Hi team, I wonder if anyone could give me some poi...
# general
s
Hi team, I wonder if anyone could give me some pointers in resolving a "Failed to cache distribution" error when running a pex generated out of my build? It's a pex that worked originally, then we replaced a dependency (tensorflow) with a different build of it (intel-tensorflow) and hit this issue.
as far as I can make out the wheel is properly packaged into the pex deps directory, but I can't even launch the pex repl using
PEX_INTERPRETER=1
without triggering that error. Just a little stumped at this point as to what to start looking for 🤔
This is building and running on ubuntu 18.04
h
Hi, can you provide the full stack trace of the error message? and in fact can you show the BUILD files+requirements.txt that demonstrates this?
s
Sure! Here's the trace from trying to launch the pex
And these are the BUILD files
sorry about the delay!
oh and this is the 3rdparty/python/requirements.txt
I should also mention that it turns out a
./pants run
of that binary will start the process, but fail immediately when loading any module trying to import tensorflow - which seems very related.
If I install the package in site packages instead (and remove it from the BUILD files) and launch with
PEX_INHERIT_PATH=1
it does work though.
pretty sure it's something funky that intel-tensorflow package is doing when it loads? I'm struggling to expose enough log information to debug it though.