[ 
https://issues.apache.org/jira/browse/SLING-13434?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Joerg Hoh updated SLING-13434:
------------------------------
    Description: 
*Summary*
A deterministic unit test reproduces stale topology properties after two 
PropertyProvider update callbacks interleave. This identifies a failure path in 
Discovery Standalone.
h2. Reproduction

{{updatedPropertyProvider}} calls {{unbindPropertyProvider(..., false)}} and 
then {{bindPropertyProvider(...)}} in separate synchronized sections. If two 
update callbacks interleave between those operations, the provider list can 
contain duplicate entries with the same service ID.

{{ProviderInfo}} snapshots properties when constructed, but equality is based 
on service ID. A later update can remove the newer duplicate while the older 
snapshot remains last; {{updatePropertiesCache()}} then preserves the stale 
value.

The test simulates that interleaving, changes the provider value from 
{{test/early}} to {{{}test/early,test/late{}}}, and verifies the topology 
property. Actual: {{{}test/early{}}}. Expected: {{{}test/early,test/late{}}}.
h2. Verification
 * {{mvn -Dtest=InterleavedPropertyProviderUpdateTest test}} — fails with the 
stale value, as expected.
 * {{mvn -Dtest=NoClusterDiscoveryServiceTest test}} — passes (6 tests).
 * {{mvn -DskipTests verify}} — passes.

The test deterministically models callback interleaving; it does not force 
concurrent OSGi/SCR threads or prove this was the production cause.
h2. Related PR

Draft test-only PR: 
[https://github.com/apache/sling-org-apache-sling-discovery-standalone/pull/3]

  was:
*Summary*
A deterministic unit test reproduces stale topology properties after two 
PropertyProvider update callbacks interleave. This identifies a failure path in 
Discovery Standalone; whether it caused the Helvetia incident remains 
unverified.
h2. Reproduction

{{updatedPropertyProvider}} calls {{unbindPropertyProvider(..., false)}} and 
then {{bindPropertyProvider(...)}} in separate synchronized sections. If two 
update callbacks interleave between those operations, the provider list can 
contain duplicate entries with the same service ID.

{{ProviderInfo}} snapshots properties when constructed, but equality is based 
on service ID. A later update can remove the newer duplicate while the older 
snapshot remains last; {{updatePropertiesCache()}} then preserves the stale 
value.

The test simulates that interleaving, changes the provider value from 
{{test/early}} to {{{}test/early,test/late{}}}, and verifies the topology 
property. Actual: {{{}test/early{}}}. Expected: {{{}test/early,test/late{}}}.
h2. Verification
 * {{mvn -Dtest=InterleavedPropertyProviderUpdateTest test}} — fails with the 
stale value, as expected.
 * {{mvn -Dtest=NoClusterDiscoveryServiceTest test}} — passes (6 tests).
 * {{mvn -DskipTests verify}} — passes.

The test deterministically models callback interleaving; it does not force 
concurrent OSGi/SCR threads or prove this was the production cause.
h2. Related PR

Draft test-only PR: 
[https://github.com/apache/sling-org-apache-sling-discovery-standalone/pull/3]


> NoClusterDiscoveryService can retain stale properties after interleaved 
> PropertyProvider updates
> ------------------------------------------------------------------------------------------------
>
>                 Key: SLING-13434
>                 URL: https://issues.apache.org/jira/browse/SLING-13434
>             Project: Sling
>          Issue Type: Bug
>            Reporter: Abhishek Garg
>            Priority: Major
>
> *Summary*
> A deterministic unit test reproduces stale topology properties after two 
> PropertyProvider update callbacks interleave. This identifies a failure path 
> in Discovery Standalone.
> h2. Reproduction
> {{updatedPropertyProvider}} calls {{unbindPropertyProvider(..., false)}} and 
> then {{bindPropertyProvider(...)}} in separate synchronized sections. If two 
> update callbacks interleave between those operations, the provider list can 
> contain duplicate entries with the same service ID.
> {{ProviderInfo}} snapshots properties when constructed, but equality is based 
> on service ID. A later update can remove the newer duplicate while the older 
> snapshot remains last; {{updatePropertiesCache()}} then preserves the stale 
> value.
> The test simulates that interleaving, changes the provider value from 
> {{test/early}} to {{{}test/early,test/late{}}}, and verifies the topology 
> property. Actual: {{{}test/early{}}}. Expected: {{{}test/early,test/late{}}}.
> h2. Verification
>  * {{mvn -Dtest=InterleavedPropertyProviderUpdateTest test}} — fails with the 
> stale value, as expected.
>  * {{mvn -Dtest=NoClusterDiscoveryServiceTest test}} — passes (6 tests).
>  * {{mvn -DskipTests verify}} — passes.
> The test deterministically models callback interleaving; it does not force 
> concurrent OSGi/SCR threads or prove this was the production cause.
> h2. Related PR
> Draft test-only PR: 
> [https://github.com/apache/sling-org-apache-sling-discovery-standalone/pull/3]



--
This message was sent by Atlassian Jira
(v10.3.26#10030026)

Reply via email to