You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Just want to kindly ask if this is a reasonable feature: when packing a pipeline, find all http(s) imports, download those imports and convert them to local imports.
I think WDL being able to use http(s) in imports is really convenient for development. However, I've seen it's not always good for production pipelines. Just to name one, it could be not FAIR. For example, when a pipeline is stabilized to scaled analysis, it seems every runs will do http requests (maybe I missed some Cromwell cache?) which might be ideal for server and get throttled thus not accessible. Or two runs at different time may pull different version of http(s) imports thus not reproducible.
I can see potential counter arguments as pipeline authors should be responsible for this and maybe avoid http(s) imports as best practice. I've also seen old issues like openwdl/wdl#226 might be related here and might provide alternative solution. But still this might be a right feature for a packaging tool?
The text was updated successfully, but these errors were encountered:
Hi @yunhailuo miniwdl has now a miniwdl zip command. That can convert binary imports. The machinery for binary reproducible packages has been implemented there too. Packages packaged with miniwdl zip are binary reproducible by default. (All the modification times are set to timepoint 0, which is more elegant than the solution in wdl-packager.)
Hi @rhpvorderman , thanks a lot for the tips. I tried that once but got an error about bad common prefix. I'll definitely try again more carefully and ask miniwdl if needed. Close this issue now.
First, thank you for the great tool!
Just want to kindly ask if this is a reasonable feature: when packing a pipeline, find all http(s) imports, download those imports and convert them to local imports.
I think WDL being able to use http(s) in imports is really convenient for development. However, I've seen it's not always good for production pipelines. Just to name one, it could be not FAIR. For example, when a pipeline is stabilized to scaled analysis, it seems every runs will do http requests (maybe I missed some Cromwell cache?) which might be ideal for server and get throttled thus not accessible. Or two runs at different time may pull different version of http(s) imports thus not reproducible.
I can see potential counter arguments as pipeline authors should be responsible for this and maybe avoid http(s) imports as best practice. I've also seen old issues like openwdl/wdl#226 might be related here and might provide alternative solution. But still this might be a right feature for a packaging tool?
The text was updated successfully, but these errors were encountered: