X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/df591042778408d03d410d5c22a669d85652d1ea..1010c9e519409f0eb36971df6c382fc7cc0e2732:/sdk/cwl/arvados_cwl/arvworkflow.py diff --git a/sdk/cwl/arvados_cwl/arvworkflow.py b/sdk/cwl/arvados_cwl/arvworkflow.py index 5aed871a12..ae90625102 100644 --- a/sdk/cwl/arvados_cwl/arvworkflow.py +++ b/sdk/cwl/arvados_cwl/arvworkflow.py @@ -15,6 +15,7 @@ from cwltool.process import shortname from cwltool.workflow import Workflow, WorkflowException from cwltool.pathmapper import adjustFileObjs, adjustDirObjs, visit_class from cwltool.builder import Builder +from cwltool.context import LoadingContext import ruamel.yaml as yaml @@ -51,13 +52,24 @@ def upload_workflow(arvRunner, tool, job_order, project_uuid, uuid=None, upload_dependencies(arvRunner, name, tool.doc_loader, packed, tool.tool["id"], False) - # TODO nowhere for submit_runner_ram to go. + if submit_runner_ram: + hints = main.get("hints", []) + found = False + for h in hints: + if h["class"] == "http://arvados.org/cwl#WorkflowRunnerResources": + h["ramMin"] = submit_runner_ram + found = True + break + if not found: + hints.append({"class": "http://arvados.org/cwl#WorkflowRunnerResources", + "ramMin": submit_runner_ram}) + main["hints"] = hints body = { "workflow": { "name": name, "description": tool.tool.get("doc", ""), - "definition":yaml.round_trip_dump(packed) + "definition":json.dumps(packed, sort_keys=True, indent=4, separators=(',',': ')) }} if project_uuid: body["workflow"]["owner_uuid"] = project_uuid @@ -89,7 +101,7 @@ def get_overall_res_req(res_reqs): if isinstance(res_req[a], int): # integer check all_res_req[a].append(res_req[a]) else: - msg = SourceLine(res_req).makeError( + msg = SourceLine(res_req, a).makeError( "Non-top-level ResourceRequirement in single container cannot have expressions") exception_msgs.append(msg) if exception_msgs: @@ -109,14 +121,16 @@ def get_overall_res_req(res_reqs): class ArvadosWorkflow(Workflow): """Wrap cwltool Workflow to override selected methods.""" - def __init__(self, arvrunner, toolpath_object, **kwargs): - super(ArvadosWorkflow, self).__init__(toolpath_object, **kwargs) + def __init__(self, arvrunner, toolpath_object, loadingContext): + super(ArvadosWorkflow, self).__init__(toolpath_object, loadingContext) self.arvrunner = arvrunner - self.work_api = kwargs["work_api"] self.wf_pdh = None + self.dynamic_resource_req = [] + self.static_resource_req = [] + self.wf_reffiles = [] + self.loadingContext = loadingContext - def job(self, joborder, output_callback, **kwargs): - kwargs["work_api"] = self.work_api + def job(self, joborder, output_callback, runtimeContext): req, _ = self.get_requirement("http://arvados.org/cwl#RunInSingleContainer") if req: with SourceLine(self.tool, None, WorkflowException, logger.isEnabledFor(logging.DEBUG)): @@ -140,57 +154,84 @@ class ArvadosWorkflow(Workflow): packed = pack(document_loader, workflowobj, uri, self.metadata) - builder = Builder() - builder.job = joborder - builder.requirements = workflowobj["requirements"] - builder.hints = workflowobj["hints"] - builder.resources = {} - - res_reqs = {"requirements": [], "hints": []} - for t in ("requirements", "hints"): - for item in packed["$graph"]: - if t in item: - if item["id"] == "#main": # evaluate potential expressions in the top-level requirements/hints - for req in item[t]: - if req["class"] == "ResourceRequirement": - eval_req = {"class": "ResourceRequirement"} - for a in max_res_pars + sum_res_pars: - if a in req: - eval_req[a] = builder.do_eval(req[a]) - res_reqs[t].append(eval_req) - else: - for req in item[t]: - if req["class"] == "ResourceRequirement": - res_reqs[t].append(req) - overall_res_req = {"requirements": get_overall_res_req(res_reqs["requirements"]), - "hints": get_overall_res_req(res_reqs["hints"])} - - new_spec = {"requirements": self.requirements, "hints": self.hints} - for t in ("requirements", "hints"): - for req in new_spec[t]: - if req["class"] == "ResourceRequirement": - new_spec[t].remove(req) - if overall_res_req[t]: - new_spec[t].append(overall_res_req[t]) + builder = Builder(joborder, + requirements=workflowobj["requirements"], + hints=workflowobj["hints"], + resources={}) + + def visit(item): + for t in ("hints", "requirements"): + if t not in item: + continue + for req in item[t]: + if req["class"] == "ResourceRequirement": + dyn = False + for k in max_res_pars + sum_res_pars: + if k in req: + if isinstance(req[k], basestring): + if item["id"] == "#main": + # only the top-level requirements/hints may contain expressions + self.dynamic_resource_req.append(req) + dyn = True + break + else: + with SourceLine(req, k, WorkflowException): + raise WorkflowException("Non-top-level ResourceRequirement in single container cannot have expressions") + if not dyn: + self.static_resource_req.append(req) + + visit_class(packed["$graph"], ("Workflow", "CommandLineTool"), visit) + + if self.static_resource_req: + self.static_resource_req = [get_overall_res_req(self.static_resource_req)] upload_dependencies(self.arvrunner, - kwargs.get("name", ""), + runtimeContext.name, document_loader, packed, uri, False) + # Discover files/directories referenced by the + # workflow (mainly "default" values) + visit_class(packed, ("File", "Directory"), self.wf_reffiles.append) + + + if self.dynamic_resource_req: + builder = Builder(joborder, + requirements=self.requirements, + hints=self.hints, + resources={}) + + # Evaluate dynamic resource requirements using current builder + rs = copy.copy(self.static_resource_req) + for dyn_rs in self.dynamic_resource_req: + eval_req = {"class": "ResourceRequirement"} + for a in max_res_pars + sum_res_pars: + if a in dyn_rs: + eval_req[a] = builder.do_eval(dyn_rs[a]) + rs.append(eval_req) + job_res_reqs = [get_overall_res_req(rs)] + else: + job_res_reqs = self.static_resource_req + with Perf(metrics, "subworkflow adjust"): joborder_resolved = copy.deepcopy(joborder) joborder_keepmount = copy.deepcopy(joborder) reffiles = [] - visit_class(joborder_keepmount, ("File", "Directory"), lambda x: reffiles.append(x)) + visit_class(joborder_keepmount, ("File", "Directory"), reffiles.append) + + mapper = ArvPathMapper(self.arvrunner, reffiles+self.wf_reffiles, runtimeContext.basedir, + "/keep/%s", + "/keep/%s/%s") - mapper = ArvPathMapper(self.arvrunner, reffiles, kwargs["basedir"], - "/keep/%s", - "/keep/%s/%s", - **kwargs) + # For containers API, we need to make sure any extra + # referenced files (ie referenced by the workflow but + # not in the inputs) are included in the mounts. + if self.wf_reffiles: + runtimeContext = runtimeContext.copy() + runtimeContext.extra_reffiles = copy.deepcopy(self.wf_reffiles) def keepmount(obj): remove_redundant_fields(obj) @@ -226,15 +267,13 @@ class ArvadosWorkflow(Workflow): "inputs": self.tool["inputs"], "outputs": self.tool["outputs"], "stdout": "cwl.output.json", - "requirements": self.requirements+[ + "requirements": self.requirements+job_res_reqs+[ + {"class": "InlineJavascriptRequirement"}, { "class": "InitialWorkDirRequirement", "listing": [{ "entryname": "workflow.cwl", - "entry": { - "class": "File", - "location": "keep:%s/workflow.cwl" % self.wf_pdh - } + "entry": '$({"class": "File", "location": "keep:%s/workflow.cwl"})' % self.wf_pdh }, { "entryname": "cwl.input.yml", "entry": json.dumps(joborder_keepmount, indent=2, sort_keys=True, separators=(',',': ')).replace("\\", "\\\\").replace('$(', '\$(').replace('${', '\${') @@ -244,8 +283,6 @@ class ArvadosWorkflow(Workflow): "arguments": ["--no-container", "--move-outputs", "--preserve-entire-environment", "workflow.cwl#main", "cwl.input.yml"], "id": "#" }) - kwargs["loader"] = self.doc_loader - kwargs["avsc_names"] = self.doc_schema - return ArvadosCommandTool(self.arvrunner, wf_runner, **kwargs).job(joborder_resolved, output_callback, **kwargs) + return ArvadosCommandTool(self.arvrunner, wf_runner, self.loadingContext).job(joborder_resolved, output_callback, runtimeContext) else: - return super(ArvadosWorkflow, self).job(joborder, output_callback, **kwargs) + return super(ArvadosWorkflow, self).job(joborder, output_callback, runtimeContext)