[jira] [Updated] (HBASE-4863) Make HBase Thrift server more configurable and add a command-line UI test

["Phabricator (Updated) (JIRA)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/HBASE-4863?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Phabricator updated HBASE-4863:
-------------------------------

    Attachment: D531.1.patch

mbautin requested code review of "[jira] [HBASE-4863] Make HBase Thrift server more configurable and add a command-line UI test".
Reviewers: JIRA, Kannan, tedyu, stack

  This started as an internal hotfix where we found out that the Thrift server spawned 15000 threads. To bound the thread pool size I added a custom thread pool server implementation called HBaseThreadPoolServer into HBase codebase, and made the following parameters configurable from both command line and as config settings: minWorkerThreads, maxWorkerThreads, and maxQueuedRequests. Under an increasing load, the server creates new threads for every connection before the pool size reaches minWorkerThreads. After that, the server puts new connections into the queue and only creates a new thread when the queue is full. If an attempt to create a new thread fails, the server drops connection. The default TThreadPoolServer would crash in that case, but it never happened because the thread pool was unbounded, so the server would hang indefinitely, consume a lot of memory, and cause huge latency spikes on the client side.

  Another part of this fix is refactoring and unit testing of the command-line part of the Thrift server. The logic there is sufficiently complicated, and the existing ThriftServer class does not test that part at all. The new TestThriftServerCmdLine test starts the Thrift server on a random port with various combinations of options and talks to it through the client API from another thread.


TEST PLAN
  Unit tests, cluster test with a Python Thrift client.
  I will post an update when I'm done with testing.

REVISION DETAIL
  https://reviews.facebook.net/D531

AFFECTED FILES
  src/main/java/org/apache/hadoop/hbase/thrift/HBaseThreadPoolServer.java
  src/main/java/org/apache/hadoop/hbase/thrift/ThriftServer.java
  src/main/java/org/apache/hadoop/hbase/util/Threads.java
  src/test/java/org/apache/hadoop/hbase/HBaseTestingUtility.java
  src/test/java/org/apache/hadoop/hbase/thrift/TestThriftServer.java
  src/test/java/org/apache/hadoop/hbase/thrift/TestThriftServerCmdLine.java
  src/test/java/org/apache/hadoop/hbase/util/TestThreads.java

MANAGE HERALD DIFFERENTIAL RULES
  https://reviews.facebook.net/herald/view/differential/

WHY DID I GET THIS EMAIL?
  https://reviews.facebook.net/herald/transcript/1167/

Tip: use the X-Herald-Rules header to filter Herald messages in your client.

                
> Make HBase Thrift server more configurable and add a command-line UI test
> -------------------------------------------------------------------------
>
>                 Key: HBASE-4863
>                 URL: https://issues.apache.org/jira/browse/HBASE-4863
>             Project: HBase
>          Issue Type: Improvement
>            Reporter: Mikhail Bautin
>            Assignee: Mikhail Bautin
>         Attachments: D531.1.patch
>
>
> This started as an internal hotfix where we found out that the Thrift server spawned 15000 threads. To bound the thread pool size I added a custom thread pool server implementation called HBaseThreadPoolServer into HBase codebase, and made the following parameters configurable from both command line and as config settings: minWorkerThreads, maxWorkerThreads, and maxQueuedRequests. Under an increasing load, the server creates new threads for every connection before the pool size reaches minWorkerThreads. After that, the server puts new connections into the queue and only creates a new thread when the queue is full. If an attempt to create a new thread fails, the server drops connection. The default TThreadPoolServer would crash in that case, but it never happened because the thread pool was unbounded, so the server would hang indefinitely, consume a lot of memory, and cause huge latency spikes on the client side.
> Another part of this fix is refactoring and unit testing of the command-line part of the Thrift server. The logic there is sufficiently complicated, and the existing ThriftServer class does not test that part at all. The new TestThriftServerCmdLine test starts the Thrift server on a random port with various combinations of options and talks to it through the client API from another thread.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira