[
https://issues.apache.org/jira/browse/THRIFT-5322?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17247309#comment-17247309
]
Juraci Paixão Kröhling commented on THRIFT-5322:
------------------------------------------------
I'm not quite sure the problem is in the message size: the message itself is
under 5k in size, but due to the fact that the message is malformed, the buffer
size at that method becomes huge. I still couldn't come up with a valid
explanation for it, and I can't check what gets parsed as a result as there's
only one field that results in this huge memory usage and I get an EOF instead
of getting a call to my handler.
The attached source file can be used to reproduce the problem. Some times, the
memory consumption in the "after" phase is 400MiB, sometimes it's ~800MiB.
> Go compact_protocol allocating unreasonable buffer size
> -------------------------------------------------------
>
> Key: THRIFT-5322
> URL: https://issues.apache.org/jira/browse/THRIFT-5322
> Project: Thrift
> Issue Type: Bug
> Components: Go - Library
> Affects Versions: 0.13.0
> Reporter: Juraci Paixão Kröhling
> Priority: Major
> Attachments: main.go
>
>
> I don't yet know all the pieces to this puzzle, and it's quite possible that
> the problem is on our side, but we use the Thrift Go library in the Jaeger
> Agent and we are seeing a case where the memory consumption for a payload of
> 4k bytes to result in a buffer allocation in the compact_protocol.go with
> unreasonable sizes. I found buffers of 1.4GiB while debugging the issue.
>
> This is the code that we are seeing this memory usage:
> [https://github.com/apache/thrift/blob/b75e88a33d67ae05ef9b5fa001d2a63a2effe377/lib/go/thrift/compact_protocol.go#L556-L577]
>
> Here's more information about this, including a reproducer and initial
> diagnostics:
> [https://github.com/jaegertracing/jaeger/issues/2638#issuecomment-741848201]
>
> As mentioned above, I'm still getting all the pieces together, but perhaps
> you've seen this before or know what might be going on. What I know for sure
> at the moment is that this happens on malformed payloads, but I would expect
> the library to have an upper limit on the buffer size.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
