Bryan Ashby created ARROW-14263:
-----------------------------------
Summary: "Invalid flatbuffers message" thrown with some serialized
RecordBatch's
Key: ARROW-14263
URL: https://issues.apache.org/jira/browse/ARROW-14263
Project: Apache Arrow
Issue Type: Bug
Components: C++, Python
Affects Versions: 5.0.0
Environment: pyarrow==5.0.0
C++ = 5.0.0
Windows 10 Pro x64
Reporter: Bryan Ashby
Attachments: record-batch-large.arrow
I'm running into various exceptions (often: "Invalid flatbuffers message") when
attempting to de-serialize RecordBatch's in Python that were generated in C++.
The same batch can be de-serialized back within C++.
*Example (C++)* (status checks omitted), but they are check in real code)*:*
{code:java}
const auto stream = arrow::io::BufferOutputStream::Create();
{
const auto writer = arrow::ipc::MakeStreamWriter(*stream, schema);
sdk::MaybeThrowError(writer); const auto writeRes =
(*writer)->WriteRecordBatch(batch);
sdk::MaybeThrowError((*writer)->Close());
sdk::MaybeThrowError(writeRes);
}auto buffer = (*stream)->Finish();std::ofstream
ofs("record-batch-large.arrow"); // we'll read this in Python
ofs.write(reinterpret_cast<const char*>((*buffer)->data()), (*buffer)->size());
ofs.close();auto backAgain = DeserializeRecordBatch((*buffer)); // all good
{code}
*Then in Python*:
{code:java}
with open("record-batch-large.arrow", "rb") as f:
data = f.read()reader = pa.RecordBatchStreamReader(data) // throws here
- "Invalid flatbuffers message"
{code}
Please see the attached .arrow file (produced above).
Any ideas?
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
