Factor out SA statement building logic into 2 separate helpers (I think
the filtering criteria for admin and nonadmin users is sufficiently
different, so 2 methods make for cleaner code here).
and invalidate `_output_hdas_and_paths` when current session is not the
same session that was used to populate `_output_hdas_and_paths`.
The Job instance may originate from a session that is associated to
another thread, and when that thread closes the session the instance
(along with other instances retrieved through loading relationships)
becomes detached.
I am not sure at all if this will fix
```
DetachedInstanceError: Instance <HistoryDatasetAssociation at 0x7fe68bbf14f0> is not bound to a Session; attribute refresh operation cannot proceed (Background on this error at: https://sqlalche.me/e/14/bhk3)
File "galaxy/jobs/runners/__init__.py", line 291, in prepare_job
job_wrapper.prepare()
File "galaxy/jobs/__init__.py", line 1248, in prepare
tool_evaluator.set_compute_environment(compute_environment, get_special=get_special)
File "galaxy/tools/evaluation.py", line 162, in set_compute_environment
self.param_dict = self.build_param_dict(
File "galaxy/tools/evaluation.py", line 204, in build_param_dict
self.__populate_output_dataset_wrappers(param_dict, output_datasets, job_working_directory)
File "galaxy/tools/evaluation.py", line 447, in __populate_output_dataset_wrappers
param_dict[name] = DatasetFilenameWrapper(
File "galaxy/tools/wrappers.py", line 403, in __init__
path_rewrite = compute_environment and compute_environment.output_path_rewrite(dataset_instance)
File "galaxy/job_execution/compute_environment.py", line 132, in output_path_rewrite
return str(self.job_io.get_output_path(dataset))
File "galaxy/job_execution/setup.py", line 226, in get_output_path
if hda.id == dataset.id:
File "sqlalchemy/orm/attributes.py", line 487, in __get__
return self.impl.get(state, dict_)
File "sqlalchemy/orm/attributes.py", line 959, in get
value = self._fire_loader_callables(state, key, passive)
File "sqlalchemy/orm/attributes.py", line 990, in _fire_loader_callables
return state._load_expired(state, passive)
File "sqlalchemy/orm/state.py", line 712, in _load_expired
self.manager.expired_attribute_loader(self, toload, passive)
File "sqlalchemy/orm/loading.py", line 1369, in load_scalar_attributes
raise orm_exc.DetachedInstanceError(
```
but it seems to make some sense. JobIO crosses thread boundaries as part
of the job wrapper getting put into threading queues.
Ideally we'd make sure that no ORM instance crosses the thread boundary
(or we systematically re-associated with a session).
I also tried flagging these patterns automatically using something like:
```
@event.listens_for(session, "persistent_to_detached")
def on_detach(sess, instance):
if not getattr(instance, "allow_detatch", False):
raise Exception(f"{instance} detached. This ain't good for how we do things ?")
```
but it seems tricky to figure out when this is fine and when it is not.
So https://github.com/galaxyproject/galaxy/pull/10121 added back in the title header to the component which is only designed for the dataset details page for an individual dataset - the original implementation striped out the title for that reason.
The target might an interesting step of a workflow, a problematic tool, a complex resulting dataset - so trying to come up with a title automically is a little challenging. My thought was therefore to keep the component rendered as simply as possible - since the page or report writer will be the one that has the context for how the component should be titled. It needs a prefix and so the big h2 "Job Metrics" is not useful or wanted I think.
https://github.com/galaxyproject/galaxy/pull/10121 did correctly preserve the removal of the title for the job parameters component - so I think this getting added back in was just an oversight.
John Davis