[
https://issues.apache.org/jira/browse/BEAM-8271?focusedWorklogId=375938&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-375938
]
ASF GitHub Bot logged work on BEAM-8271:
----------------------------------------
Author: ASF GitHub Bot
Created on: 22/Jan/20 22:26
Start Date: 22/Jan/20 22:26
Worklog Time Spent: 10m
Work Description: chadrik commented on issue #10595: [BEAM-8271] Properly
encode/decode StateGetRequest/Response continuation token
URL: https://github.com/apache/beam/pull/10595#issuecomment-577415527
> Actually, I think it's more correct to let continuation_token by bytes
everywhere (possibly fixing the uses if need be).
Here's the ticket I made: https://issues.apache.org/jira/browse/BEAM-8271.
It's been awhile, but my instinct was that this is an issue in the proto
file: all other id-like fields throughout these protos are strings. Also, the
token itself is created using string operations (`'token_%x' %
len(self._continuations)`) so it certainly feels like this wants to be a string
right up until it's encoded and stored in the proto message.
IIRC, my first attempt to solve this was to preserve the `bytes` typing, but
I think there are other calls that create the continuation token from strings,
and it started getting pretty ugly (converting from bytes to str to do some
string formatting, then back to bytes in multiple places, etc), so I looked at
how this was solved in Java, and based my changes here on that. It keeps the
ugliness isolated and at a minimum.
Again, it was awhile ago, so I may be misremembering something. If I
haven't convinced you yet, I can show you what an alternate solution looks like
using bytes, but first please do have a look at other ids in the .proto files
and at the Java equivalent for this.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
Issue Time Tracking
-------------------
Worklog Id: (was: 375938)
Time Spent: 20m (was: 10m)
> StateGetRequest/Response continuation_token should be string
> ------------------------------------------------------------
>
> Key: BEAM-8271
> URL: https://issues.apache.org/jira/browse/BEAM-8271
> Project: Beam
> Issue Type: Improvement
> Components: beam-model
> Reporter: Chad Dombrova
> Assignee: Chad Dombrova
> Priority: Major
> Time Spent: 20m
> Remaining Estimate: 0h
>
> I've been working on adding typing to the python code and I came across a
> discrepancy between regarding the type of the continuation token. The .proto
> defines it as bytes, but the code treats it as a string (i.e. unicode):
>
> {code:java}
> // A request to get state.
> message StateGetRequest {
> // (Optional) If specified, signals to the runner that the response
> // should resume from the following continuation token.
> //
> // If unspecified, signals to the runner that the response should start
> // from the beginning of the logical continuable stream.
> bytes continuation_token = 1;
> }
> // A response to get state representing a logical byte stream which can be
> // continued using the state API.
> message StateGetResponse {
> // (Optional) If specified, represents a token which can be used with the
> // state API to get the next chunk of this logical byte stream. The end of
> // the logical byte stream is signalled by this field being unset.
> bytes continuation_token = 1;
> // Represents a part of a logical byte stream. Elements within
> // the logical byte stream are encoded in the nested context and
> // concatenated together.
> bytes data = 2;
> }
> {code}
> From FnApiRunner.StateServicer:
> {code:python}
> def blocking_get(self, state_key, continuation_token=None):
> with self._lock:
> full_state = self._state[self._to_key(state_key)]
> if self._use_continuation_tokens:
> # The token is "nonce:index".
> if not continuation_token:
> token_base = 'token_%x' % len(self._continuations)
> self._continuations[token_base] = tuple(full_state)
> return b'', '%s:0' % token_base
> else:
> token_base, index = continuation_token.split(':')
> ix = int(index)
> full_state = self._continuations[token_base]
> if ix == len(full_state):
> return b'', None
> else:
> return full_state[ix], '%s:%d' % (token_base, ix + 1)
> else:
> assert not continuation_token
> return b''.join(full_state), None
> {code}
> This could be a problem in python3.
> All other id values are string, whereas bytes is reserved for data, so I
> think that the proto should be changed to string.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
