[Crunch2] CWL workflow failed with Crunch2, but passed with Crunch1
#1 Updated by Radhika Chippada about 5 years ago
I ran Jiayong's varscan CWL workflow (main-goCallVars.cwl) in 9tee4.
When I used the command "arvados-cwl-runner --local --wait --debug main-goCallVars.cwl test-input-yml/main-goCallVars-job.yml" the pipeline instance https://workbench.9tee4.arvadosapi.com/pipeline_instances/9tee4-d1hrv-wgvlm9an4a6caw6 was created and succeeded. It had 7 successful jobs.
When I it ran using "--api containers", i.e. Crunch2, it created a container request https://workbench.9tee4.arvadosapi.com/container_requests/9tee4-xvhdp-hdeyli0ct23pwsf . This had two children (freebayes and varscan) and the freebayes CR is failing.
(venv)email@example.com:~/goCallVars/goCallVars$ arvados-cwl-runner --local --wait --debug --api containers main-goCallVars.cwl test-input-yml/main-goCallVars-job.yml (venv)firstname.lastname@example.org:~/goCallVars/goCallVars$ arvados-cwl-runner --api containers --submit --no-wait main-goCallVars.cwl test-input-yml/main-goCallVars-job.yml
#2 Updated by Brett Smith about 5 years ago
9tee4 is currently trying and failing to dispatch a container for one of these workflows because it does not specify a number of CPUs in its runtime constraints. Once we upgrade the API server here, this container request wouldn't even be valid.
I think the bug here is that arvados-cwl-runner needs to fill in default constraints in cases where the workflow doesn't specify them. Otherwise, it might end up trying to submit invalid container requests to the API server.