[
https://issues.apache.org/jira/browse/BEAM-2595?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16083496#comment-16083496
]
Andrea Pierleoni commented on BEAM-2595:
----------------------------------------
yes, sorry forgot the stack trace
{code:}
Traceback (most recent call last):
File "/Users/andreap/work/code/library_dataflow/main.py", line 347, in
<module>
run()
File "/Users/andreap/work/code/library_dataflow/main.py", line 343, in run
write_disposition=beam.io.BigQueryDisposition.WRITE_TRUNCATE)
File
"/Users/andreap/library_dataflow/venv/lib/python2.7/site-packages/apache_beam/pvalue.py",
line 100, in __or__
return self.pipeline.apply(ptransform, self)
File
"/Users/andreap/library_dataflow/venv/lib/python2.7/site-packages/apache_beam/pipeline.py",
line 265, in apply
pvalueish_result = self.runner.apply(transform, pvalueish)
File
"/Users/andreap/library_dataflow/venv/lib/python2.7/site-packages/apache_beam/runners/runner.py",
line 150, in apply
return m(transform, input)
File
"/Users/andreap/library_dataflow/venv/lib/python2.7/site-packages/apache_beam/runners/runner.py",
line 156, in apply_PTransform
return transform.expand(input)
File "/Users/andreap/work/code/library_dataflow/beam2_1.py", line 213, in
expand
return pcoll | 'WriteToBigQuery' >> ParDo(bigquery_write_fn)
File
"/Users/andreap/library_dataflow/venv/lib/python2.7/site-packages/apache_beam/transforms/core.py",
line 620, in __init__
super(ParDo, self).__init__(fn, *args, **kwargs)
File
"/Users/andreap/library_dataflow/venv/lib/python2.7/site-packages/apache_beam/transforms/ptransform.py",
line 515, in __init__
self.fn = pickler.loads(pickler.dumps(self.fn))
File
"/Users/andreap/library_dataflow/venv/lib/python2.7/site-packages/apache_beam/internal/pickler.py",
line 225, in loads
return dill.loads(s)
File
"/Users/andreap/library_dataflow/venv/lib/python2.7/site-packages/dill/dill.py",
line 277, in loads
return load(file)
File
"/Users/andreap/library_dataflow/venv/lib/python2.7/site-packages/dill/dill.py",
line 266, in load
obj = pik.load()
File
"/usr/local/Cellar/python/2.7.13/Frameworks/Python.framework/Versions/2.7/lib/python2.7/pickle.py",
line 864, in load
dispatch[key](self)
File
"/usr/local/Cellar/python/2.7.13/Frameworks/Python.framework/Versions/2.7/lib/python2.7/pickle.py",
line 1195, in load_appends
list.extend(stack[mark + 1:])
File
"/Users/andreap/library_dataflow/venv/lib/python2.7/site-packages/apitools/base/protorpclite/messages.py",
line 1147, in extend
self.__field.validate(sequence)
AttributeError: 'FieldList' object has no attribute '_FieldList__field'
{code}
for some reasons once deserialized the seuquence does not have the
'_FieldList__field' attribute.
I don't believe it is introduced in 2.1.0 per se, but a (very useful) class
using it is coming with 2.1.0 and it may flag as a real problem in production.
Definitely for us.
> WriteToBigQuery does not work with nested json schema
> -----------------------------------------------------
>
> Key: BEAM-2595
> URL: https://issues.apache.org/jira/browse/BEAM-2595
> Project: Beam
> Issue Type: Bug
> Components: sdk-py
> Affects Versions: 2.1.0
> Environment: mac os local runner, Python
> Reporter: Andrea Pierleoni
> Assignee: Sourabh Bajaj
> Priority: Minor
> Labels: gcp
> Fix For: 2.1.0
>
>
> I am trying to use the new `WriteToBigQuery` PTransform added to
> `apache_beam.io.gcp.bigquery` in version 2.1.0-RC1
> I need to write to a bigquery table with nested fields.
> The only way to specify nested schemas in bigquery is with teh json schema.
> None of the classes in `apache_beam.io.gcp.bigquery` are able to parse the
> json schema, but they accept a schema as an instance of the class
> `apache_beam.io.gcp.internal.clients.bigquery.TableFieldSchema`
> I am composing the `TableFieldSchema` as suggested here
> [https://stackoverflow.com/questions/36127537/json-table-schema-to-bigquery-tableschema-for-bigquerysink/45039436#45039436],
> and it looks fine when passed to the PTransform `WriteToBigQuery`.
> The problem is that the base class `PTransformWithSideInputs` try to pickle
> and unpickle the function
> [https://github.com/apache/beam/blob/master/sdks/python/apache_beam/transforms/ptransform.py#L515]
> (that includes the TableFieldSchema instance) and for some reason when the
> class is unpickled some `FieldList` instance are converted to simple lists,
> and the pickling validation fails.
> Would it be possible to extend the test coverage to nested json objects for
> bigquery?
> They are also relatively easy to parse into a TableFieldSchema.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
