Madhusoodan created PHOENIX-5583:
------------------------------------
Summary: Index tasks gettting submitted to a stopped task runner
Key: PHOENIX-5583
URL: https://issues.apache.org/jira/browse/PHOENIX-5583
Project: Phoenix
Issue Type: Bug
Affects Versions: 4.14.1, 4.14.0, 5.0.0-alpha
Environment: Linux (RHel 7.1)
Hbase Version 1.3.1
Phoenix version tried: 4.14.0 (Code hasn't changed in the later releases)
Reporter: Madhusoodan
Fix For: 5.1.0
In the org.apache.phoenix.hbase.index.parallel.BaseTaskRunner the falg
"stopped" is never set to true. Due to this, the tasks are getting submitted to
underlying ExecutorService even though it's stopped causing a
java.util.concurrent.RejectedExecutionException
The scenario is something like this
# A table is altered
# While the regions of the tables are getting closed, some clients are writing
to the table.
# Due to the closing the regions the taskrunner executor pool is stopped
# After this phoenix coprocessors are firing the index writes and these index
write tasks are getting submitted to the taskrunner causing the rejections.
# In 4.14.0 the condition worsens and causes the RS down scenarios as the
policy is to kill the RS if there is a index write failure.
I believe you won't ask for issue reproduction steps because the flaw is
evident from the code.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
