Hey team, I want to file a bug that I believe was ...
# dev-metaflow
q
Hey team, I want to file a bug that I believe was introduced by this PR. Specifically this line is the culprit:
Copy code
# Use Decimals to compare and convert to string here so
# that numbers that can't be exactly represented as
# floats (e.g. 0.8) still look "nice". We don't care
# about precision more that .001 for resources anyway.
result[k] = str(max(float(my_val or 0), float(v or 0)))
With this change we are no longer able to deploy to AWS SFN. Example:
@resource(cpp=1, memory=2048)
On
step-functions deploy
the following error is received (not attaching full stack trace for brevity):
Copy code
if int(gpu) > 0:
ValueError: invalid literal for int() with base 10: '0.0'
The reason this happens is because
gpu=0
, which gets converted to
"0.0"
by the above culprit line. The gpu resolver fails to parse it as an int any more and breaks
1
a
indeed, thanks for reporting
i think we should have a fix for this out today
🙌 1
hey sorry for the delay, this should fix it https://github.com/Netflix/metaflow/pull/1118
should be fixed in just-released 2.7.7 ( already available on pip, soon on conda-forge)
👍 1