X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/56555817311d4cadf261cd3aa380361690e6fe65..cc095ef1f8c86b05a8e58f15c4086a2caa861ec5:/sdk/cwl/arvados_cwl/arvcontainer.py diff --git a/sdk/cwl/arvados_cwl/arvcontainer.py b/sdk/cwl/arvados_cwl/arvcontainer.py index 0c7ad732d2..3703fc6217 100644 --- a/sdk/cwl/arvados_cwl/arvcontainer.py +++ b/sdk/cwl/arvados_cwl/arvcontainer.py @@ -17,26 +17,29 @@ import uuid import math import arvados_cwl.util -import ruamel.yaml as yaml +import ruamel.yaml from cwltool.errors import WorkflowException from cwltool.process import UnsupportedRequirement, shortname -from cwltool.pathmapper import adjustFileObjs, adjustDirObjs, visit_class -from cwltool.utils import aslist +from cwltool.utils import aslist, adjustFileObjs, adjustDirObjs, visit_class from cwltool.job import JobBase import arvados.collection from .arvdocker import arv_docker_get_image from . import done -from .runner import Runner, arvados_jobs_image, packed_workflow, trim_anonymous_location, remove_redundant_fields +from .runner import Runner, arvados_jobs_image, packed_workflow, trim_anonymous_location, remove_redundant_fields, make_builder from .fsaccess import CollectionFetcher from .pathmapper import NoFollowPathMapper, trim_listing from .perf import Perf +from ._version import __version__ logger = logging.getLogger('arvados.cwl-runner') metrics = logging.getLogger('arvados.cwl-runner.metrics') +def cleanup_name_for_collection(name): + return name.replace("/", " ") + class ArvadosContainer(JobBase): """Submit and manage a Crunch container request for executing a CWL CommandLineTool.""" @@ -57,7 +60,13 @@ class ArvadosContainer(JobBase): def update_pipeline_component(self, r): pass - def run(self, runtimeContext): + def _required_env(self): + env = {} + env["HOME"] = self.outdir + env["TMPDIR"] = self.tmpdir + return env + + def run(self, toplevelRuntimeContext): # ArvadosCommandTool subclasses from cwltool.CommandLineTool, # which calls makeJobRunner() to get a new ArvadosContainer # object. The fields that define execution such as @@ -67,15 +76,23 @@ class ArvadosContainer(JobBase): runtimeContext = self.job_runtime - container_request = { - "command": self.command_line, - "name": self.name, - "output_path": self.outdir, - "cwd": self.outdir, - "priority": runtimeContext.priority, - "state": "Committed", - "properties": {}, - } + if runtimeContext.submit_request_uuid: + container_request = self.arvrunner.api.container_requests().get( + uuid=runtimeContext.submit_request_uuid + ).execute(num_retries=self.arvrunner.num_retries) + else: + container_request = {} + + container_request["command"] = self.command_line + container_request["name"] = self.name + container_request["output_path"] = self.outdir + container_request["cwd"] = self.outdir + container_request["priority"] = runtimeContext.priority + container_request["state"] = "Committed" + container_request.setdefault("properties", {}) + + container_request["properties"]["cwl_input"] = self.joborder + runtime_constraints = {} if runtimeContext.project_uuid: @@ -134,6 +151,8 @@ class ArvadosContainer(JobBase): mounts[targetdir]["path"] = path prevdir = targetdir + "/" + intermediate_collection_info = arvados_cwl.util.get_intermediate_collection_info(self.name, runtimeContext.current_container, runtimeContext.intermediate_output_ttl) + with Perf(metrics, "generatefiles %s" % self.name): if self.generatefiles["listing"]: vwd = arvados.collection.Collection(api_client=self.arvrunner.api, @@ -149,21 +168,28 @@ class ArvadosContainer(JobBase): with Perf(metrics, "createfiles %s" % self.name): for f, p in sorteditems: if not p.target: - pass - elif p.type in ("File", "Directory", "WritableFile", "WritableDirectory"): + continue + + if p.target.startswith("/"): + dst = p.target[len(self.outdir)+1:] if p.target.startswith(self.outdir+"/") else p.target[1:] + else: + dst = p.target + + if p.type in ("File", "Directory", "WritableFile", "WritableDirectory"): if p.resolved.startswith("_:"): - vwd.mkdirs(p.target) + vwd.mkdirs(dst) else: source, path = self.arvrunner.fs_access.get_collection(p.resolved) - vwd.copy(path, p.target, source_collection=source) + vwd.copy(path or ".", dst, source_collection=source) elif p.type == "CreateFile": if self.arvrunner.secret_store.has_secret(p.resolved): - secret_mounts["%s/%s" % (self.outdir, p.target)] = { + mountpoint = p.target if p.target.startswith("/") else os.path.join(self.outdir, p.target) + secret_mounts[mountpoint] = { "kind": "text", "content": self.arvrunner.secret_store.retrieve(p.resolved) } else: - with vwd.open(p.target, "w") as n: + with vwd.open(dst, "w") as n: n.write(p.resolved) def keepemptydirs(p): @@ -178,22 +204,25 @@ class ArvadosContainer(JobBase): if not runtimeContext.current_container: runtimeContext.current_container = arvados_cwl.util.get_current_container(self.arvrunner.api, self.arvrunner.num_retries, logger) - info = arvados_cwl.util.get_intermediate_collection_info(self.name, runtimeContext.current_container, runtimeContext.intermediate_output_ttl) - vwd.save_new(name=info["name"], + vwd.save_new(name=intermediate_collection_info["name"], owner_uuid=runtimeContext.project_uuid, ensure_unique_name=True, - trash_at=info["trash_at"], - properties=info["properties"]) + trash_at=intermediate_collection_info["trash_at"], + properties=intermediate_collection_info["properties"]) prev = None for f, p in sorteditems: if (not p.target or self.arvrunner.secret_store.has_secret(p.resolved) or (prev is not None and p.target.startswith(prev))): continue - mountpoint = "%s/%s" % (self.outdir, p.target) + if p.target.startswith("/"): + dst = p.target[len(self.outdir)+1:] if p.target.startswith(self.outdir+"/") else p.target[1:] + else: + dst = p.target + mountpoint = p.target if p.target.startswith("/") else os.path.join(self.outdir, p.target) mounts[mountpoint] = {"kind": "collection", "portable_data_hash": vwd.portable_data_hash(), - "path": p.target} + "path": dst} if p.type.startswith("Writable"): mounts[mountpoint]["writable"] = True prev = p.target + "/" @@ -217,13 +246,11 @@ class ArvadosContainer(JobBase): "path": "%s/%s" % (self.outdir, self.stdout)} (docker_req, docker_is_req) = self.get_requirement("DockerRequirement") - if not docker_req: - docker_req = {"dockerImageId": "arvados/jobs"} container_request["container_image"] = arv_docker_get_image(self.arvrunner.api, docker_req, runtimeContext.pull_image, - runtimeContext.project_uuid) + runtimeContext) network_req, _ = self.get_requirement("NetworkAccess") if network_req: @@ -233,10 +260,22 @@ class ArvadosContainer(JobBase): if api_req: runtime_constraints["API"] = True + use_disk_cache = (self.arvrunner.api.config()["Containers"].get("DefaultKeepCacheRAM", 0) == 0) + + keep_cache_type_req, _ = self.get_requirement("http://arvados.org/cwl#KeepCacheTypeRequirement") + if keep_cache_type_req: + if "keepCacheType" in keep_cache_type_req: + if keep_cache_type_req["keepCacheType"] == "ram_cache": + use_disk_cache = False + runtime_req, _ = self.get_requirement("http://arvados.org/cwl#RuntimeConstraints") if runtime_req: if "keep_cache" in runtime_req: - runtime_constraints["keep_cache_ram"] = math.ceil(runtime_req["keep_cache"] * 2**20) + if use_disk_cache: + # If DefaultKeepCacheRAM is zero it means we should use disk cache. + runtime_constraints["keep_cache_disk"] = math.ceil(runtime_req["keep_cache"] * 2**20) + else: + runtime_constraints["keep_cache_ram"] = math.ceil(runtime_req["keep_cache"] * 2**20) if "outputDirType" in runtime_req: if runtime_req["outputDirType"] == "local_output_dir": # Currently the default behavior. @@ -260,6 +299,33 @@ class ArvadosContainer(JobBase): if self.output_ttl < 0: raise WorkflowException("Invalid value %d for output_ttl, cannot be less than zero" % container_request["output_ttl"]) + + if self.arvrunner.api._rootDesc["revision"] >= "20210628": + storage_class_req, _ = self.get_requirement("http://arvados.org/cwl#OutputStorageClass") + if storage_class_req and storage_class_req.get("intermediateStorageClass"): + container_request["output_storage_classes"] = aslist(storage_class_req["intermediateStorageClass"]) + else: + container_request["output_storage_classes"] = runtimeContext.intermediate_storage_classes.strip().split(",") + + cuda_req, _ = self.get_requirement("http://commonwl.org/cwltool#CUDARequirement") + if cuda_req: + runtime_constraints["cuda"] = { + "device_count": resources.get("cudaDeviceCount", 1), + "driver_version": cuda_req["cudaVersionMin"], + "hardware_capability": aslist(cuda_req["cudaComputeCapability"])[0] + } + + if runtimeContext.enable_preemptible is False: + scheduling_parameters["preemptible"] = False + else: + preemptible_req, _ = self.get_requirement("http://arvados.org/cwl#UsePreemptible") + if preemptible_req: + scheduling_parameters["preemptible"] = preemptible_req["usePreemptible"] + elif runtimeContext.enable_preemptible is True: + scheduling_parameters["preemptible"] = True + elif runtimeContext.enable_preemptible is None: + pass + if self.timelimit is not None and self.timelimit > 0: scheduling_parameters["max_run_time"] = self.timelimit @@ -267,7 +333,7 @@ class ArvadosContainer(JobBase): if runtimeContext.submit_runner_cluster: extra_submit_params["cluster_id"] = runtimeContext.submit_runner_cluster - container_request["output_name"] = "Output for step %s" % (self.name) + container_request["output_name"] = cleanup_name_for_collection("Output from step %s" % (self.name)) container_request["output_ttl"] = self.output_ttl container_request["mounts"] = mounts container_request["secret_mounts"] = secret_mounts @@ -276,11 +342,29 @@ class ArvadosContainer(JobBase): enable_reuse = runtimeContext.enable_reuse if enable_reuse: + reuse_req, _ = self.get_requirement("WorkReuse") + if reuse_req: + enable_reuse = reuse_req["enableReuse"] reuse_req, _ = self.get_requirement("http://arvados.org/cwl#ReuseRequirement") if reuse_req: enable_reuse = reuse_req["enableReuse"] container_request["use_existing"] = enable_reuse + properties_req, _ = self.get_requirement("http://arvados.org/cwl#ProcessProperties") + if properties_req: + for pr in properties_req["processProperties"]: + container_request["properties"][pr["propertyName"]] = self.builder.do_eval(pr["propertyValue"]) + + output_properties_req, _ = self.get_requirement("http://arvados.org/cwl#OutputCollectionProperties") + if output_properties_req: + if self.arvrunner.api._rootDesc["revision"] >= "20220510": + container_request["output_properties"] = {} + for pr in output_properties_req["outputProperties"]: + container_request["output_properties"][pr["propertyName"]] = self.builder.do_eval(pr["propertyValue"]) + else: + logger.warning("%s API revision is %s, revision %s is required to support setting properties on output collections.", + self.arvrunner.label(self), self.arvrunner.api._rootDesc["revision"], "20220510") + if runtimeContext.runnerjob.startswith("arvwf:"): wfuuid = runtimeContext.runnerjob[6:runtimeContext.runnerjob.index("#")] wfrecord = self.arvrunner.api.workflows().get(uuid=wfuuid).execute(num_retries=self.arvrunner.num_retries) @@ -310,8 +394,9 @@ class ArvadosContainer(JobBase): logger.info("%s reused container %s", self.arvrunner.label(self), response["container_uuid"]) else: logger.info("%s %s state is %s", self.arvrunner.label(self), response["uuid"], response["state"]) - except Exception: - logger.exception("%s got an error", self.arvrunner.label(self)) + except Exception as e: + logger.exception("%s error submitting container\n%s", self.arvrunner.label(self), e) + logger.debug("Container request was %s", container_request) self.output_callback({}, "permanentFail") def done(self, record): @@ -332,6 +417,10 @@ class ArvadosContainer(JobBase): processStatus = "success" else: processStatus = "permanentFail" + + if rcode == 137: + logger.warning("%s Container may have been killed for using too much RAM. Try resubmitting with a higher 'ramMin'.", + self.arvrunner.label(self)) else: processStatus = "permanentFail" @@ -358,6 +447,13 @@ class ArvadosContainer(JobBase): if container["output"]: outputs = done.done_outputs(self, container, "/tmp", self.outdir, "/keep") + + properties = record["properties"].copy() + properties["cwl_output"] = outputs + self.arvrunner.api.container_requests().update( + uuid=self.uuid, + body={"container_request": {"properties": properties}} + ).execute(num_retries=self.arvrunner.num_retries) except WorkflowException as e: # Only include a stack trace if in debug mode. # A stack trace may obfuscate more useful output about the workflow. @@ -374,7 +470,7 @@ class ArvadosContainer(JobBase): class RunnerContainer(Runner): """Submit and manage a container that runs arvados-cwl-runner.""" - def arvados_job_spec(self, runtimeContext): + def arvados_job_spec(self, runtimeContext, git_info): """Create an Arvados container request for this workflow. The returned dict can be used to create a container passed as @@ -401,7 +497,7 @@ class RunnerContainer(Runner): "cwd": "/var/spool/cwl", "priority": self.priority, "state": "Committed", - "container_image": arvados_jobs_image(self.arvrunner, self.jobs_image), + "container_image": arvados_jobs_image(self.arvrunner, self.jobs_image, runtimeContext), "mounts": { "/var/lib/cwl/cwl.input.json": { "kind": "json", @@ -422,7 +518,7 @@ class RunnerContainer(Runner): "ram": 1024*1024 * (math.ceil(self.submit_runner_ram) + math.ceil(self.collection_cache_size)), "API": True }, - "use_existing": self.enable_reuse, + "use_existing": False, # Never reuse the runner container - see #15497. "properties": {} } @@ -435,16 +531,33 @@ class RunnerContainer(Runner): "kind": "collection", "portable_data_hash": "%s" % workflowcollection } + elif self.embedded_tool.tool.get("id", "").startswith("arvwf:"): + uuid, frg = urllib.parse.urldefrag(self.embedded_tool.tool["id"]) + workflowpath = "/var/lib/cwl/workflow.json#" + frg + packedtxt = self.loadingContext.loader.fetch_text(uuid) + yaml = ruamel.yaml.YAML(typ='safe', pure=True) + packed = yaml.load(packedtxt) + container_req["mounts"]["/var/lib/cwl/workflow.json"] = { + "kind": "json", + "content": packed + } + container_req["properties"]["template_uuid"] = self.embedded_tool.tool["id"][6:33] else: - packed = packed_workflow(self.arvrunner, self.embedded_tool, self.merged_map) + main = self.loadingContext.loader.idx["_:main"] + #print(json.dumps(main, indent=2)) workflowpath = "/var/lib/cwl/workflow.json#main" container_req["mounts"]["/var/lib/cwl/workflow.json"] = { "kind": "json", - "content": packed + "content": main } - if self.embedded_tool.tool.get("id", "").startswith("arvwf:"): - container_req["properties"]["template_uuid"] = self.embedded_tool.tool["id"][6:33] + container_req["properties"].update({k.replace("http://arvados.org/cwl#", "arv:"): v for k, v in git_info.items()}) + + properties_req, _ = self.embedded_tool.get_requirement("http://arvados.org/cwl#ProcessProperties") + if properties_req: + builder = make_builder(self.job_order, self.embedded_tool.hints, self.embedded_tool.requirements, runtimeContext, self.embedded_tool.metadata) + for pr in properties_req["processProperties"]: + container_req["properties"][pr["propertyName"]] = builder.do_eval(pr["propertyValue"]) # --local means execute the workflow instead of submitting a container request # --api=containers means use the containers API @@ -459,6 +572,7 @@ class RunnerContainer(Runner): "--api=containers", "--no-log-timestamps", "--disable-validate", + "--disable-color", "--eval-timeout=%s" % self.arvrunner.eval_timeout, "--thread-count=%s" % self.arvrunner.thread_count, "--enable-reuse" if self.enable_reuse else "--disable-reuse", @@ -474,24 +588,42 @@ class RunnerContainer(Runner): if runtimeContext.debug: command.append("--debug") - if runtimeContext.storage_classes != "default": + if runtimeContext.storage_classes != "default" and runtimeContext.storage_classes: command.append("--storage-classes=" + runtimeContext.storage_classes) - if self.on_error: + if runtimeContext.intermediate_storage_classes != "default" and runtimeContext.intermediate_storage_classes: + command.append("--intermediate-storage-classes=" + runtimeContext.intermediate_storage_classes) + + if runtimeContext.on_error: command.append("--on-error=" + self.on_error) - if self.intermediate_output_ttl: - command.append("--intermediate-output-ttl=%d" % self.intermediate_output_ttl) + if runtimeContext.intermediate_output_ttl: + command.append("--intermediate-output-ttl=%d" % runtimeContext.intermediate_output_ttl) - if self.arvrunner.trash_intermediate: + if runtimeContext.trash_intermediate: command.append("--trash-intermediate") - if self.arvrunner.project_uuid: - command.append("--project-uuid="+self.arvrunner.project_uuid) + if runtimeContext.project_uuid: + command.append("--project-uuid="+runtimeContext.project_uuid) if self.enable_dev: command.append("--enable-dev") + if runtimeContext.enable_preemptible is True: + command.append("--enable-preemptible") + + if runtimeContext.enable_preemptible is False: + command.append("--disable-preemptible") + + if runtimeContext.varying_url_params: + command.append("--varying-url-params="+runtimeContext.varying_url_params) + + if runtimeContext.prefer_cached_downloads: + command.append("--prefer-cached-downloads") + + if self.fast_parser: + command.append("--fast-parser") + command.extend([workflowpath, "/var/lib/cwl/cwl.input.json"]) container_req["command"] = command @@ -501,9 +633,9 @@ class RunnerContainer(Runner): def run(self, runtimeContext): runtimeContext.keepprefix = "keep:" - job_spec = self.arvados_job_spec(runtimeContext) - if self.arvrunner.project_uuid: - job_spec["owner_uuid"] = self.arvrunner.project_uuid + job_spec = self.arvados_job_spec(runtimeContext, self.git_info) + if runtimeContext.project_uuid: + job_spec["owner_uuid"] = runtimeContext.project_uuid extra_submit_params = {} if runtimeContext.submit_runner_cluster: @@ -529,6 +661,17 @@ class RunnerContainer(Runner): logger.info("%s submitted container_request %s", self.arvrunner.label(self), response["uuid"]) + workbench1 = self.arvrunner.api.config()["Services"]["Workbench1"]["ExternalURL"] + workbench2 = self.arvrunner.api.config()["Services"]["Workbench2"]["ExternalURL"] + url = "" + if workbench2: + url = "{}processes/{}".format(workbench2, response["uuid"]) + elif workbench1: + url = "{}container_requests/{}".format(workbench1, response["uuid"]) + if url: + logger.info("Monitor workflow progress at %s", url) + + def done(self, record): try: container = self.arvrunner.api.containers().get(