[jira] Updated: (CASSANDRA-1331) Any request after a TApplicationException hangs

["Jonathan Ellis (JIRA)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/CASSANDRA-1331?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Jonathan Ellis updated CASSANDRA-1331:
--------------------------------------

    Affects Version/s:     (was: 0.6.3)
             Priority: Minor  (was: Major)

this is a thrift bug, but I don't see what is causing it.

if you look at the generated thrift code sending the error (Cassandra.get_count_args.read), validate() is not called until all the data has been read.  The only thing remaining should be the thrift "footer", and if you look at Cassandra.get_count.process which is calling read(), it calls readMessageEnd before returning the ApplicationError as well.

> Any request after a TApplicationException hangs
> -----------------------------------------------
>
>                 Key: CASSANDRA-1331
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-1331
>             Project: Cassandra
>          Issue Type: Bug
>          Components: Core
>            Reporter: Apache
>            Priority: Minor
>
> Observed that the design of request validation can return without consuming the complete request. The remnant is then read by the next request and produces a large read size.
> readMessageBegin size: 134218752
> Sample test case in python:
> from cassandra.ttypes import *
> from cassandra import Cassandra
> from thrift import Thrift
> from thrift.transport import TTransport
> from thrift.transport import TSocket
> from thrift.protocol.TBinaryProtocol import TBinaryProtocolAccelerated
> socket = TSocket.TSocket("127.0.0.1", 9160)
> transport = TTransport.TBufferedTransport(socket)
> protocol = TBinaryProtocolAccelerated(transport)
> client = Cassandra.Client(protocol)
> transport.open()
> client.transport = transport
> # don't specify a column_family to force a TApplicationException
> parent = ColumnParent()
> try:
>     client.get_count("ignore_keyspace", "ignore_key", parent, 1)
>     print "ERROR: we didn't see the problem"
> except TApplicationException as e:
>     message = "Required field 'column_family' was not present!"
>     if e.message.startswith(message):
>         print "OK, we got the error we were looking for."
>         print "The server input buffer was only partially read,"
>         print "  up to the validation error, so our next request"
>         print "  will start reading the stale data causing a hang."
>         print "hanging..."
>         client.get_count("ignore_keyspace", "ignore_key", parent, 1)
> client.transport.close()

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.