X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/e1655f12ed1cfec6b5763e1db81e7ec8dd0f7a14..40ad5fb3f699c348222c90692b9ae27a8f716af9:/sdk/cwl/arvados_cwl/__init__.py

diff --git a/sdk/cwl/arvados_cwl/__init__.py b/sdk/cwl/arvados_cwl/__init__.py
index d534abd997..8f2102c6c3 100644
--- a/sdk/cwl/arvados_cwl/__init__.py
+++ b/sdk/cwl/arvados_cwl/__init__.py
@@ -1,5 +1,7 @@
 #!/usr/bin/env python
 
+# Implement cwl-runner interface for submitting and running jobs on Arvados.
+
 import argparse
 import arvados
 import arvados.events
@@ -20,6 +22,8 @@ import re
 import os
 import sys
 import functools
+import json
+import pkg_resources  # part of setuptools
 
 from cwltool.process import get_feature, adjustFiles, scandeps
 from arvados.api import OrderedJsonModel
@@ -27,16 +31,14 @@ from arvados.api import OrderedJsonModel
 logger = logging.getLogger('arvados.cwl-runner')
 logger.setLevel(logging.INFO)
 
-crunchrunner_pdh = "83db29f08544e1c319572a6bd971088a+140"
-crunchrunner_download = "https://cloud.curoverse.com/collections/download/qr1hi-4zz18-n3m1yxd0vx78jic/1i1u2qtq66k1atziv4ocfgsg5nu5tj11n4r6e0bhvjg03rix4m/crunchrunner"
-certs_download = "https://cloud.curoverse.com/collections/download/qr1hi-4zz18-n3m1yxd0vx78jic/1i1u2qtq66k1atziv4ocfgsg5nu5tj11n4r6e0bhvjg03rix4m/ca-certificates.crt"
-
 tmpdirre = re.compile(r"^\S+ \S+ \d+ \d+ stderr \S+ \S+ crunchrunner: \$\(task\.tmpdir\)=(.*)")
 outdirre = re.compile(r"^\S+ \S+ \d+ \d+ stderr \S+ \S+ crunchrunner: \$\(task\.outdir\)=(.*)")
 keepre = re.compile(r"^\S+ \S+ \d+ \d+ stderr \S+ \S+ crunchrunner: \$\(task\.keep\)=(.*)")
 
 
 def arv_docker_get_image(api_client, dockerRequirement, pull_image, project_uuid):
+    """Check if a Docker image is available in Keep, if not, upload it using arv-keepdocker."""
+
     if "dockerImageId" not in dockerRequirement and "dockerPull" in dockerRequirement:
         dockerRequirement["dockerImageId"] = dockerRequirement["dockerPull"]
 
@@ -54,12 +56,14 @@ def arv_docker_get_image(api_client, dockerRequirement, pull_image, project_uuid
         if image_tag:
             args.append(image_tag)
         logger.info("Uploading Docker image %s", ":".join(args[1:]))
-        arvados.commands.keepdocker.main(args)
+        arvados.commands.keepdocker.main(args, stdout=sys.stderr)
 
     return dockerRequirement["dockerImageId"]
 
 
 class CollectionFsAccess(cwltool.process.StdFsAccess):
+    """Implement the cwltool FsAccess interface for Arvados Collections."""
+
     def __init__(self, basedir):
         self.collections = {}
         self.basedir = basedir
@@ -116,6 +120,8 @@ class CollectionFsAccess(cwltool.process.StdFsAccess):
             return os.path.exists(self._abs(fn))
 
 class ArvadosJob(object):
+    """Submit and manage a Crunch job for executing a CWL CommandLineTool."""
+
     def __init__(self, runner):
         self.arvrunner = runner
         self.running = False
@@ -153,6 +159,8 @@ class ArvadosJob(object):
         (docker_req, docker_is_req) = get_feature(self, "DockerRequirement")
         if docker_req and kwargs.get("use_container") is not False:
             runtime_constraints["docker_image"] = arv_docker_get_image(self.arvrunner.api, docker_req, pull_image, self.arvrunner.project_uuid)
+        else:
+            runtime_constraints["docker_image"] = "arvados/jobs"
 
         resources = self.builder.resources
         if resources is not None:
@@ -160,16 +168,26 @@ class ArvadosJob(object):
             runtime_constraints["min_ram_mb_per_node"] = resources.get("ram")
             runtime_constraints["min_scratch_mb_per_node"] = resources.get("tmpdirSize", 0) + resources.get("outdirSize", 0)
 
+        filters = [["repository", "=", "arvados"],
+                   ["script", "=", "crunchrunner"],
+                   ["script_version", "in git", "9e5b98e8f5f4727856b53447191f9c06e3da2ba6"]]
+        if not self.arvrunner.ignore_docker_for_reuse:
+            filters.append(["docker_image_locator", "in docker", runtime_constraints["docker_image"]])
+
         try:
-            response = self.arvrunner.api.jobs().create(body={
-                "owner_uuid": self.arvrunner.project_uuid,
-                "script": "crunchrunner",
-                "repository": "arvados",
-                "script_version": "master",
-                "minimum_script_version": "9e5b98e8f5f4727856b53447191f9c06e3da2ba6",
-                "script_parameters": {"tasks": [script_parameters], "crunchrunner": crunchrunner_pdh+"/crunchrunner"},
-                "runtime_constraints": runtime_constraints
-            }, find_or_create=kwargs.get("enable_reuse", True)).execute(num_retries=self.arvrunner.num_retries)
+            response = self.arvrunner.api.jobs().create(
+                body={
+                    "owner_uuid": self.arvrunner.project_uuid,
+                    "script": "crunchrunner",
+                    "repository": "arvados",
+                    "script_version": "master",
+                    "minimum_script_version": "9e5b98e8f5f4727856b53447191f9c06e3da2ba6",
+                    "script_parameters": {"tasks": [script_parameters]},
+                    "runtime_constraints": runtime_constraints
+                },
+                filters=filters,
+                find_or_create=kwargs.get("enable_reuse", True)
+            ).execute(num_retries=self.arvrunner.num_retries)
 
             self.arvrunner.jobs[response["uuid"]] = self
 
@@ -214,7 +232,15 @@ class ArvadosJob(object):
                     tmpdir = None
                     outdir = None
                     keepdir = None
-                    for l in log.readlines():
+                    for l in log:
+                        # Determine the tmpdir, outdir and keepdir paths from
+                        # the job run.  Unfortunately, we can't take the first
+                        # values we find (which are expected to be near the
+                        # top) and stop scanning because if the node fails and
+                        # the job restarts on a different node these values
+                        # will different runs, and we need to know about the
+                        # final run that actually produced output.
+
                         g = tmpdirre.match(l)
                         if g:
                             tmpdir = g.group(1)
@@ -225,13 +251,39 @@ class ArvadosJob(object):
                         if g:
                             keepdir = g.group(1)
 
-                        # It turns out if the job fails and restarts it can
-                        # come up on a different compute node, so we have to
-                        # read the log to the end to be sure instead of taking the
-                        # easy way out.
-                        #
-                        #if tmpdir and outdir and keepdir:
-                        #    break
+                    colname = "Output %s of %s" % (record["output"][0:7], self.name)
+
+                    # check if collection already exists with same owner, name and content
+                    collection_exists = self.arvrunner.api.collections().list(
+                        filters=[["owner_uuid", "=", self.arvrunner.project_uuid],
+                                 ['portable_data_hash', '=', record["output"]],
+                                 ["name", "=", colname]]
+                    ).execute(num_retries=self.arvrunner.num_retries)
+
+                    if not collection_exists["items"]:
+                        # Create a collection located in the same project as the
+                        # pipeline with the contents of the output.
+                        # First, get output record.
+                        collections = self.arvrunner.api.collections().list(
+                            limit=1,
+                            filters=[['portable_data_hash', '=', record["output"]]],
+                            select=["manifest_text"]
+                        ).execute(num_retries=self.arvrunner.num_retries)
+
+                        if not collections["items"]:
+                            raise WorkflowException(
+                                "Job output '%s' cannot be found on API server" % (
+                                    record["output"]))
+
+                        # Create new collection in the parent project
+                        # with the output contents.
+                        self.arvrunner.api.collections().create(body={
+                            "owner_uuid": self.arvrunner.project_uuid,
+                            "name": colname,
+                            "portable_data_hash": record["output"],
+                            "manifest_text": collections["items"][0]["manifest_text"]
+                        }, ensure_unique_name=True).execute(
+                            num_retries=self.arvrunner.num_retries)
 
                     self.builder.outdir = outdir
                     self.builder.pathmapper.keepdir = keepdir
@@ -248,24 +300,129 @@ class ArvadosJob(object):
             del self.arvrunner.jobs[record["uuid"]]
 
 
+class RunnerJob(object):
+    """Submit and manage a Crunch job that runs crunch_scripts/cwl-runner."""
+
+    def __init__(self, runner, tool, job_order, enable_reuse):
+        self.arvrunner = runner
+        self.tool = tool
+        self.job_order = job_order
+        self.running = False
+        self.enable_reuse = enable_reuse
+
+    def update_pipeline_component(self, record):
+        pass
+
+    def upload_docker(self, tool):
+        if isinstance(tool, cwltool.draft2tool.CommandLineTool):
+            (docker_req, docker_is_req) = get_feature(tool, "DockerRequirement")
+            if docker_req:
+                arv_docker_get_image(self.arvrunner.api, docker_req, True, self.arvrunner.project_uuid)
+        elif isinstance(tool, cwltool.workflow.Workflow):
+            for s in tool.steps:
+                self.upload_docker(s.embedded_tool)
+
+    def run(self, dry_run=False, pull_image=True, **kwargs):
+        self.upload_docker(self.tool)
+
+        workflowfiles = set()
+        jobfiles = set()
+        workflowfiles.add(self.tool.tool["id"])
+
+        self.name = os.path.basename(self.tool.tool["id"])
+
+        def visitFiles(files, path):
+            files.add(path)
+            return path
+
+        document_loader, _, _ = cwltool.process.get_schema()
+        def loadref(b, u):
+            return document_loader.resolve_ref(u, base_url=b)[0]
+
+        sc = scandeps("", self.tool.tool,
+                      set(("$import", "run")),
+                      set(("$include", "$schemas", "path")),
+                      loadref)
+        adjustFiles(sc, functools.partial(visitFiles, workflowfiles))
+        adjustFiles(self.job_order, functools.partial(visitFiles, jobfiles))
+
+        workflowmapper = ArvPathMapper(self.arvrunner, workflowfiles, "",
+                                       "%s",
+                                       "%s/%s",
+                                       name=self.name,
+                                       **kwargs)
+
+        jobmapper = ArvPathMapper(self.arvrunner, jobfiles, "",
+                                  "%s",
+                                  "%s/%s",
+                                  name=os.path.basename(self.job_order.get("id", "#")),
+                                  **kwargs)
+
+        adjustFiles(self.job_order, lambda p: jobmapper.mapper(p)[1])
+
+        if "id" in self.job_order:
+            del self.job_order["id"]
+
+        self.job_order["cwl:tool"] = workflowmapper.mapper(self.tool.tool["id"])[1]
+
+        response = self.arvrunner.api.jobs().create(body={
+            "owner_uuid": self.arvrunner.project_uuid,
+            "script": "cwl-runner",
+            "script_version": "master",
+            "repository": "arvados",
+            "script_parameters": self.job_order,
+            "runtime_constraints": {
+                "docker_image": "arvados/jobs"
+            }
+        }, find_or_create=self.enable_reuse).execute(num_retries=self.arvrunner.num_retries)
+
+        self.arvrunner.jobs[response["uuid"]] = self
+
+        logger.info("Submitted job %s", response["uuid"])
+
+        if response["state"] in ("Complete", "Failed", "Cancelled"):
+            self.done(response)
+
+    def done(self, record):
+        if record["state"] == "Complete":
+            processStatus = "success"
+        else:
+            processStatus = "permanentFail"
+
+        outputs = None
+        try:
+            try:
+                outc = arvados.collection.Collection(record["output"])
+                with outc.open("cwl.output.json") as f:
+                    outputs = json.load(f)
+            except Exception as e:
+                logger.error("While getting final output object: %s", e)
+            self.arvrunner.output_callback(outputs, processStatus)
+        finally:
+            del self.arvrunner.jobs[record["uuid"]]
+
 class ArvPathMapper(cwltool.pathmapper.PathMapper):
+    """Convert container-local paths to and from Keep collection ids."""
+
     def __init__(self, arvrunner, referenced_files, basedir,
                  collection_pattern, file_pattern, name=None, **kwargs):
         self._pathmap = arvrunner.get_uploaded()
-        uploadfiles = []
+        uploadfiles = set()
 
         pdh_path = re.compile(r'^keep:[0-9a-f]{32}\+\d+/.+')
 
         for src in referenced_files:
             if isinstance(src, basestring) and pdh_path.match(src):
                 self._pathmap[src] = (src, collection_pattern % src[5:])
+            if "#" in src:
+                src = src[:src.index("#")]
             if src not in self._pathmap:
                 ab = cwltool.pathmapper.abspath(src, basedir)
                 st = arvados.commands.run.statfile("", ab, fnPattern=file_pattern)
                 if kwargs.get("conformance_test"):
                     self._pathmap[src] = (src, ab)
                 elif isinstance(st, arvados.commands.run.UploadFile):
-                    uploadfiles.append((src, ab, st))
+                    uploadfiles.add((src, ab, st))
                 elif isinstance(st, arvados.commands.run.ArvFile):
                     self._pathmap[src] = (ab, st.fn)
                 else:
@@ -296,6 +453,8 @@ class ArvPathMapper(cwltool.pathmapper.PathMapper):
 
 
 class ArvadosCommandTool(cwltool.draft2tool.CommandLineTool):
+    """Wrap cwltool CommandLineTool to override selected methods."""
+
     def __init__(self, arvrunner, toolpath_object, **kwargs):
         super(ArvadosCommandTool, self).__init__(toolpath_object, **kwargs)
         self.arvrunner = arvrunner
@@ -311,6 +470,9 @@ class ArvadosCommandTool(cwltool.draft2tool.CommandLineTool):
 
 
 class ArvCwlRunner(object):
+    """Execute a CWL tool or workflow, submit crunch jobs, wait for them to
+    complete, and report output."""
+
     def __init__(self, api_client):
         self.api = api_client
         self.jobs = {}
@@ -329,13 +491,15 @@ class ArvCwlRunner(object):
     def output_callback(self, out, processStatus):
         if processStatus == "success":
             logger.info("Overall job status is %s", processStatus)
-            self.api.pipeline_instances().update(uuid=self.pipeline["uuid"],
-                                                 body={"state": "Complete"}).execute(num_retries=self.num_retries)
+            if self.pipeline:
+                self.api.pipeline_instances().update(uuid=self.pipeline["uuid"],
+                                                     body={"state": "Complete"}).execute(num_retries=self.num_retries)
 
         else:
             logger.warn("Overall job status is %s", processStatus)
-            self.api.pipeline_instances().update(uuid=self.pipeline["uuid"],
-                                                 body={"state": "Failed"}).execute(num_retries=self.num_retries)
+            if self.pipeline:
+                self.api.pipeline_instances().update(uuid=self.pipeline["uuid"],
+                                                     body={"state": "Failed"}).execute(num_retries=self.num_retries)
         self.final_output = out
 
 
@@ -366,97 +530,27 @@ class ArvCwlRunner(object):
     def add_uploaded(self, src, pair):
         self.uploaded[src] = pair
 
-    def upload_docker(self, tool):
-        if isinstance(tool, cwltool.draft2tool.CommandLineTool):
-            (docker_req, docker_is_req) = get_feature(tool, "DockerRequirement")
-            if docker_req:
-                arv_docker_get_image(self.api, docker_req, True, self.project_uuid)
-        elif isinstance(tool, cwltool.workflow.Workflow):
-            for s in tool.steps:
-                self.upload_docker(s.embedded_tool)
-
-    def submit(self, tool, job_order, input_basedir, args, **kwargs):
-        self.upload_docker(tool)
-
-        workflowfiles = set()
-        jobfiles = set()
-        workflowfiles.add(tool.tool["id"])
-
-        def visitFiles(files, path):
-            files.add(path)
-            return path
-
-        document_loader, _, _ = cwltool.process.get_schema()
-        def loadref(b, u):
-            return document_loader.resolve_ref(u, base_url=b)[0]
-
-        adjustFiles(scandeps("", tool.tool,
-                             set(("run",)),
-                             set(("$schemas", "path")),
-                             loadref),
-                    functools.partial(visitFiles, workflowfiles))
-        adjustFiles(job_order, functools.partial(visitFiles, jobfiles))
-
-        workflowmapper = ArvPathMapper(self, workflowfiles, "",
-                                       "%s",
-                                       "%s/%s",
-                                       name=os.path.basename(tool.tool["id"]),
-                                       **kwargs)
-
-        jobmapper = ArvPathMapper(self, jobfiles, "",
-                                  "%s",
-                                  "%s/%s",
-                                  name=os.path.basename(job_order.get("id", "#")),
-                                  **kwargs)
-
-        adjustFiles(job_order, lambda p: jobmapper.mapper(p)[1])
-
-        if "id" in job_order:
-            del job_order["id"]
-
-        job_order["cwl:tool"] = workflowmapper.mapper(tool.tool["id"])[1]
-
-        response = self.api.jobs().create(body={
-            "script": "cwl-runner",
-            "script_version": "8654-arv-jobs-cwl-runner",
-            "repository": "arvados",
-            "script_parameters": job_order,
-            "runtime_constraints": {
-                "docker_image": "arvados/jobs"
-            }
-        }, find_or_create=args.enable_reuse).execute(num_retries=self.num_retries)
-
-        logger.info("Submitted job %s", response["uuid"])
-
-        return
+    def arvExecutor(self, tool, job_order, input_basedir, args, **kwargs):
+        self.debug = args.debug
 
+        if args.quiet:
+            logger.setLevel(logging.WARN)
+            logging.getLogger('arvados.arv-run').setLevel(logging.WARN)
 
-    def arvExecutor(self, tool, job_order, input_basedir, args, **kwargs):
         useruuid = self.api.users().current().execute()["uuid"]
         self.project_uuid = args.project_uuid if args.project_uuid else useruuid
+        self.pipeline = None
 
         if args.submit:
-            return self.submit(tool, job_order, input_basedir, args, **kwargs)
+            runnerjob = RunnerJob(self, tool, job_order, args.enable_reuse)
+            if not args.wait:
+                runnerjob.run()
+                return
 
         events = arvados.events.subscribe(arvados.api('v1'), [["object_uuid", "is_a", "arvados#job"]], self.on_message)
 
         self.debug = args.debug
-
-        try:
-            self.api.collections().get(uuid=crunchrunner_pdh).execute()
-        except arvados.errors.ApiError as e:
-            import httplib2
-            h = httplib2.Http(ca_certs=arvados.util.ca_certs_path())
-            resp, content = h.request(crunchrunner_download, "GET")
-            resp2, content2 = h.request(certs_download, "GET")
-            with arvados.collection.Collection() as col:
-                with col.open("crunchrunner", "w") as f:
-                    f.write(content)
-                with col.open("ca-certificates.crt", "w") as f:
-                    f.write(content2)
-
-                col.save_new("crunchrunner binary", ensure_unique_name=True)
-
+        self.ignore_docker_for_reuse = args.ignore_docker_for_reuse
         self.fs_access = CollectionFsAccess(input_basedir)
 
         kwargs["fs_access"] = self.fs_access
@@ -468,24 +562,27 @@ class ArvCwlRunner(object):
         if kwargs.get("conformance_test"):
             return cwltool.main.single_job_executor(tool, job_order, input_basedir, args, **kwargs)
         else:
-            components = {}
-            if "cwl_runner_job" in kwargs:
-                components[os.path.basename(tool.tool["id"])] = {"job": kwargs["cwl_runner_job"]}
+            if args.submit:
+                jobiter = iter((runnerjob,))
+            else:
+                components = {}
+                if "cwl_runner_job" in kwargs:
+                    components[os.path.basename(tool.tool["id"])] = {"job": kwargs["cwl_runner_job"]}
 
-            self.pipeline = self.api.pipeline_instances().create(
-                body={
-                    "owner_uuid": self.project_uuid,
-                    "name": shortname(tool.tool["id"]),
-                    "components": components,
-                    "state": "RunningOnClient"}).execute(num_retries=self.num_retries)
+                self.pipeline = self.api.pipeline_instances().create(
+                    body={
+                        "owner_uuid": self.project_uuid,
+                        "name": shortname(tool.tool["id"]),
+                        "components": components,
+                        "state": "RunningOnClient"}).execute(num_retries=self.num_retries)
 
-            logger.info("Pipeline instance %s", self.pipeline["uuid"])
+                logger.info("Pipeline instance %s", self.pipeline["uuid"])
 
-            jobiter = tool.job(job_order,
-                               input_basedir,
-                               self.output_callback,
-                               docker_outdir="$(task.outdir)",
-                               **kwargs)
+                jobiter = tool.job(job_order,
+                                   input_basedir,
+                                   self.output_callback,
+                                   docker_outdir="$(task.outdir)",
+                                   **kwargs)
 
             try:
                 self.cond.acquire()
@@ -517,17 +614,29 @@ class ArvCwlRunner(object):
                     logger.error("Interrupted, marking pipeline as failed")
                 else:
                     logger.error("Caught unhandled exception, marking pipeline as failed.  Error was: %s", sys.exc_info()[0], exc_info=(sys.exc_info()[1] if self.debug else False))
-                self.api.pipeline_instances().update(uuid=self.pipeline["uuid"],
-                                                     body={"state": "Failed"}).execute(num_retries=self.num_retries)
+                if self.pipeline:
+                    self.api.pipeline_instances().update(uuid=self.pipeline["uuid"],
+                                                         body={"state": "Failed"}).execute(num_retries=self.num_retries)
             finally:
                 self.cond.release()
 
             return self.final_output
 
+def versionstring():
+    """Print version string of key packages for provenance and debugging."""
+
+    arvcwlpkg = pkg_resources.require("arvados-cwl-runner")
+    arvpkg = pkg_resources.require("arvados-python-client")
+    cwlpkg = pkg_resources.require("cwltool")
+
+    return "%s %s, %s %s, %s %s" % (sys.argv[0], arvcwlpkg[0].version,
+                                    "arvados-python-client", arvpkg[0].version,
+                                    "cwltool", cwlpkg[0].version)
 
 def main(args, stdout, stderr, api_client=None):
     args.insert(0, "--leave-outputs")
     parser = cwltool.main.arg_parser()
+
     exgroup = parser.add_mutually_exclusive_group()
     exgroup.add_argument("--enable-reuse", action="store_true",
                         default=True, dest="enable_reuse",
@@ -535,10 +644,24 @@ def main(args, stdout, stderr, api_client=None):
     exgroup.add_argument("--disable-reuse", action="store_false",
                         default=True, dest="enable_reuse",
                         help="")
+
     parser.add_argument("--project-uuid", type=str, help="Project that will own the workflow jobs")
-    parser.add_argument("--submit", action="store_true", help="Submit job and print job uuid.",
+    parser.add_argument("--ignore-docker-for-reuse", action="store_true",
+                        help="Ignore Docker image version when deciding whether to reuse past jobs.",
                         default=False)
 
+    exgroup = parser.add_mutually_exclusive_group()
+    exgroup.add_argument("--submit", action="store_true", help="Submit workflow to run on Arvados.",
+                        default=True, dest="submit")
+    exgroup.add_argument("--local", action="store_false", help="Run workflow on local host (submits jobs to Arvados).",
+                        default=True, dest="submit")
+
+    exgroup = parser.add_mutually_exclusive_group()
+    exgroup.add_argument("--wait", action="store_true", help="After submitting workflow runner job, wait for completion.",
+                        default=True, dest="wait")
+    exgroup.add_argument("--no-wait", action="store_false", help="Submit workflow runner job and exit.",
+                        default=True, dest="wait")
+
     try:
         if api_client is None:
             api_client=arvados.api('v1', model=OrderedJsonModel())
@@ -552,4 +675,5 @@ def main(args, stdout, stderr, api_client=None):
                              stderr=stderr,
                              executor=runner.arvExecutor,
                              makeTool=runner.arvMakeTool,
-                             parser=parser)
+                             parser=parser,
+                             versionfunc=versionstring)