Find a better name for Row.commit_modifications() or fold the behavior into

See discussion in #1547 for context.

The name modifications is intended to map to the RPC method name ReadModifyWriteRow but the word modification is a synonym of mutation, so this method is poorly named. The method actually sends changes that

increment an int (stored in BT as 8 bytes)
append bytes to a cell value

so we should either

Fold it into Row.commit() as commit(append_only=True)
Rename it to something like Row.commit_cell_updates (or a better name than that)

@jonparrott

@jonparrott @jgeewax @tseaver I'd love to have you all chime in.

I like the first option, but unsure about complexity/edge cases.

For option two, would commit_append be a better name?

I like the first also. commit() does all commits, commit(append_only=True) does just the appending commits. Seems super clear to me.

What does the user expect the state-of-the-world to be after commit(append_only=True), assuming there were any pending non-append modifications?

I think the best way forward is to actually create Row(..., append=True) and then make it so that in the append case, only append_cell_value and increment_cell_value will work.

This way Row.commit() does have to worry about sending up to 3 API requests to the RPC methods MutateRow (no filter), CheckAndMutateRow (filter) and ReadModifyWriteRow (append).

No issues with complexity, just worried that people would be confused. Here's some code snippets for how things work right now

>>> row1 = table.row(b'row-key')
>>> row2 = table.row(b'row-key', filter_=filter_)
>>>
>>> row1.set_cell(u'fam', b'col', b'val')
>>> row2.set_cell(u'fam', b'col', b'val')
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "gcloud/bigtable/row.py", line 166, in set_cell
    self._get_mutations(state).append(mutation_pb)
  File "gcloud/bigtable/row.py", line 97, in _get_mutations
    raise ValueError('A filter is set on the current row, but no '
ValueError: A filter is set on the current row, but no state given for the mutation
>>>
>>> # Use state
>>> row1.set_cell(u'fam', b'col', b'val', state=True)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "gcloud/bigtable/row.py", line 166, in set_cell
    self._get_mutations(state).append(mutation_pb)
  File "gcloud/bigtable/row.py", line 102, in _get_mutations
    raise ValueError('No filter was set on the current row, but a '
ValueError: No filter was set on the current row, but a state was given for the mutation
>>> row2.set_cell(u'fam', b'col', b'val', state=True)
>>>
>>> # state also an argument in delete(), delete_cell(), delete_cells()
>>>
>>> row1.append_cell_value(u'fam', b'col', b'extra-bytes')
>>> row2.append_cell_value(u'fam', b'col', b'extra-bytes')
>>>
>>> row1.increment_cell_value(u'fam', b'col', 123)
>>> row2.increment_cell_value(u'fam', b'col', 123)

@tseaver

@tseaver Having a hard time answering your question. Can you give a scenario of what you have in mind?

If commit() doesn't take the append_only argument (which I thought was the option being preferred), then my question is moot.

[For rows created with append_only], only append_cell_value and increment_cell_value will work.

That seems pretty natural: the user can't get the row into a state where uncommitted changes remain after a call to commit().

Yup. Will implement and send out for PR.

Moving Bigtable Row.commit_modifications() into commit().

chore: resolve docstring issue where oneof reference is missing (#1548)

chore: cleanup owlbot and release please files (#1548)

feat(experimental): Implement `__init__` method for _AsyncReadObjectS…

docs: add samples from spanner/cloud-client (#117)

fix: pandas.cut returns labels index for numeric breaks when labels=F…

Navigation Menu

Repository navigation

Find a better name for Row.commit_modifications() or fold the behavior into #1548

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Find a better name for Row.commit_modifications() or fold the behavior into #1548

Description

Activity

dhermes commented on Feb 29, 2016

theacodes commented on Feb 29, 2016

jgeewax commented on Feb 29, 2016

tseaver commented on Feb 29, 2016

dhermes commented on Feb 29, 2016

dhermes commented on Feb 29, 2016

tseaver commented on Feb 29, 2016

dhermes commented on Feb 29, 2016

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions