[BEAM-5156] Apache Beam on dataflow runner can't find Tensorflow for workers

• Type: Bug
• Status: Resolved
• Priority: P2
• Resolution: Fixed
• Fix Version/s: 2.17.0
• Component/s: sdk-py-core
• Labels: triaged
• Environment: google cloud compute instance running linux
• Assignee: Ankur Goenka
• Reporter: Thomas Johns
• Created: 15/Aug/18 22:04
• Updated: 16/May/20 14:00
• Resolved: 08/Jan/20 00:50

Description

Adding serialized tensorflow model to apache beam pipeline with python sdk but it can not find any version of tensorflow when applied to dataflow runner although it is not a problem locally. Tried various versions of tensorflow from 1.6 to 1.10. I thought it might be a conflicting package some where so I removed all other packages and tried to just install tensorflow and same problem.

Could not find a version that satisfies the requirement tensorflow==1.6.0 (from -r reqtest.txt (line 59)) (from versions: )No matching distribution found for tensorflow==1.6.0 (from -r reqtest.txt (line 59))