Brian Forbis created THRIFT-4564:
------------------------------------
Summary: TBufferedTransport can leave corrupt data in the buffer
Key: THRIFT-4564
URL: https://issues.apache.org/jira/browse/THRIFT-4564
Project: Thrift
Issue Type: Bug
Components: Node.js - Library
Affects Versions: 0.11.0, 0.10.0
Reporter: Brian Forbis
I ran into the following issue on Thrift 0.10.0 using Buffered Transport and
Binary Protocol
IDL File:
{code:java}
struct Foo {
1: string name
}
service SocketTestService {
i16 countFoos(1: list<Foo> foos)
}
{code}
Client Call:
{code:java}
async function main() {
numFoos = await client.countFoos([
new thriftTypes.Foo(),
null,
new thriftTypes.Foo(),
]);
}
{code}
The issue I ran into was a client-side thrown exception:
{code:java}
TypeError: Cannot read property 'write' of null
at SocketTestService_countFoos_args.write
(/Users/bforbis/git/bforbis/thrift-socket-test/gen10/SocketTestService.js:81:15)
at exports.Client.SocketTestServiceClient.send_countFoos
(/Users/bforbis/git/bforbis/thrift-socket-test/gen10/SocketTestService.js:178:8)
at exports.Client.SocketTestServiceClient.countFoos
(/Users/bforbis/git/bforbis/thrift-socket-test/gen10/SocketTestService.js:165:10)
at main (/Users/bforbis/git/bforbis/thrift-socket-test/client.js:110:28)
at <anonymous>
at process._tickCallback (internal/process/next_tick.js:188:7)
{code}
The cause of the issue is because I pass in *null* as a parameter to my Foo
list, which does not have a write() method to serialize itself when writing to
the buffer. That's okay on its own, but what ends up happening here is we have
*half* of a message still written to our buffered transport, meaning that a
subsequent RPC call on that same transport will fail to be processed on the
server side.
{code:java}
Error: Invalid type: -128
at TBinaryProtocol.skip
(/Users/bforbis/git/bforbis/thrift-socket-test/node_modules/thrift/lib/nodejs/lib/thrift/binary_protocol.js:364:13)
at module.exports.Foo.Foo.read
(/Users/bforbis/git/bforbis/thrift-socket-test/gen11/socket-test_types.js:47:15)
at SocketTestService_countFoos_args.read
(/Users/bforbis/git/bforbis/thrift-socket-test/gen11/SocketTestService.js:51:17)
at exports.Processor.SocketTestServiceProcessor.process_countFoos
(/Users/bforbis/git/bforbis/thrift-socket-test/gen11/SocketTestService.js:224:8)
at exports.Processor.SocketTestServiceProcessor.process
(/Users/bforbis/git/bforbis/thrift-socket-test/gen11/SocketTestService.js:210:39)
at
/Users/bforbis/git/bforbis/thrift-socket-test/node_modules/thrift/lib/nodejs/lib/thrift/server.js:55:21
at Socket.<anonymous>
(/Users/bforbis/git/bforbis/thrift-socket-test/node_modules/thrift/lib/nodejs/lib/thrift/buffered_transport.js:48:5)
at emitOne (events.js:121:20)
at Socket.emit (events.js:211:7)
at addChunk (_stream_readable.js:263:12)
{code}
*Proposal*: We should allow applications to have global persistent socket
connections that don't corrupt their buffer if there is a serialization error
when doing an RPC all on the client. Calls to send_${RPC_NAME} should be
wrapped in a try / catch in case there are client side errors in the processing
code. In the case that an exception is thrown, the buffer in buffered transport
should be emptied rather than leaving it full.
*Side Note*: The cause of the client-side exception has been fixed in Thrift
0.11.0 in this [Pull
Request|https://github.com/apache/thrift/commit/de112fbb0d7f2139ef107211e82e03b574f890d0#diff-a686530169a8a14044fae0f95d54578c],
which casts undef to a new struct. But I fear that this issue where hanging
data on the buffered transport isn't cleaned up could come up again.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
