[bigtable] InvalidChunk

[junghoahnsc]

1. OS type and version
   Ubuntu 14.04

2. Python version and virtual environment information python --version
   Python 2.7.6

3. google-cloud-python version pip show google-cloud, pip show google-<service> or pip freeze

Name: google-cloud-bigtable
Version: 0.22.0

4. Stacktrace if available

Traceback (most recent call last):
  File "bt.py", line 13, in <module>
    row = bt_table.read_row('mykey')
  File "/home/jungho.ahn/tmp/test/local/lib/python2.7/site-packages/google/cloud/bigtable/table.py", line 234, in read_row
    rows_data.consume_all()
  File "/home/jungho.ahn/tmp/test/local/lib/python2.7/site-packages/google/cloud/bigtable/row_data.py", line 323, in consume_all
    self.consume_next()
  File "/home/jungho.ahn/tmp/test/local/lib/python2.7/site-packages/google/cloud/bigtable/row_data.py", line 275, in consume_next
    self._validate_chunk(chunk)
  File "/home/jungho.ahn/tmp/test/local/lib/python2.7/site-packages/google/cloud/bigtable/row_data.py", line 390, in _validate_chunk
    self._validate_chunk_row_in_progress(chunk)
  File "/home/jungho.ahn/tmp/test/local/lib/python2.7/site-packages/google/cloud/bigtable/row_data.py", line 370, in _validate_chunk_row_in_progress
    _raise_if(not chunk.timestamp_micros or not chunk.value)
  File "/home/jungho.ahn/tmp/test/local/lib/python2.7/site-packages/google/cloud/bigtable/row_data.py", line 441, in _raise_if
    raise InvalidChunk(*args)
google.cloud.bigtable.row_data.InvalidChunk

5. Steps to reproduce
   See the code example
6. Code example

from google.cloud import bigtable

bt_client = bigtable.client.Client(project='myproject', read_only=True)
bt_instance = bt_client.instance(instance_id='myinstance')
bt_table = bt_instance.table(table_id='mytable')

row = bt_table.read_row('mykey')
print row.cells

This exception is raised all the time. Did I use incorrectly?

Thanks,

[dhermes]

It's tough to say without actually seeing the failed chunk. Can you run this with a debugger? It may be easiest to just dump it in a script and run python -m pdb script_name.py or to run it in IPython and use the %debug magic after it fails.

[junghoahnsc]

(Pdb) print chunk
timestamp_micros: 1485886500000000
(Pdb) print not chunk.value
True

Actually the value is an empty string.

[dhermes]

Gotcha, ISTM the chunk is equivalent to:

>>> from google.cloud.bigtable._generated import bigtable_pb2
>>> Chunk = bigtable_pb2.ReadRowsResponse.CellChunk
>>> chunk = Chunk(timestamp_micros=1485886500000000)
>>> chunk
timestamp_micros: 1485886500000000

>>> 

[dhermes]

@tseaver The culprit is exactly as described by @junghoahnsc. I don't quite recognize the validation since it was re-worked for v2 while I was on "paternity leave". Is there a reason we don't allow empty cells on read?

UPDATE: I just verified that writing an empty cell works:

>>> row = table.row(b'row-key')
>>> row.set_cell(u'col-fam-id', b'col-name', b'') 
>>> row.commit()

but then reading this row doesn't fail either (because the chunk type must be different):

>>> partial_result = table.read_row(b'row-key')
>>> cells = partial_result[u'col-fam-id'][b'col-name']
>>> cells[0].value   
''
>>> cells[0].timestamp
datetime.datetime(2017, 2, 2, 22, 5, 4, 333000, tzinfo=<UTC>)

@junghoahnsc The state of the chunks is also relevant here, so I need a little more info to reproduce.

[tseaver]

Too much water under the bridge since June :). I think we need to ask @garye, @mbrukman, or @sduskis. The context is:

• The state machine is in the "row in progress" state (no current partial cell)
• It is validating the chunk which would trigger a new cell (is doesn't reset or commit the row).
• It is asserting that there must be both a timestamp and a value set on that chunk.

[yuanbaisc]

Hi, I am trying to run this step by step:
The PartialRowData object's state is "row in progress"
The chunk it currently processing has only "qualifier" and "timestamp_micros". "reset_now" is False, no "commit_row" field, "value" is empty.

[yuanbaisc]

The error is triggered in _validate_chunk_row_in_progress method:
if not chunk.HasField('commit_row') and not chunk.reset_row:
_raise_if(not chunk.timestamp_micros or not chunk.value)

[garye]

Yeah I think the chunk and timestamp check there is invalid.

[junghoahnsc]

Is there any workaround until it is fixed?
@dhermes do you need any more information from our side? Actually our bigtable was written by Go/Java apps and now we're trying to read it in Python.

[garye]

Is there any way you can validate the same read using Go or Java? Just to gain confidence that the fix I suggested will work... And/or can you just remove those checks from your local copy and verify?

[sduskis]

A timestamp of 0 is valid. I think that there needs to be a chunk.value.

[garye]

Timestamp of 0 is fine and empty value for chunk.value is fine, but I think "not chunk.value" is true if it's an empty string which is wrong.

[sduskis]

empty values are not fine in these cases. An empty cell isn't a thing in BT. If there's no reset, there needs to be a value, IIRC.

[dhermes]

@junghoahnsc @yuanbaisc Can you consume the response using the low-level API?

>>> # NOT: partial_result = table.read_row(b'row-key')
>>> from google.cloud.bigtable.table import _create_row_request
>>> request_pb = _create_row_request(table.name, row_key=b'row-key')
>>> response_iterator = client._data_stub.ReadRows(request_pb)
>>> all_responses = list(response_iterator)

UPDATE: If there is private data returned, you can redact it or feel free to email me (email in my GitHub profile)