aws-batch: support Snakemake `--report`

[joverlee521]

Context

Snakemake has removed the --stats option in v8, so I'm looking into the --report option for long term workflow stats.

The Snakemake report must be generated after the workflow has finished. I thought this would be as simple as attaching/downloading an old AWS Batch job then running nextstrain build . --report.

When I did this for ncov-ingest, I saw a bunch of warnings along the lines of:

Missing metadata for file data/gisaid/metadata.tsv. Maybe metadata was deleted or it was created using an older version of Snakemake. This is a non critical warning.

I then realized we are explicitly excluding Snakemake state in the downloads from AWS Batch:

cli/nextstrain/cli/runner/aws_batch/s3.py

Lines 113 to 124 in 8ed779c

	# We don't want the remote Snakemake state to interfere locally…
	".snakemake/",
	# Ignore Python bytecode
	"*.pyc",
	"__pycache__/",
	])
	included = path_matcher([
	# But we do want the Snakemake logs to come over.
	".snakemake/log/",
	])

Possible solutions

1. Include .snakemake/metadata in the downloads from AWS Batch so that users can generate the Snakemake report locally.
2. Automatically generate the Snakemake report within the AWS Batch job so that users can download the rendered report

[2] definitely seems like the nicer option and maybe should be applied across all runtimes for nextstrain build?

[tsibley]

Hmm. Downloading the Snakemake state locally may fix this problem, but it can/will cause other problems. I don't know if it'd be ok if we scope it down to not all of .snakemake/ but just .snakemake/metadata/ as you suggest in 1.

What's the effect of the warnings? Is there useful information missing from the report? Or is just wanting to suppress the noise from Snakemake?

[joverlee521]

What's the effect of the warnings? Is there useful information missing from the report? Or is just wanting to suppress the noise from Snakemake?

The generated report does not include any runtime info:

[tsibley]

Ah, looking more closely at the contents of .snakemake/metadata/, I do think we want to start downloading it by default. It's mostly info used to determine if Snakemake needs to re-run rules based on inputs/outputs, and thus is akin to the file mtimes which we already preserve on download.

[tsibley]

1. Include .snakemake/metadata in the downloads from AWS Batch so that users can generate the Snakemake report locally.

We should do this, per above. I'll open a PR.

2. Automatically generate the Snakemake report within the AWS Batch job so that users can download the rendered report

[2] definitely seems like the nicer option and maybe should be applied across all runtimes for nextstrain build?

We could also do this as well, but it requires a little more consideration about how/where/when. Would you open it as a separate issue if you'd like to see it?

[tsibley]

This will also need a new docker-base image, as the same exclusions of .snakemake/ are recapitulated there:

https://github.com/nextstrain/docker-base/blob/ccac0787cbc6118d0e39518e2220f650e121b8bd/entrypoint-aws-batch#L43-L44

[joverlee521]

2. Automatically generate the Snakemake report within the AWS Batch job so that users can download the rendered report
   [2] definitely seems like the nicer option and maybe should be applied across all runtimes for nextstrain build?

We could also do this as well, but it requires a little more consideration about how/where/when. Would you open it as a separate issue if you'd like to see it?

Hmm, maybe this doesn't need to be built into the Nextsrain CLI. It could just be a separate step in the pathogen-repo-build workflow so we have reports for our automated pathogens.