output viral-ngs version as string in WDL workflows

[tomkinsc]

This calls reports.py --version in a new WDL task at the end of each workflow so the viral-ngs version is reported among workflow outputs as a WDL string. It would be more efficient to call reports.py --version within all of the existing tasks to avoid the extra startup overhead of a new task call, but that would add quite a bit of repetition among all of the tasks.

[notestaff]

For workflows run on DNAnexus, the version can be baked into the workflow as a constant, the way the docker tag is. It is also currently possible to determine the version from 'dx describe analysis-xxxxx' output. It might also make sense to attach the viral-ngs version to the created workflows and apps as DNAnexus properties, and/or to create global workflows which are explicitly versioned.

One other option is to use git smudge filters to bake the git hash as a constant into the WDL files. Not sure if dxWDL would be smart enough to avoid spinning up an extra job in that case.

[dpark01]

Yeah been thinking about this a little... agree with the need to have this version more clearly stamped in a way we can access. A few thoughts:

• I do worry a bit about the extra spinup cost of the task at the end of every workflow (it's usually minimally about 5 mins to just pull the docker image).
• Agree with @notestaff that maybe DNAnexus properties is something we could utilize, but that only addresses the dxWDL compiled outputs, not Cromwell/Firecloud use cases.
• instead of git smudge filters, why not just coopt the version string sed hackery we're already doing during deployment? So that does a search-replace of all the docker tags in the runtime WDL blocks. Similar to @notestaff 's idea, why not just add an additional constant String to the output section of every task that repeats the same docker tag in the runtime block (and would get seded by the above script automatically)? This is probably the cleanest route.

Just a note that I still often ponder a future in which we don't need to have the exact same docker image for each WDL task (we could then strip the images down quite a bit since they don't have to be so monolithic), but I'm not sure if that future is compatible with retaining compatibility with Snakemake... anyway, a bit off-topic.

[tomkinsc]

Switching to containers may not break Snakemake after all: https://snakemake.readthedocs.io/en/stable/snakefiles/deployment.html#singularity

[notestaff]

Besides slimming down the images, per-task Docker images would enable much more reuse of past results when re-running old analyses with newer viral-ngs versions. E.g. assemble_denovo_with_deplete could reuse the deplete part if only the assemble code changed.

[tomkinsc]

Thanks for the review, everyone. The latest commit switches to a sed-replaced string replacement per-task. Here's an example CI execution of a workflow, with the output string shown. There's some redundancy since each task outputs the same value, but that would allow us to mix-and-match task versions in workflows while keeping track of their source versions.