Merge branch '10081-cwl-run-same-job' into 9988-cwl-arv-hints

[arvados.git] / sdk / cwl / arvados_cwl / __init__.py

diff --git a/sdk/cwl/arvados_cwl/__init__.py b/sdk/cwl/arvados_cwl/__init__.py
index cf968809b9016d1df1ab408bd75b4192b4ad1eb5..cd38003daef65d1c64007818c193ce2bd3398513 100644 (file)
--- a/sdk/cwl/arvados_cwl/__init__.py
+++ b/sdk/cwl/arvados_cwl/__init__.py
@@ -8,21 +8,29 @@ import logging
 import os
 import sys
 import threading
 import os
 import sys
 import threading

+import hashlib
+from functools import partial

 import pkg_resources  # part of setuptools
 
 from cwltool.errors import WorkflowException
 import cwltool.main
 import cwltool.workflow
 import pkg_resources  # part of setuptools
 
 from cwltool.errors import WorkflowException
 import cwltool.main
 import cwltool.workflow

+import schema_salad

 
 import arvados
 
 import arvados

-import arvados.events
+import arvados.config

 
 from .arvcontainer import ArvadosContainer, RunnerContainer
 from .arvjob import ArvadosJob, RunnerJob, RunnerTemplate
 from .arvtool import ArvadosCommandTool
 
 from .arvcontainer import ArvadosContainer, RunnerContainer
 from .arvjob import ArvadosJob, RunnerJob, RunnerTemplate
 from .arvtool import ArvadosCommandTool

+from .arvworkflow import ArvadosWorkflow, upload_workflow

 from .fsaccess import CollectionFsAccess
 from .fsaccess import CollectionFsAccess

+from .perf import Perf
+from cwltool.pack import pack

 
 from cwltool.process import shortname, UnsupportedRequirement
 
 from cwltool.process import shortname, UnsupportedRequirement

+from cwltool.pathmapper import adjustFileObjs
+from cwltool.draft2tool import compute_checksums

 from arvados.api import OrderedJsonModel
 
 logger = logging.getLogger('arvados.cwl-runner')
 from arvados.api import OrderedJsonModel
 
 logger = logging.getLogger('arvados.cwl-runner')


@@ -45,6 +53,9 @@ class ArvCwlRunner(object):
         self.num_retries = 4
         self.uuid = None
         self.work_api = work_api
         self.num_retries = 4
         self.uuid = None
         self.work_api = work_api

+        self.stop_polling = threading.Event()
+        self.poll_api = None
+        self.pipeline = None

 
         if self.work_api is None:
             # todo: autodetect API to use.
 
         if self.work_api is None:
             # todo: autodetect API to use.


@@ -53,9 +64,12 @@ class ArvCwlRunner(object):
         if self.work_api not in ("containers", "jobs"):
             raise Exception("Unsupported API '%s'" % self.work_api)
 
         if self.work_api not in ("containers", "jobs"):
             raise Exception("Unsupported API '%s'" % self.work_api)
 

-    def arvMakeTool(self, toolpath_object, **kwargs):
+    def arv_make_tool(self, toolpath_object, **kwargs):
+        kwargs["work_api"] = self.work_api

         if "class" in toolpath_object and toolpath_object["class"] == "CommandLineTool":
         if "class" in toolpath_object and toolpath_object["class"] == "CommandLineTool":

-            return ArvadosCommandTool(self, toolpath_object, work_api=self.work_api, **kwargs)
+            return ArvadosCommandTool(self, toolpath_object, **kwargs)
+        elif "class" in toolpath_object and toolpath_object["class"] == "Workflow":
+            return ArvadosWorkflow(self, toolpath_object, **kwargs)

         else:
             return cwltool.workflow.defaultMakeTool(toolpath_object, **kwargs)
 
         else:
             return cwltool.workflow.defaultMakeTool(toolpath_object, **kwargs)
 


@@ -89,28 +103,73 @@ class ArvCwlRunner(object):
                         self.cond.acquire()
                         j = self.processes[uuid]
                         logger.info("Job %s (%s) is %s", j.name, uuid, event["properties"]["new_attributes"]["state"])
                         self.cond.acquire()
                         j = self.processes[uuid]
                         logger.info("Job %s (%s) is %s", j.name, uuid, event["properties"]["new_attributes"]["state"])

-                        j.done(event["properties"]["new_attributes"])
+                        with Perf(logger, "done %s" % j.name):
+                            j.done(event["properties"]["new_attributes"])

                         self.cond.notify()
                     finally:
                         self.cond.release()
 
                         self.cond.notify()
                     finally:
                         self.cond.release()
 

+    def poll_states(self):
+        """Poll status of jobs or containers listed in the processes dict.
+
+        Runs in a separate thread.
+        """
+
+        while True:
+            self.stop_polling.wait(15)
+            if self.stop_polling.is_set():
+                break
+            with self.lock:
+                keys = self.processes.keys()
+            if not keys:
+                continue
+
+            if self.work_api == "containers":
+                table = self.poll_api.containers()
+            elif self.work_api == "jobs":
+                table = self.poll_api.jobs()
+
+            try:
+                proc_states = table.list(filters=[["uuid", "in", keys]]).execute(num_retries=self.num_retries)
+            except Exception as e:
+                logger.warn("Error checking states on API server: %s", e)
+                continue
+
+            for p in proc_states["items"]:
+                self.on_message({
+                    "object_uuid": p["uuid"],
+                    "event_type": "update",
+                    "properties": {
+                        "new_attributes": p
+                    }
+                })
+

     def get_uploaded(self):
         return self.uploaded.copy()
 
     def add_uploaded(self, src, pair):
         self.uploaded[src] = pair
 
     def get_uploaded(self):
         return self.uploaded.copy()
 
     def add_uploaded(self, src, pair):
         self.uploaded[src] = pair
 

-    def arvExecutor(self, tool, job_order, **kwargs):
+    def check_writable(self, obj):
+        if isinstance(obj, dict):
+            if obj.get("writable"):
+                raise UnsupportedRequirement("InitialWorkDir feature 'writable: true' not supported")
+            for v in obj.itervalues():
+                self.check_writable(v)
+        if isinstance(obj, list):
+            for v in obj:
+                self.check_writable(v)
+
+    def arv_executor(self, tool, job_order, **kwargs):

         self.debug = kwargs.get("debug")
 
         self.debug = kwargs.get("debug")
 

-        if kwargs.get("quiet"):
-            logger.setLevel(logging.WARN)
-            logging.getLogger('arvados.arv-run').setLevel(logging.WARN)
+        tool.visit(self.check_writable)

 
         useruuid = self.api.users().current().execute()["uuid"]
         self.project_uuid = kwargs.get("project_uuid") if kwargs.get("project_uuid") else useruuid
         self.pipeline = None
 
         useruuid = self.api.users().current().execute()["uuid"]
         self.project_uuid = kwargs.get("project_uuid") if kwargs.get("project_uuid") else useruuid
         self.pipeline = None

-        self.fs_access = CollectionFsAccess(kwargs["basedir"], api_client=self.api)
+        make_fs_access = kwargs.get("make_fs_access") or partial(CollectionFsAccess, api_client=self.api)
+        self.fs_access = make_fs_access(kwargs["basedir"])

 
         if kwargs.get("create_template"):
             tmpl = RunnerTemplate(self, tool, job_order, kwargs.get("enable_reuse"))
 
         if kwargs.get("create_template"):
             tmpl = RunnerTemplate(self, tool, job_order, kwargs.get("enable_reuse"))


@@ -118,19 +177,23 @@ class ArvCwlRunner(object):
             # cwltool.main will write our return value to stdout.
             return tmpl.uuid
 
             # cwltool.main will write our return value to stdout.
             return tmpl.uuid
 

-        self.debug = kwargs.get("debug")
+        if kwargs.get("create_workflow") or kwargs.get("update_workflow"):
+            return upload_workflow(self, tool, job_order, self.project_uuid, kwargs.get("update_workflow"))
+

         self.ignore_docker_for_reuse = kwargs.get("ignore_docker_for_reuse")
 
         self.ignore_docker_for_reuse = kwargs.get("ignore_docker_for_reuse")
 

-        kwargs["fs_access"] = self.fs_access
+        kwargs["make_fs_access"] = make_fs_access

         kwargs["enable_reuse"] = kwargs.get("enable_reuse")
         kwargs["use_container"] = True
         kwargs["tmpdir_prefix"] = "tmp"
         kwargs["on_error"] = "continue"
         kwargs["enable_reuse"] = kwargs.get("enable_reuse")
         kwargs["use_container"] = True
         kwargs["tmpdir_prefix"] = "tmp"
         kwargs["on_error"] = "continue"

+        kwargs["compute_checksum"] = kwargs.get("compute_checksum")

 
         if self.work_api == "containers":
             kwargs["outdir"] = "/var/spool/cwl"
             kwargs["docker_outdir"] = "/var/spool/cwl"
             kwargs["tmpdir"] = "/tmp"
 
         if self.work_api == "containers":
             kwargs["outdir"] = "/var/spool/cwl"
             kwargs["docker_outdir"] = "/var/spool/cwl"
             kwargs["tmpdir"] = "/tmp"

+            kwargs["docker_tmpdir"] = "/tmp"

         elif self.work_api == "jobs":
             kwargs["outdir"] = "$(task.outdir)"
             kwargs["docker_outdir"] = "$(task.outdir)"
         elif self.work_api == "jobs":
             kwargs["outdir"] = "$(task.outdir)"
             kwargs["docker_outdir"] = "$(task.outdir)"


@@ -162,10 +225,9 @@ class ArvCwlRunner(object):
             runnerjob.run()
             return runnerjob.uuid
 
             runnerjob.run()
             return runnerjob.uuid
 

-        if self.work_api == "containers":
-            events = arvados.events.subscribe(arvados.api('v1'), [["object_uuid", "is_a", "arvados#container"]], self.on_message)
-        if self.work_api == "jobs":
-            events = arvados.events.subscribe(arvados.api('v1'), [["object_uuid", "is_a", "arvados#job"]], self.on_message)
+        self.poll_api = arvados.api('v1')
+        self.polling_thread = threading.Thread(target=self.poll_states)
+        self.polling_thread.start()

 
         if runnerjob:
             jobiter = iter((runnerjob,))
 
         if runnerjob:
             jobiter = iter((runnerjob,))


@@ -184,7 +246,8 @@ class ArvCwlRunner(object):
 
             for runnable in jobiter:
                 if runnable:
 
             for runnable in jobiter:
                 if runnable:

-                    runnable.run(**kwargs)
+                    with Perf(logger, "run"):
+                        runnable.run(**kwargs)

                 else:
                     if self.processes:
                         self.cond.wait(1)
                 else:
                     if self.processes:
                         self.cond.wait(1)


@@ -195,7 +258,6 @@ class ArvCwlRunner(object):
             while self.processes:
                 self.cond.wait(1)
 
             while self.processes:
                 self.cond.wait(1)
 

-            events.close()

         except UnsupportedRequirement:
             raise
         except:
         except UnsupportedRequirement:
             raise
         except:


@@ -211,6 +273,8 @@ class ArvCwlRunner(object):
                                                      body={"priority": "0"}).execute(num_retries=self.num_retries)
         finally:
             self.cond.release()
                                                      body={"priority": "0"}).execute(num_retries=self.num_retries)
         finally:
             self.cond.release()

+            self.stop_polling.set()
+            self.polling_thread.join()

 
         if self.final_status == "UnsupportedRequirement":
             raise UnsupportedRequirement("Check log for details.")
 
         if self.final_status == "UnsupportedRequirement":
             raise UnsupportedRequirement("Check log for details.")


@@ -221,6 +285,9 @@ class ArvCwlRunner(object):
         if self.final_output is None:
             raise WorkflowException("Workflow did not return a result.")
 
         if self.final_output is None:
             raise WorkflowException("Workflow did not return a result.")
 

+        if kwargs.get("compute_checksum"):
+            adjustFileObjs(self.final_output, partial(compute_checksums, self.fs_access))
+

         return self.final_output
 
 
         return self.final_output
 
 


@@ -265,7 +332,7 @@ def arg_parser():  # type: () -> argparse.ArgumentParser
                         default=True, dest="enable_reuse",
                         help="")
 
                         default=True, dest="enable_reuse",
                         help="")
 

-    parser.add_argument("--project-uuid", type=str, help="Project that will own the workflow jobs, if not provided, will go to home project.")
+    parser.add_argument("--project-uuid", type=str, metavar="UUID", help="Project that will own the workflow jobs, if not provided, will go to home project.")

     parser.add_argument("--ignore-docker-for-reuse", action="store_true",
                         help="Ignore Docker image version when deciding whether to reuse past jobs.",
                         default=False)
     parser.add_argument("--ignore-docker-for-reuse", action="store_true",
                         help="Ignore Docker image version when deciding whether to reuse past jobs.",
                         default=False)


@@ -276,6 +343,8 @@ def arg_parser():  # type: () -> argparse.ArgumentParser
     exgroup.add_argument("--local", action="store_false", help="Run workflow on local host (submits jobs to Arvados).",
                         default=True, dest="submit")
     exgroup.add_argument("--create-template", action="store_true", help="Create an Arvados pipeline template.")
     exgroup.add_argument("--local", action="store_false", help="Run workflow on local host (submits jobs to Arvados).",
                         default=True, dest="submit")
     exgroup.add_argument("--create-template", action="store_true", help="Create an Arvados pipeline template.")

+    exgroup.add_argument("--create-workflow", action="store_true", help="Create an Arvados workflow.")
+    exgroup.add_argument("--update-workflow", type=str, metavar="UUID", help="Update existing Arvados workflow with uuid.")

 
     exgroup = parser.add_mutually_exclusive_group()
     exgroup.add_argument("--wait", action="store_true", help="After submitting workflow runner job, wait for completion.",
 
     exgroup = parser.add_mutually_exclusive_group()
     exgroup.add_argument("--wait", action="store_true", help="After submitting workflow runner job, wait for completion.",


@@ -287,20 +356,36 @@ def arg_parser():  # type: () -> argparse.ArgumentParser
                         default=None, dest="work_api",
                         help="Select work submission API, one of 'jobs' or 'containers'.")
 
                         default=None, dest="work_api",
                         help="Select work submission API, one of 'jobs' or 'containers'.")
 

+    parser.add_argument("--compute-checksum", action="store_true", default=False,
+                        help="Compute checksum of contents while collecting outputs",
+                        dest="compute_checksum")
+

     parser.add_argument("workflow", type=str, nargs="?", default=None, help="The workflow to execute")
     parser.add_argument("job_order", nargs=argparse.REMAINDER, help="The input object to the workflow.")
 
     return parser
 
     parser.add_argument("workflow", type=str, nargs="?", default=None, help="The workflow to execute")
     parser.add_argument("job_order", nargs=argparse.REMAINDER, help="The input object to the workflow.")
 
     return parser
 

+def add_arv_hints():
+    cache = {}
+    res = pkg_resources.resource_stream(__name__, 'arv-cwl-schema.yml')
+    cache["http://arvados.org/cwl"] = res.read()
+    res.close()
+    _, cwlnames, _, _ = cwltool.process.get_schema("v1.0")
+    _, extnames, _, _ = schema_salad.schema.load_schema("http://arvados.org/cwl", cache=cache)
+    for n in extnames.names:
+        if not cwlnames.has_name("http://arvados.org/cwl#"+n, ""):
+            cwlnames.add_name("http://arvados.org/cwl#"+n, "", extnames.get_name(n, ""))

 
 def main(args, stdout, stderr, api_client=None):
     parser = arg_parser()
 
     job_order_object = None
     arvargs = parser.parse_args(args)
 
 def main(args, stdout, stderr, api_client=None):
     parser = arg_parser()
 
     job_order_object = None
     arvargs = parser.parse_args(args)

-    if arvargs.create_template and not arvargs.job_order:
+    if (arvargs.create_template or arvargs.create_workflow or arvargs.update_workflow) and not arvargs.job_order:

         job_order_object = ({}, "")
 
         job_order_object = ({}, "")
 

+    add_arv_hints()
+

     try:
         if api_client is None:
             api_client=arvados.api('v1', model=OrderedJsonModel())
     try:
         if api_client is None:
             api_client=arvados.api('v1', model=OrderedJsonModel())


@@ -309,13 +394,21 @@ def main(args, stdout, stderr, api_client=None):
         logger.error(e)
         return 1
 
         logger.error(e)
         return 1
 

+    if arvargs.debug:
+        logger.setLevel(logging.DEBUG)
+
+    if arvargs.quiet:
+        logger.setLevel(logging.WARN)
+        logging.getLogger('arvados.arv-run').setLevel(logging.WARN)
+

     arvargs.conformance_test = None
     arvargs.use_container = True
 
     return cwltool.main.main(args=arvargs,
                              stdout=stdout,
                              stderr=stderr,
     arvargs.conformance_test = None
     arvargs.use_container = True
 
     return cwltool.main.main(args=arvargs,
                              stdout=stdout,
                              stderr=stderr,

-                             executor=runner.arvExecutor,
-                             makeTool=runner.arvMakeTool,
+                             executor=runner.arv_executor,
+                             makeTool=runner.arv_make_tool,

                              versionfunc=versionstring,
                              versionfunc=versionstring,

-                             job_order_object=job_order_object)
+                             job_order_object=job_order_object,
+                             make_fs_access=partial(CollectionFsAccess, api_client=api_client))