>
> Key: GOBBLIN-1921
> URL: https://issues.apache.org/jira/browse/GOBBLIN-1921
> Project: Apache Gobblin
> Issue Type: Bug
> Components: gobblin-service
> Reporter: Urmi Mustafi
> Assignee: Abhishek Tiwari
> Priority: Major
> Time Spent: 0.5h
> Remaining Estimate: 0h
>
> Reminder flow trigger events were being improperly handled and interpreted as
> new events because they are triggered {{linger}} time after the original
> trigger where {{epsilon < linger}} and we use {{epsilon}} to determine event
> distinctness. With reminder events being considered distinct events, we were
> launching excess concurrent flows that were then being cancelled. Now we
> handle reminder events differently from normal event triggers to ensure
> they're properly evaluated. Because of db laundering, reminder events are
> easy to handle - if they're older than the currently worked upon event in the
> database they can be skipped and if they're equal to the current event in the
> database they are handled like normal. Reminder events should never be newer
> than the current event in the lease arbiter table because db laundering
> always results in increasing event times.Â
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
