[
https://issues.apache.org/jira/browse/CASSANDRA-21497?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
C. Scott Andreas updated CASSANDRA-21497:
-----------------------------------------
Description:
On trunk at {{{}1332e217{}}},
{{PlacementsChangeListenerTest#testPlacementChange}} fails intermittently with
"Conflicting replica added (expected unique endpoints)".
testPlacementChange builds random placements with
OwnershipUtils.randomPlacements, then for each range removes the first write
replica and adds a replica for endpoint .99 at that range.
randomPlacements assigns replicas via MembershipUtils.randomEndpoint, which
returns 127.0.0.[1..254]. Endpoint .99 is inside that domain, so a range's
write group can already contain it; removing a different replica and then
adding endpoint 99 produces a duplicate, and EndpointsForRange.Builder rejects
it.
The seed is System.nanoTime(), so the failure appears at a low rate. A search
over 200,000 seeds reproduced it on 2,447 of them (about 1.2%).
The current test adds a full replica at endpoint .99:
`Replica add = Replica.fullReplica(MembershipUtils.endpoint(99),
remove.range());`
Either excluding .99 from the domain selected for other replicas; or
restricting the domain such that a valid address exists above the domain could
be reasonable approaches to prevent the flake.
To reproduce this failure locally, seed OwnershipUtils#randomPlacements with
j.u.Random(135L):
{{DataPlacements before = OwnershipUtils.randomPlacements(new
java.util.Random(135L)).withLastModified(e);}}
Then run: {{ant test -Dtest.name=PlacementsChangeListenerTest}}
The test will fail with:
{{}}
{{testPlacementChange errors with java.lang.IllegalArgumentException:
Conflicting replica added (expected unique endpoints):
Full(/127.0.0.99:7000,...); existing: Full(/127.0.0.99:7000,...).}}
was:
On trunk at {{{}1332e217{}}},
{{PlacementsChangeListenerTest#testPlacementChange}} fails intermittently with
"Conflicting replica added (expected unique endpoints)".
testPlacementChange builds random placements with
OwnershipUtils.randomPlacements, then for each range removes the first write
replica and adds a replica for endpoint .99 at that range.
randomPlacements assigns replicas via MembershipUtils.randomEndpoint, which
returns 127.0.0.[1..254]. Endpoint .99 is inside that domain, so a range's
write group can already contain it; removing a different replica and then
adding endpoint 99 produces a duplicate, and EndpointsForRange.Builder rejects
it.
The seed is System.nanoTime(), so the failure appears at a low rate. A search
over 200,000 seeds reproduced it on 2,447 of them (about 1.2%).
The current test adds a full replica at endpoint .99:
`Replica add = Replica.fullReplica(MembershipUtils.endpoint(99),
remove.range());`
Either excluding .99 from the domain selected for other replicas; or
restricting the domain such that a valid address exists above the domain could
be reasonable approaches to prevent the flake.
To reproduce this failure locally, seed OwnershipUtils#randomPlacements with
j.u.Random(135L):
{{DataPlacements before = OwnershipUtils.randomPlacements(new
java.util.Random(135L)).withLastModified(e);}}
Then run: {{ant test -Dtest.name=PlacementsChangeListenerTest}}
The test will fail with:
{{testPlacementChange errors with java.lang.IllegalArgumentException:
Conflicting replica added (expected unique endpoints):
Full(/127.0.0.99:7000,...); existing: Full(/127.0.0.99:7000,...).}}
> Flaky test: Reproducible failing seed in
> PlacementsChangeListenerTest#testPlacementChange
> -----------------------------------------------------------------------------------------
>
> Key: CASSANDRA-21497
> URL: https://issues.apache.org/jira/browse/CASSANDRA-21497
> Project: Apache Cassandra
> Issue Type: Bug
> Components: Transactional Cluster Metadata
> Reporter: C. Scott Andreas
> Assignee: Sam Tunnicliffe
> Priority: Normal
>
> On trunk at {{{}1332e217{}}},
> {{PlacementsChangeListenerTest#testPlacementChange}} fails intermittently
> with "Conflicting replica added (expected unique endpoints)".
> testPlacementChange builds random placements with
> OwnershipUtils.randomPlacements, then for each range removes the first write
> replica and adds a replica for endpoint .99 at that range.
> randomPlacements assigns replicas via MembershipUtils.randomEndpoint, which
> returns 127.0.0.[1..254]. Endpoint .99 is inside that domain, so a range's
> write group can already contain it; removing a different replica and then
> adding endpoint 99 produces a duplicate, and EndpointsForRange.Builder
> rejects it.
> The seed is System.nanoTime(), so the failure appears at a low rate. A search
> over 200,000 seeds reproduced it on 2,447 of them (about 1.2%).
> The current test adds a full replica at endpoint .99:
> `Replica add = Replica.fullReplica(MembershipUtils.endpoint(99),
> remove.range());`
> Either excluding .99 from the domain selected for other replicas; or
> restricting the domain such that a valid address exists above the domain
> could be reasonable approaches to prevent the flake.
> To reproduce this failure locally, seed OwnershipUtils#randomPlacements with
> j.u.Random(135L):
> {{DataPlacements before = OwnershipUtils.randomPlacements(new
> java.util.Random(135L)).withLastModified(e);}}
> Then run: {{ant test -Dtest.name=PlacementsChangeListenerTest}}
> The test will fail with:
> {{}}
> {{testPlacementChange errors with java.lang.IllegalArgumentException:
> Conflicting replica added (expected unique endpoints):
> Full(/127.0.0.99:7000,...); existing: Full(/127.0.0.99:7000,...).}}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
