Arvados

Run pipeline with File input (pipeline_templates/qr1hi-p5p6p-xt9iw6rmhvm93q5) -> Select a file (selected into a clippy before hand) -> Run it -> Pipeline failing with 404 (pipeline_instances/qr1hi-d1hrv-udkzowfb2qr65nb)

stderr arvados.errors.ApiError: <HttpError 404 when requesting https://qr1hi.arvadosapi.com/arvados/v1/collections/qr1hi-4zz18-pudu13a9iuwvgwg%2Ffile1_in_main.txt?alt=json returned "Path not found”

The URL makes this look like the problem could be excessive URL-encoding. Is this a problem specific to Clippy? If so I'm inclined to call this a nice-to-have until we either retire Clippy or decided that we're not going to.

This is not an Arvados bug. This error is happening because of a mistake in the pipeline template: it says that the input parameter to the do_hash script is a File, but do_hash is actually expecting a Collection—it calls one_task_per_input_file to work iteratively over each file in the Collection. Ultimately do_hash crashes trying to treat the File spec as a Collection.

There's lots of discussion to be had about ways we might prevent this or make it clearer to the user—CollectionReader could support path syntax, or Arvados could have more information about how scripts expect their inputs to be formatted—but those need further discussion and speccing out as stories. Since I expect those to be handled separately, I'm closing this.