[
https://issues.apache.org/jira/browse/THRIFT-4600?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17938084#comment-17938084
]
Jeffrey Tolar commented on THRIFT-4600:
---------------------------------------
This would be helpful for us: we're talking to a Hive cluster via a load
balancer. There's a very high chance that subsequent TCP connections will hit a
different backend server, which causes problems for Hive.
I believe the Hive protocol involves first calling {{{}OpenSession{}}}; that
returns a handle that seems to be valid only on the server that created it -
requests to other servers using that handle will fail.
Additionally, Hive makes use of signed cookies: the secret used to sign the
cookie [is randomly generated during
startup|https://github.com/apache/hive/blob/f06aa9541cdfc314831c2026fcddd552042e3ea4/service/src/java/org/apache/hive/service/auth/HttpAuthService.java#L48-L49],
which means that cookies generated by one server won't be accepted by another.
I've got a monkey-patch for the Thrift client that works to allow connection
reuse, but it'd be great if it could be addressed upstream - it looks like
[https://github.com/apache/thrift/pull/2204] still applies cleanly. I've tested
that patch locally, and it successfully allows connections to be reused (and
resolves the issue we see when connections aren't reused).
> Don't close the connection in flush for python THttpClient.py
> -------------------------------------------------------------
>
> Key: THRIFT-4600
> URL: https://issues.apache.org/jira/browse/THRIFT-4600
> Project: Thrift
> Issue Type: Bug
> Components: Python - Library
> Affects Versions: 0.11.0
> Reporter: Finn Isaac Norris Colman
> Priority: Major
> Attachments: 1572.patch
>
> Time Spent: 50m
> Remaining Estimate: 0h
>
> In thrift/lib/py/src/transport/THttpClient.py in the flush method it closes
> and reopens the connection:
> {noformat}
> if self.isOpen():
> self.close()
> self.open(){noformat}
>
> This code is closing the connection on every flush request. The flush request
> occurs for every message send, which means that keep alive cannot be used.
> Removing this code from flush means that keep alive can be used as the user
> now has control over when the connection is closed.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
