Merge pull request #18594 from mvdbeek/include_invocation_in_exception

[24.1] Include workflow invocation id in exception logs

lib/galaxy/schema/invocation.py

@@ -78,6 +78,21 @@ class FailureReason(str, Enum):
     unexpected_failure = "unexpected_failure"
 
 
+# The reasons below are attached to the invocation and user-actionable.
+# Not included are `unexpected_failure` and `expression_evaluation_failed`.
+# If expression evaluation fails we're not attaching the templated
+# expression to the invocation, as it could contain secrets.
+# If the failure reason is not in `FAILURE_REASONS_EXPECTED` we should
+# log an exception so admins can debug and/or submit bug reports.
+FAILURE_REASONS_EXPECTED = (
+    FailureReason.dataset_failed,
+    FailureReason.collection_failed,
+    FailureReason.job_failed,
+    FailureReason.output_not_found,
+    FailureReason.when_not_boolean,
+)
+
+
 class CancelReason(str, Enum):
     """Possible reasons for a cancelled workflow."""
 

lib/galaxy/workflow/run.py

@@ -25,6 +25,7 @@
 )
 from galaxy.schema.invocation import (
     CancelReason,
+    FAILURE_REASONS_EXPECTED,
     FailureReason,
     InvocationCancellationHistoryDeleted,
     InvocationFailureCollectionFailed,
@@ -251,8 +252,12 @@ def invoke(self) -> Dict[int, Any]:
                 step_delayed = delayed_steps = True
                 self.progress.mark_step_outputs_delayed(step, why=de.why)
             except Exception as e:
-                log.exception(
-                    "Failed to schedule %s, problem occurred on %s.",
+                log_function = log.exception
+                if isinstance(e, modules.FailWorkflowEvaluation) and e.why.reason in FAILURE_REASONS_EXPECTED:
+                    log_function = log.info
+                log_function(
+                    "Failed to schedule %s for %s, problem occurred on %s.",
+                    self.workflow_invocation.log_str(),
                     self.workflow_invocation.workflow.log_str(),
                     step.log_str(),
                 )