[
https://issues.apache.org/jira/browse/SENTRY-2491?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Na Li updated SENTRY-2491:
--------------------------
Status: Patch Available (was: Open)
> Sentry High availability unit tests run into deadlock sometimes
> ---------------------------------------------------------------
>
> Key: SENTRY-2491
> URL: https://issues.apache.org/jira/browse/SENTRY-2491
> Project: Sentry
> Issue Type: Bug
> Components: Sentry
> Affects Versions: 2.2.0
> Reporter: Na Li
> Assignee: Na Li
> Priority: Major
> Attachments: SENTRY-2491.001.patch
>
>
> In sentry unit tests, we don't create schema before running a test. Instead,
> we use dataNucleus to create sentry tables when they are accessed. This
> creates potential deadlock when running test for Sentry HA setup.
> For example, the following shows the event sequence that causes deadlock
> 1) thread_1 gets shared lock of SYSTABLES in order to read table
> SENTRY_HMS_NOTIFICATION_ID
> 2) thread_2 gets shared lock of SYSTABLES in order to read table
> SENTRY_HMS_NOTIFICATION_ID
> 3) thread_1 tries to get execution lock to create table
> SENTRY_HMS_NOTIFICATION_ID,
> and wait for execution lock because thread_2 got shared lock already.
> 4) thread_2 tries to get execution lock to create table
> SENTRY_HMS_NOTIFICATION_ID,
> and wait for execution lock because thread_1 got shared lock already.
> The solution is to let the instances of sentry service start with delay.
> Specifically,
> let HMS follower threads separate as far as possible, i.e., half of the
> interval.
>
> This deadlock only exists in unit tests, and does not exist in production
> because schema is created before starting Sentry services. Therefore, there
> is no table creation after service starts.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
