[
https://issues.apache.org/jira/browse/IMPALA-13427?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17887627#comment-17887627
]
ASF subversion and git services commented on IMPALA-13427:
----------------------------------------------------------
Commit 4c21084e20b0a10ba5c552f22a1a34c74e20ede2 in impala's branch
refs/heads/master from wzhou-code
[ https://gitbox.apache.org/repos/asf?p=impala.git;h=4c21084e2 ]
IMPALA-13427: Make connect timeout of statestore HA RPC tunable
In some deployment environment, unsuccessful socket connection attempts
take long time to return error if connect timeout of the socket is set
as 0. This causes statestored to tie up in unsuccessful connection
attempts during initialization if its peer is not ready when statestore
HA is enabled.
Currently socket connect timeout of Thrift RPC is always set as 0.
This patch makes socket connect timeout of Thrift RPC tunable, and
adds a flag variable statestore_ha_client_rpc_conn_timeout_ms for the
underlying socket connect timeout for Statestore HA RPC. Its default
value equals 0, which equals the default value of TSocket.connTimeout_.
Testing:
- Added new test cases for statestore HA with non zero value for
statestore_ha_client_rpc_conn_timeout_ms.
- Passed core tests.
Change-Id: Ie5840a76b4a34b4714c47b86f6366328b5ceed3a
Reviewed-on: http://gerrit.cloudera.org:8080/21900
Reviewed-by: Norbert Luksa <[email protected]>
Tested-by: Impala Public Jenkins <[email protected]>
> Make the statestore HA rpc's socket connection timeout tunable
> --------------------------------------------------------------
>
> Key: IMPALA-13427
> URL: https://issues.apache.org/jira/browse/IMPALA-13427
> Project: IMPALA
> Issue Type: Improvement
> Components: Backend
> Reporter: Wenzhe Zhou
> Assignee: Wenzhe Zhou
> Priority: Major
>
> In some deployment environment, unsuccessful socket connection attempts take
> long time to return error if connTimeout of the socket is set as 0. This
> causes a daemon to tie up in unsuccessful connection attempts during
> initialization if its peer is not ready.
> When statestored HA is enabled, statestoreds do HA hand shack to negotiate
> active role during initialization. If two statestored are initialized
> serially, the first statestored may tie up in unsuccessful connection
> attempts since the second statestored is not started.
> Currently socket connection timeout of Thrift RPC is always set as 0. To
> avoid this issue, we need to make the statestore HA rpc's socket connection
> timeout tunable.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
