onError/onSuccess for rules

[dsnezhkov]

Hello,

I have a long running workflows, and want to report the progress of execution to an external dashboard.

It would be nice to have rule-level onerror/onsuccess hooks so that a more granular reporting can be done. This call back can be used to execute pre- and post- logic.

I currently work around it by emulating start/stop of a rule by wrapping processing action in a call to an external utility (which in turn makes an HTTP call to the dashboard) but was hoping this can be done natively.

Thank you.

[SichongP]

Doesn't snakemake already output progress of rule execution? At least it reports error/success I think. Would something that periodically probes an output log file of snakemake work?

[dsnezhkov]

I think snakemake run logs are not overly parser friendly for the environment I am in, both in structure and in timing (flush of output into a file). It would be cleaner to hook into the return code via callback. That way there is no separate process that watches the logs, and the callback is just another task I can instruct snakemake to execute on. Perhaps I may be missing some logging options?

[schaefed]

Hello Everybody,

first of all: Thanks for the great tool!

We are using Snakemake to organize scientific, ETL-like pipelines and it has proven to be extremely valuable. Our workflows run regularly and integrate quite a few failure-prone steps (e.g. data transfer over mobile networks, http-requests, limited computing ressources) and are monitored by non-technical folks. In order to provide fine-grained reporting, a feature like rule-based onerror/onsuccess handlers would be really useful.

Is there any chance to get some momentum behind this enhancement request? If so and with a somewhat clear prospect to get such a feature merged eventually, I would be happy to take on the work on this and prepare a MR. What do you think?

[gibsramen]

I would also be interested in such a feature. I have a few core steps in a workflow where it would be convenient to have a rule-level onerror/onsuccess.

[nh13]

I would also add that I wouldn't mind having this!

[msto]

A rule can complete successfully (exit code 0) but its outputs can be still be wrong. (e.g. if the invoked tool is buggy, if the user specified an incorrect argument, if the user provided incorrect inputs to the rule, etc)

One helpful use case for rule-level onsuccess would be to assert that the output data matches expectation.