Stefan Egli created SLING-6945:
----------------------------------
Summary: avoid unnecessary error when OakViewChecker is deactivated
Key: SLING-6945
URL: https://issues.apache.org/jira/browse/SLING-6945
Project: Sling
Issue Type: Improvement
Components: Extensions
Affects Versions: Discovery Base 2.0.0
Reporter: Stefan Egli
Fix For: Discovery Base 2.0.2
The following series of log messages has been noticed:
{noformat}06.06.2017 11:47:41.774 *INFO* [Discovery-AsyncEventSender]
org.apache.sling.discovery.oak.pinger.OakViewChecker triggerAsyncConnectorPing:
firing job to trigger heartbeat
06.06.2017 11:47:41.811 *INFO* [OsgiInstallerImpl]
org.apache.sling.discovery.oak.pinger.OakViewChecker deactivate: deactivated
slingId: c2caefd0-7aab-4fbb-a120-985d6174d7ea, this:
org.apache.sling.discovery.oak.pinger.OakViewChecker@7a9f2f75
06.06.2017 11:47:41.812 *ERROR*
[sling-default-3-discovery.connectors.common.runner.c2caefd0-7aab-4fbb-a120-985d6174d7ea490b04a4-42f2-464a-ab15-15520e78e3a3]
org.apache.sling.discovery.oak.pinger.OakViewChecker announcementRegistry is
null
06.06.2017 11:47:45.339 *INFO* [OsgiInstallerImpl]
org.apache.sling.discovery.oak Service
[org.apache.sling.discovery.oak.pinger.OakViewChecker,10971,
[org.apache.sling.discovery.oak.pinger.OakViewChecker]] ServiceEvent
UNREGISTERING
06.06.2017 11:47:50.753 *INFO* [OsgiInstallerImpl]
org.apache.sling.discovery.oak Service
[org.apache.sling.discovery.oak.pinger.OakViewChecker,14874,
[org.apache.sling.discovery.oak.pinger.OakViewChecker]] ServiceEvent REGISTERED
06.06.2017 11:47:50.757 *INFO* [OsgiInstallerImpl]
org.apache.sling.discovery.oak.pinger.OakViewChecker doActivate: activated with
runtimeId: 82b0e439-6040-44d0-abd7-6c0cbbb62339, slingId:
c2caefd0-7aab-4fbb-a120-985d6174d7ea
{noformat}
Or in other words:
* {{11:47:41.774}} a topology event triggering a heartbeat (a normal,
low-level operation) happens
* {{11:47:41.811}} OakViewChecker deactivate
* {{11:47:41.812}} the heartbeat that was triggered before is now executed
asynchronously (via scheduler) and fails because deactivate already {{null}}-ed
the {{announcementRegistry}}
So this means {{OakViewChecker.deactivate}} didn't stop an already scheduled,
asynchronous heartbeat - which then subsequently fails because deactivate has
already happened.
Or, put slightly differently, {{deactivate}} and {{heartbeatAndCheckView}} are
not synchronized (due to SLING-3365).
As a simple measure {{BaseViewChecker.doCheckView}} should be made thread-safe,
ie it should be aware that deactivate can be called concurrently.
And the log.error should probably be lowered to log.info
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)