Merge branch '10649-squeue-stderr' closes #10649

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

diff --git a/sdk/cwl/arvados_cwl/arvjob.py b/sdk/cwl/arvados_cwl/arvjob.py
index c3e4cbe926c21a5c8964e761f9a8d8445b8cb39b..93f84f8fd173e1ffa3fd41b1632396e3148ab178 100644 (file)
--- a/sdk/cwl/arvados_cwl/arvjob.py
+++ b/sdk/cwl/arvados_cwl/arvjob.py
@@ -139,7 +139,7 @@ class ArvadosJob(object):
                 with Perf(metrics, "done %s" % self.name):
                     self.done(response)
         except Exception as e:
-            logger.error("Got error %s" % str(e))
+            logger.exception("Job %s error" % (self.name))
             self.output_callback({}, "permanentFail")
 
     def update_pipeline_component(self, record):
@@ -204,13 +204,18 @@ class ArvadosJob(object):
                         outputs = done.done(self, record, dirs["tmpdir"],
                                             dirs["outdir"], dirs["keep"])
             except WorkflowException as e:
-                logger.error("Error while collecting job outputs:\n%s", e, exc_info=(e if self.arvrunner.debug else False))
+                logger.error("Error while collecting output for job %s:\n%s", self.name, e, exc_info=(e if self.arvrunner.debug else False))
                 processStatus = "permanentFail"
-                outputs = None
             except Exception as e:
-                logger.exception("Got unknown exception while collecting job outputs:")
+                logger.exception("Got unknown exception while collecting output for job %s:", self.name)
+                processStatus = "permanentFail"
+
+            # Note: Currently, on error output_callback is expecting an empty dict,
+            # anything else will fail.
+            if not isinstance(outputs, dict):
+                logger.error("Unexpected output type %s '%s'", type(outputs), outputs)
+                outputs = {}
                 processStatus = "permanentFail"
-                outputs = None
 
             self.output_callback(outputs, processStatus)
         finally:
@@ -252,7 +257,8 @@ class RunnerJob(Runner):
             "repository": "arvados",
             "script_parameters": self.job_order,
             "runtime_constraints": {
-                "docker_image": arvados_jobs_image(self.arvrunner)
+                "docker_image": arvados_jobs_image(self.arvrunner),
+                "min_ram_mb_per_node": self.submit_runner_ram
             }
         }
 
@@ -303,7 +309,7 @@ class RunnerTemplate(object):
         'string': 'text',
     }
 
-    def __init__(self, runner, tool, job_order, enable_reuse, uuid):
+    def __init__(self, runner, tool, job_order, enable_reuse, uuid, submit_runner_ram=0):
         self.runner = runner
         self.tool = tool
         self.job = RunnerJob(
@@ -312,7 +318,8 @@ class RunnerTemplate(object):
             job_order=job_order,
             enable_reuse=enable_reuse,
             output_name=None,
-            output_tags=None)
+            output_tags=None,
+            submit_runner_ram=submit_runner_ram)
         self.uuid = uuid
 
     def pipeline_component_spec(self):