[
https://issues.apache.org/jira/browse/BEAM-10824?focusedWorklogId=481006&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-481006
]
ASF GitHub Bot logged work on BEAM-10824:
-----------------------------------------
Author: ASF GitHub Bot
Created on: 09/Sep/20 19:16
Start Date: 09/Sep/20 19:16
Worklog Time Spent: 10m
Work Description: tvalentyn commented on a change in pull request #12756:
URL: https://github.com/apache/beam/pull/12756#discussion_r485853260
##########
File path: sdks/python/setup.py
##########
@@ -165,6 +165,7 @@ def get_version():
'requests>=2.24.0,<3.0.0',
'typing>=3.7.0,<3.8.0; python_full_version < "3.5.3"',
'typing-extensions>=3.7.0,<3.8.0',
+ 'mmh3>=2.5.1,<2.5.2',
Review comment:
(I thought I already added this comment here but for some reason I don't
see it...)
Let's make the upper bound more flexible `mmh3>=2.5.1,<3.0`, or remove the
obligatory dependency on mmh3 as we do for snappy. @aaltay do you have a
preference on this?
For the record, Windows tests on the PR are passing. AFAIK, previously we
didn't add a dep on mmh3 we observed installation errors on Google internal
Windows test.
----------------------------------------------------------------
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: 481006)
Remaining Estimate: 11h 40m (was: 11h 50m)
Time Spent: 12h 20m (was: 12h 10m)
> Hash in stats.ApproximateUniqueCombineFn NON-deterministic
> ----------------------------------------------------------
>
> Key: BEAM-10824
> URL: https://issues.apache.org/jira/browse/BEAM-10824
> Project: Beam
> Issue Type: Bug
> Components: sdk-py-core
> Reporter: Monica Song
> Priority: P1
> Labels: hash
> Original Estimate: 24h
> Time Spent: 12h 20m
> Remaining Estimate: 11h 40m
>
> The python hash() function is non-deterministic. As a result, different
> workers will map identical values to different hashes. This leads to
> overestimation of the number of unique values (by several magnitudes, in my
> experience x1000) in a distributed processing model.
> [https://github.com/apache/beam/blob/master/sdks/python/apache_beam/transforms/stats.py#L218]
>
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
