Hi Metaflow, I’m wondering if there will be support for

alpine

images. Our security hardening policies will be much easier to work with if we’re able to run in the alpine image

3.10.9-alpine3.17

. @wide-battery-59962 @cold-actor-8608 @quiet-waiter-82239

You should be able to use these today with the image keyword for batch and kubernetes decorator.

@square-wire-39606 thanks for the quick reply. I have been able to get my code to work on

3.10.8-buster

and

3.10.8-bullseye

, but running

3.10.8-alpine3.17

results in this error:

2023-01-17 13:39:50.551 [1673984330327840/start/1 (pid 24157)] [pod t-66kzw-k7vvh] Task is starting (Pod is pending, Container is terminated - StartError - failed to create containerd task: failed to create shim task: OCI runtime create failed: runc create failed: unable to start container process: exec: "bash": executable file not found in $PATH: unknown)...

When I replace

bash

with

sh

, I get the following error:

[2023-01-17, 13:42:25 ] {{pod_manager.py:197}} INFO - sh: metaflow_CapacityManagementTrainingFlow_linux-64_f61c1642322721c49557ef07da6e069319773a80/bin/python: not found

. Any advice here?

@aloof-application-80784, does you flow use

@conda

?

yes

The error message you see about ".../python not found" is caused by a missed code package inside the container. If you'd check the generated command, you could find a long shell command with tools called which are likely missed in the docker image you used. Especially with alpine images, e.g. if there is no

tar

, then a downloaded code package is not unpacked, hence no python could be found. Flows with

@conda

require even more tools IIRC, something like wget. The real error messages are hidden but you could run the same command inside docker container locally. This way I managed to debug what was missed inside my custom docker images.

thanks @fancy-eve-65019! i’ll test out a private docker image