19385: Clean up debug prints
[arvados.git] / sdk / cwl / arvados_cwl / arvcontainer.py
index 9a0c53d86c44130a7e0528d07590c2dfdc3b4fea..742906c616514b9bcd2317cfeb4a11697f1c8522 100644 (file)
@@ -2,40 +2,48 @@
 #
 # SPDX-License-Identifier: Apache-2.0
 
+from future import standard_library
+standard_library.install_aliases()
+from builtins import str
+
 import logging
 import json
 import os
-import urllib
+import urllib.request, urllib.parse, urllib.error
 import time
 import datetime
 import ciso8601
 import uuid
+import math
 
-from arvados_cwl.util import get_current_container, get_intermediate_collection_info
-import ruamel.yaml as yaml
+import arvados_cwl.util
+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."""
 
-    def __init__(self, runner,
+    def __init__(self, runner, job_runtime,
                  builder,   # type: Builder
                  joborder,  # type: Dict[Text, Union[Dict[Text, Any], List, Text]]
                  make_path_mapper,  # type: Callable[..., PathMapper]
@@ -45,13 +53,20 @@ class ArvadosContainer(JobBase):
     ):
         super(ArvadosContainer, self).__init__(builder, joborder, make_path_mapper, requirements, hints, name)
         self.arvrunner = runner
+        self.job_runtime = job_runtime
         self.running = False
         self.uuid = None
 
     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
@@ -59,19 +74,29 @@ class ArvadosContainer(JobBase):
         # ArvadosContainer object by CommandLineTool.job() before
         # run() is called.
 
-        container_request = {
-            "command": self.command_line,
-            "name": self.name,
-            "output_path": self.outdir,
-            "cwd": self.outdir,
-            "priority": runtimeContext.priority,
-            "state": "Committed",
-            "properties": {},
-        }
+        runtimeContext = self.job_runtime
+
+        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 self.arvrunner.project_uuid:
-            container_request["owner_uuid"] = self.arvrunner.project_uuid
+        if runtimeContext.project_uuid:
+            container_request["owner_uuid"] = runtimeContext.project_uuid
 
         if self.arvrunner.secret_store.has_secret(self.command_line):
             raise WorkflowException("Secret material leaked on command line, only file literals may contain secrets")
@@ -81,17 +106,17 @@ class ArvadosContainer(JobBase):
 
         resources = self.builder.resources
         if resources is not None:
-            runtime_constraints["vcpus"] = resources.get("cores", 1)
-            runtime_constraints["ram"] = resources.get("ram") * 2**20
+            runtime_constraints["vcpus"] = math.ceil(resources.get("cores", 1))
+            runtime_constraints["ram"] = math.ceil(resources.get("ram") * 2**20)
 
         mounts = {
             self.outdir: {
                 "kind": "tmp",
-                "capacity": resources.get("outdirSize", 0) * 2**20
+                "capacity": math.ceil(resources.get("outdirSize", 0) * 2**20)
             },
             self.tmpdir: {
                 "kind": "tmp",
-                "capacity": resources.get("tmpdirSize", 0) * 2**20
+                "capacity": math.ceil(resources.get("tmpdirSize", 0) * 2**20)
             }
         }
         secret_mounts = {}
@@ -115,6 +140,8 @@ class ArvadosContainer(JobBase):
                 "kind": "collection",
                 "portable_data_hash": pdh
             }
+            if pdh in self.pathmapper.pdh_to_uuid:
+                mounts[targetdir]["uuid"] = self.pathmapper.pdh_to_uuid[pdh]
             if len(sp) == 2:
                 if tp == "Directory":
                     path = sp[1]
@@ -124,37 +151,46 @@ 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,
                                                     keep_client=self.arvrunner.keep_client,
                                                     num_retries=self.arvrunner.num_retries)
-                generatemapper = NoFollowPathMapper([self.generatefiles], "", "",
+                generatemapper = NoFollowPathMapper(self.generatefiles["listing"], "", "",
                                                     separateDirs=False)
 
-                sorteditems = sorted(generatemapper.items(), None, key=lambda n: n[1].target)
+                sorteditems = sorted(generatemapper.items(), key=lambda n: n[1].target)
 
                 logger.debug("generatemapper is %s", sorteditems)
 
                 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:
-                                    n.write(p.resolved.encode("utf-8"))
+                                with vwd.open(dst, "w") as n:
+                                    n.write(p.resolved)
 
                 def keepemptydirs(p):
                     if isinstance(p, arvados.collection.RichCollectionBase):
@@ -167,23 +203,26 @@ class ArvadosContainer(JobBase):
                 keepemptydirs(vwd)
 
                 if not runtimeContext.current_container:
-                    runtimeContext.current_container = get_current_container(self.arvrunner.api, self.arvrunner.num_retries, logger)
-                info = get_intermediate_collection_info(self.name, runtimeContext.current_container, runtimeContext.intermediate_output_ttl)
-                vwd.save_new(name=info["name"],
-                             owner_uuid=self.arvrunner.project_uuid,
+                    runtimeContext.current_container = arvados_cwl.util.get_current_container(self.arvrunner.api, self.arvrunner.num_retries, logger)
+                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 + "/"
@@ -207,22 +246,36 @@ 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,
-                                                                     self.arvrunner.project_uuid)
+                                                                    docker_req,
+                                                                    runtimeContext.pull_image,
+                                                                    runtimeContext)
+
+        network_req, _ = self.get_requirement("NetworkAccess")
+        if network_req:
+            runtime_constraints["API"] = network_req["networkAccess"]
 
         api_req, _ = self.get_requirement("http://arvados.org/cwl#APIRequirement")
         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"] = 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.
@@ -246,10 +299,41 @@ 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.timelimit is not None:
+
+        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
 
-        container_request["output_name"] = "Output for step %s" % (self.name)
+        extra_submit_params = {}
+        if runtimeContext.submit_runner_cluster:
+            extra_submit_params["cluster_id"] = runtimeContext.submit_runner_cluster
+
+        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
@@ -258,11 +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)
@@ -276,11 +378,13 @@ class ArvadosContainer(JobBase):
             if runtimeContext.submit_request_uuid:
                 response = self.arvrunner.api.container_requests().update(
                     uuid=runtimeContext.submit_request_uuid,
-                    body=container_request
+                    body=container_request,
+                    **extra_submit_params
                 ).execute(num_retries=self.arvrunner.num_retries)
             else:
                 response = self.arvrunner.api.container_requests().create(
-                    body=container_request
+                    body=container_request,
+                    **extra_submit_params
                 ).execute(num_retries=self.arvrunner.num_retries)
 
             self.uuid = response["uuid"]
@@ -291,7 +395,8 @@ class ArvadosContainer(JobBase):
             else:
                 logger.info("%s %s state is %s", self.arvrunner.label(self), response["uuid"], response["state"])
         except Exception as e:
-            logger.error("%s got error %s" % (self.arvrunner.label(self), str(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):
@@ -312,11 +417,15 @@ 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"
 
-            if processStatus == "permanentFail":
-                logc = arvados.collection.CollectionReader(container["log"],
+            if processStatus == "permanentFail" and record["log_uuid"]:
+                logc = arvados.collection.CollectionReader(record["log_uuid"],
                                                            api_client=self.arvrunner.api,
                                                            keep_client=self.arvrunner.keep_client,
                                                            num_retries=self.arvrunner.num_retries)
@@ -328,7 +437,7 @@ class ArvadosContainer(JobBase):
             if record["output_uuid"]:
                 if self.arvrunner.trash_intermediate or self.arvrunner.intermediate_output_ttl:
                     # Compute the trash time to avoid requesting the collection record.
-                    trash_at = ciso8601.parse_datetime_unaware(record["modified_at"]) + datetime.timedelta(0, self.arvrunner.intermediate_output_ttl)
+                    trash_at = ciso8601.parse_datetime_as_naive(record["modified_at"]) + datetime.timedelta(0, self.arvrunner.intermediate_output_ttl)
                     aftertime = " at %s" % trash_at.strftime("%Y-%m-%d %H:%M:%S UTC") if self.arvrunner.intermediate_output_ttl else ""
                     orpart = ", or" if self.arvrunner.trash_intermediate and self.arvrunner.intermediate_output_ttl else ""
                     oncomplete = " upon successful completion of the workflow" if self.arvrunner.trash_intermediate else ""
@@ -338,12 +447,21 @@ 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.
             logger.error("%s unable to collect output from %s:\n%s",
                          self.arvrunner.label(self), container["output"], e, exc_info=(e if self.arvrunner.debug else False))
             processStatus = "permanentFail"
-        except Exception as e:
-            logger.exception("%s while getting output object: %s", self.arvrunner.label(self), e)
+        except Exception:
+            logger.exception("%s while getting output object:", self.arvrunner.label(self))
             processStatus = "permanentFail"
         finally:
             self.output_callback(outputs, processStatus)
@@ -352,7 +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
@@ -379,7 +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",
@@ -396,16 +514,16 @@ class RunnerContainer(Runner):
             },
             "secret_mounts": secret_mounts,
             "runtime_constraints": {
-                "vcpus": self.submit_runner_cores,
-                "ram": 1024*1024 * self.submit_runner_ram,
+                "vcpus": math.ceil(self.submit_runner_cores),
+                "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": {}
         }
 
-        if self.tool.tool.get("id", "").startswith("keep:"):
-            sp = self.tool.tool["id"].split('/')
+        if self.embedded_tool.tool.get("id", "").startswith("keep:"):
+            sp = self.embedded_tool.tool["id"].split('/')
             workflowcollection = sp[0][5:]
             workflowname = "/".join(sp[1:])
             workflowpath = "/var/lib/cwl/workflow/%s" % workflowname
@@ -413,16 +531,34 @@ 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.tool, self.merged_map)
+            main = self.loadingContext.loader.idx["_:main"]
+            if main.get("id") == "_:main":
+                del main["id"]
             workflowpath = "/var/lib/cwl/workflow.json#main"
             container_req["mounts"]["/var/lib/cwl/workflow.json"] = {
                 "kind": "json",
-                "content": packed
+                "content": main
             }
-            if self.tool.tool.get("id", "").startswith("arvwf:"):
-                container_req["properties"]["template_uuid"] = self.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
@@ -431,14 +567,17 @@ class RunnerContainer(Runner):
         # --eval-timeout is the timeout for javascript invocation
         # --parallel-task-count is the number of threads to use for job submission
         # --enable/disable-reuse sets desired job reuse
+        # --collection-cache-size sets aside memory to store collections
         command = ["arvados-cwl-runner",
                    "--local",
                    "--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"]
+                   "--enable-reuse" if self.enable_reuse else "--disable-reuse",
+                   "--collection-cache-size=%s" % self.collection_cache_size]
 
         if self.output_name:
             command.append("--output-name=" + self.output_name)
@@ -450,20 +589,41 @@ 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"])
 
@@ -474,18 +634,27 @@ 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:
+            extra_submit_params["cluster_id"] = runtimeContext.submit_runner_cluster
 
         if runtimeContext.submit_request_uuid:
+            if "cluster_id" in extra_submit_params:
+                # Doesn't make sense for "update" and actually fails
+                del extra_submit_params["cluster_id"]
             response = self.arvrunner.api.container_requests().update(
                 uuid=runtimeContext.submit_request_uuid,
-                body=job_spec
+                body=job_spec,
+                **extra_submit_params
             ).execute(num_retries=self.arvrunner.num_retries)
         else:
             response = self.arvrunner.api.container_requests().create(
-                body=job_spec
+                body=job_spec,
+                **extra_submit_params
             ).execute(num_retries=self.arvrunner.num_retries)
 
         self.uuid = response["uuid"]
@@ -493,13 +662,25 @@ 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(
                 uuid=record["container_uuid"]
             ).execute(num_retries=self.arvrunner.num_retries)
-        except Exception as e:
-            logger.exception("%s while getting runner container: %s", self.arvrunner.label(self), e)
+            container["log"] = record["log_uuid"]
+        except Exception:
+            logger.exception("%s while getting runner container", self.arvrunner.label(self))
             self.arvrunner.output_callback({}, "permanentFail")
         else:
             super(RunnerContainer, self).done(container)