[
https://issues.apache.org/jira/browse/THRIFT-5108?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17043077#comment-17043077
]
Mario Emmenlauer commented on THRIFT-5108:
------------------------------------------
This is actually not a bug as I first thought. The code in the relevant section
could be more readable and I will submit a PR soon. However the actual problem
is that I do not understand how the transport is meant to be used.
I have implemented my own C++ thrift server on top of boost::beast websockets.
I use a {{TMemoryBuffer}} as the output buffer of the {{process()}} method,
then {{borrow()}} the pointer to send the data through the websocket and
finally I {{consume()}} the data.
But the buffer becomes fuller and fuller until it reaches 2GB and throws. Does
{{consume()}} not free the write-buffer for subsequent writes? How should one
"clear" the buffer for the next write? I could destroy and re-create the
Transport and the Protocol after every {{process()}}, but that must be
inefficient?
> Memory allocation error in TBufferTransports
> --------------------------------------------
>
> Key: THRIFT-5108
> URL: https://issues.apache.org/jira/browse/THRIFT-5108
> Project: Thrift
> Issue Type: Improvement
> Components: C++ - Library
> Affects Versions: 0.13.0
> Reporter: Mario Emmenlauer
> Assignee: Mario Emmenlauer
> Priority: Major
>
> When using the memory buffered transport I randomly get
> {{TTransportException}} in {{TBufferTransports.cpp}} method
>
> {{void TMemoryBuffer::ensureCanWrite(uint32_t len) {}}
> {{ ...}}
> {{ // Grow the buffer as necessary.}}
> {{ uint64_t new_size = bufferSize_;}}
> {{ while (len > avail) {}}
> {{ new_size = new_size > 0 ? new_size * 2 : 1;}}
> {{ if (new_size > maxBufferSize_) {}}
> {{ throwTTransportException(TTransportException::BAD_ARGS,}}
> {{ "Internal buffer size overflow when requesting a buffer of
> size " + std::to_string(new_size));}}
> {{ }}}
> {{ avail = available_write() + (static_cast<uint32_t>(new_size) -
> bufferSize_);}}
> }
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
