aaltay commented on pull request #15105: URL: https://github.com/apache/beam/pull/15105#issuecomment-920544932
I still do not think this is the right solution. The reason is: Ideally user's would be responsible for their dependencies being frozen but that is not the case, and that is not how user's are operating today. Even with frozen dependencies it is possible for sub-dependencies to change, or even dependencies change with the same version (as far as i know, pypi deos not prevent this). Additionally, with the GCS staging folder changing dependencies would require an active user action, with pypi it could chage without any user interaction. (More issues, pypi does not have SLOs as high as GCS, and Dataflow pipelines failing at startup because of pypi dependencies is common user issue.) I understand the use you are describing. I would think that custom containers would actually result in a faster experience. Note that with custom containers that containers is built only once in the local machines, without that the same container will be built N times, one for each worker. The suggestion to add support for remote package to the extra_package sounds good to me. I actually thought this is already the case (see: [1]). Is it not working? It seems like it does not support gs:// scheme but supports https://. You could add gs:// support there, or I believe you could get https url for files that exists on GCS and use that instead. [1] https://github.com/apache/beam/blob/abfd4c662a6701feef078c36551c15bf4303eef4/sdks/python/apache_beam/runners/portability/stager.py#L577 -- 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. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
