def __init__(self, dataset, datatypes_registry=None, tool=None, name=None, dataset_path=None, identifier=None): if not dataset: try: # TODO: allow this to work when working with grouping ext = tool.inputs[name].extensions[0] except: ext = 'data' self.dataset = wrap_with_safe_string( NoneDataset(datatypes_registry=datatypes_registry, ext=ext), no_wrap_classes=ToolParameterValueWrapper) else: # Tool wrappers should not normally be accessing .dataset directly, # so we will wrap it and keep the original around for file paths # Should we name this .value to maintain consistency with most other ToolParameterValueWrapper? self.unsanitized = dataset self.dataset = wrap_with_safe_string( dataset, no_wrap_classes=ToolParameterValueWrapper) self.metadata = self.MetadataWrapper(dataset.metadata) self.datatypes_registry = datatypes_registry self.false_path = getattr(dataset_path, "false_path", None) self.false_extra_files_path = getattr(dataset_path, "false_extra_files_path", None) self._element_identifier = identifier
def __populate_output_dataset_wrappers(self, param_dict, output_datasets, output_paths, job_working_directory): output_dataset_paths = dataset_path_rewrites( output_paths ) for name, hda in output_datasets.items(): # Write outputs to the working directory (for security purposes) # if desired. real_path = hda.file_name if real_path in output_dataset_paths: dataset_path = output_dataset_paths[ real_path ] param_dict[name] = DatasetFilenameWrapper( hda, dataset_path=dataset_path ) try: open( dataset_path.false_path, 'w' ).close() except EnvironmentError: pass # May well not exist - e.g. Pulsar. else: param_dict[name] = DatasetFilenameWrapper( hda ) # Provide access to a path to store additional files # TODO: path munging for cluster/dataset server relocatability param_dict[name].files_path = os.path.abspath(os.path.join( job_working_directory, "dataset_%s_files" % (hda.dataset.id) )) for child in hda.children: param_dict[ "_CHILD___%s___%s" % ( name, child.designation ) ] = DatasetFilenameWrapper( child ) for out_name, output in self.tool.outputs.iteritems(): if out_name not in param_dict and output.filters: # Assume the reason we lack this output is because a filter # failed to pass; for tool writing convienence, provide a # NoneDataset ext = getattr( output, "format", None ) # populate only for output datasets (not collections) param_dict[ out_name ] = NoneDataset( datatypes_registry=self.app.datatypes_registry, ext=ext )
def __init__(self, dataset, datatypes_registry=None, tool=None, name=None, dataset_path=None, identifier=None, formats=None): if not dataset: try: # TODO: allow this to work when working with grouping ext = tool.inputs[name].extensions[0] except Exception: ext = 'data' self.dataset = wrap_with_safe_string( NoneDataset(datatypes_registry=datatypes_registry, ext=ext), no_wrap_classes=ToolParameterValueWrapper) else: # Tool wrappers should not normally be accessing .dataset directly, # so we will wrap it and keep the original around for file paths # Should we name this .value to maintain consistency with most other ToolParameterValueWrapper? if formats: target_ext, converted_dataset = dataset.find_conversion_destination( formats) if target_ext and converted_dataset: dataset = converted_dataset self.unsanitized = dataset self.dataset = wrap_with_safe_string( dataset, no_wrap_classes=ToolParameterValueWrapper) self.metadata = self.MetadataWrapper(dataset.metadata) if hasattr(dataset, 'tags'): self.groups = { tag.user_value.lower() for tag in dataset.tags if tag.user_tname == 'group' } else: # May be a 'FakeDatasetAssociation' self.groups = set() self.datatypes_registry = datatypes_registry self.false_path = getattr(dataset_path, "false_path", None) self.false_extra_files_path = getattr(dataset_path, "false_extra_files_path", None) self._element_identifier = identifier
def __init__(self, dataset, datatypes_registry=None, tool=None, name=None, dataset_path=None): if not dataset: try: # TODO: allow this to work when working with grouping ext = tool.inputs[name].extensions[0] except: ext = 'data' self.dataset = NoneDataset(datatypes_registry=datatypes_registry, ext=ext) else: self.dataset = dataset self.metadata = self.MetadataWrapper(dataset.metadata) self.false_path = getattr(dataset_path, "false_path", None) self.false_extra_files_path = getattr(dataset_path, "false_extra_files_path", None)
def execute(self, tool, trans, incoming={}, return_job=False, set_output_hid=True, set_output_history=True, history=None, job_params=None, rerun_remap_job_id=None, mapping_over_collection=False, execution_cache=None): """ Executes a tool, creating job and tool outputs, associating them, and submitting the job to the job queue. If history is not specified, use trans.history as destination for tool's output datasets. """ self._check_access(tool, trans) app = trans.app if execution_cache is None: execution_cache = ToolExecutionCache(trans) current_user_roles = execution_cache.current_user_roles history, inp_data, inp_dataset_collections = self._collect_inputs( tool, trans, incoming, history, current_user_roles) out_data = odict() out_collections = {} out_collection_instances = {} # Deal with input dataset names, 'dbkey' and types input_names = [] # format='input" previously would give you a random extension from # the input extensions, now it should just give "input" as the output # format. input_ext = 'data' if tool.profile < 16.04 else "input" input_dbkey = incoming.get("dbkey", "?") for name, data in reversed(inp_data.items()): if not data: data = NoneDataset(datatypes_registry=app.datatypes_registry) continue # Convert LDDA to an HDA. if isinstance(data, LibraryDatasetDatasetAssociation): data = data.to_history_dataset_association(None) inp_data[name] = data else: # HDA if data.hid: input_names.append('data %s' % data.hid) if tool.profile < 16.04: input_ext = data.ext if data.dbkey not in [None, '?']: input_dbkey = data.dbkey identifier = getattr(data, "element_identifier", None) if identifier is not None: incoming["%s|__identifier__" % name] = identifier # Collect chromInfo dataset and add as parameters to incoming (chrom_info, db_dataset) = app.genome_builds.get_chrom_info( input_dbkey, trans=trans, custom_build_hack_get_len_from_fasta_conversion=tool.id != 'CONVERTER_fasta_to_len') if db_dataset: inp_data.update({"chromInfo": db_dataset}) incoming["chromInfo"] = chrom_info # Determine output dataset permission/roles list existing_datasets = [inp for inp in inp_data.values() if inp] if existing_datasets: output_permissions = app.security_agent.guess_derived_permissions_for_datasets( existing_datasets) else: # No valid inputs, we will use history defaults output_permissions = app.security_agent.history_get_default_permissions( history) # Build name for output datasets based on tool name and input names on_text = on_text_for_names(input_names) # Add the dbkey to the incoming parameters incoming["dbkey"] = input_dbkey # wrapped params are used by change_format action and by output.label; only perform this wrapping once, as needed wrapped_params = WrappedParameters(trans, tool, incoming) # Keep track of parent / child relationships, we'll create all the # datasets first, then create the associations parent_to_child_pairs = [] child_dataset_names = set() object_store_populator = ObjectStorePopulator(app) def handle_output(name, output, hidden=None): if output.parent: parent_to_child_pairs.append((output.parent, name)) child_dataset_names.add(name) # What is the following hack for? Need to document under what # conditions can the following occur? ([email protected]) # HACK: the output data has already been created # this happens i.e. as a result of the async controller if name in incoming: dataid = incoming[name] data = trans.sa_session.query( app.model.HistoryDatasetAssociation).get(dataid) assert data is not None out_data[name] = data else: ext = determine_output_format(output, wrapped_params.params, inp_data, inp_dataset_collections, input_ext) data = app.model.HistoryDatasetAssociation(extension=ext, create_dataset=True, flush=False) if hidden is None: hidden = output.hidden if hidden: data.visible = False trans.sa_session.add(data) trans.app.security_agent.set_all_dataset_permissions( data.dataset, output_permissions, new=True) # Must flush before setting object store id currently. # TODO: optimize this. trans.sa_session.flush() object_store_populator.set_object_store_id(data) # This may not be neccesary with the new parent/child associations data.designation = name # Copy metadata from one of the inputs if requested. # metadata source can be either a string referencing an input # or an actual object to copy. metadata_source = output.metadata_source if metadata_source: if isinstance(metadata_source, string_types): metadata_source = inp_data.get(metadata_source) if metadata_source is not None: data.init_meta(copy_from=metadata_source) else: data.init_meta() # Take dbkey from LAST input data.dbkey = str(input_dbkey) # Set state data.blurb = "queued" # Set output label data.name = self.get_output_name(output, data, tool, on_text, trans, incoming, history, wrapped_params.params, job_params) # Store output out_data[name] = data if output.actions: # Apply pre-job tool-output-dataset actions; e.g. setting metadata, changing format output_action_params = dict(out_data) output_action_params.update(incoming) output.actions.apply_action(data, output_action_params) # Also set the default values of actions of type metadata self.set_metadata_defaults(output, data, tool, on_text, trans, incoming, history, wrapped_params.params, job_params) # Flush all datasets at once. return data for name, output in tool.outputs.items(): if not filter_output(output, incoming): if output.collection: collections_manager = app.dataset_collections_service # As far as I can tell - this is always true - but just verify assert set_output_history, "Cannot create dataset collection for this kind of tool." element_identifiers = [] input_collections = dict([ (k, v[0][0]) for k, v in inp_dataset_collections.iteritems() ]) known_outputs = output.known_outputs( input_collections, collections_manager.type_registry) # Just to echo TODO elsewhere - this should be restructured to allow # nested collections. for output_part_def in known_outputs: # Add elements to top-level collection, unless nested... current_element_identifiers = element_identifiers current_collection_type = output.structure.collection_type for parent_id in (output_part_def.parent_ids or []): # TODO: replace following line with formal abstractions for doing this. current_collection_type = ":".join( current_collection_type.split(":")[1:]) name_to_index = dict( map( lambda (index, value): (value["name"], index), enumerate(current_element_identifiers))) if parent_id not in name_to_index: if parent_id not in current_element_identifiers: index = len(current_element_identifiers) current_element_identifiers.append( dict( name=parent_id, collection_type= current_collection_type, src="new_collection", element_identifiers=[], )) else: index = name_to_index[parent_id] current_element_identifiers = current_element_identifiers[ index]["element_identifiers"] effective_output_name = output_part_def.effective_output_name element = handle_output(effective_output_name, output_part_def.output_def, hidden=True) # TODO: this shouldn't exist in the top-level of the history at all # but for now we are still working around that by hiding the contents # there. # Following hack causes dataset to no be added to history... child_dataset_names.add(effective_output_name) if set_output_history: history.add_dataset(element, set_hid=set_output_hid, quota=False) trans.sa_session.add(element) trans.sa_session.flush() current_element_identifiers.append({ "__object__": element, "name": output_part_def.element_identifier, }) log.info(element_identifiers) if output.dynamic_structure: assert not element_identifiers # known_outputs must have been empty element_kwds = dict(elements=collections_manager. ELEMENTS_UNINITIALIZED) else: element_kwds = dict( element_identifiers=element_identifiers) collection_type = output.structure.collection_type if collection_type is None: collection_type_source = output.structure.collection_type_source if collection_type_source is None: # TODO: Not a new problem, but this should be determined # sooner. raise Exception( "Could not determine collection type to create." ) if collection_type_source not in input_collections: raise Exception( "Could not find collection type source with name [%s]." % collection_type_source) collection_type = input_collections[ collection_type_source].collection.collection_type if mapping_over_collection: dc = collections_manager.create_dataset_collection( trans, collection_type=collection_type, **element_kwds) out_collections[name] = dc else: hdca_name = self.get_output_name( output, None, tool, on_text, trans, incoming, history, wrapped_params.params, job_params) hdca = collections_manager.create( trans, history, name=hdca_name, collection_type=collection_type, trusted_identifiers=True, **element_kwds) # name here is name of the output element - not name # of the hdca. out_collection_instances[name] = hdca else: handle_output_timer = ExecutionTimer() handle_output(name, output) log.info("Handled output named %s for tool %s %s" % (name, tool.id, handle_output_timer)) add_datasets_timer = ExecutionTimer() # Add all the top-level (non-child) datasets to the history unless otherwise specified datasets_to_persist = [] for name in out_data.keys(): if name not in child_dataset_names and name not in incoming: # don't add children; or already existing datasets, i.e. async created data = out_data[name] datasets_to_persist.append(data) if set_output_history: # Set HID and add to history. # This is brand new and certainly empty so don't worry about quota. # TOOL OPTIMIZATION NOTE - from above loop to the job create below 99%+ # of execution time happens within in history.add_datasets. history.add_datasets(trans.sa_session, datasets_to_persist, set_hid=set_output_hid, quota=False, flush=False) else: for data in datasets_to_persist: trans.sa_session.add(data) # Add all the children to their parents for parent_name, child_name in parent_to_child_pairs: parent_dataset = out_data[parent_name] child_dataset = out_data[child_name] parent_dataset.children.append(child_dataset) log.info("Added output datasets to history %s" % add_datasets_timer) job_setup_timer = ExecutionTimer() # Create the job object job, galaxy_session = self._new_job_for_session(trans, tool, history) self._record_inputs(trans, tool, job, incoming, inp_data, inp_dataset_collections, current_user_roles) self._record_outputs(job, out_data, out_collections, out_collection_instances) job.object_store_id = object_store_populator.object_store_id if job_params: job.params = dumps(job_params) job.set_handler(tool.get_job_handler(job_params)) trans.sa_session.add(job) # Now that we have a job id, we can remap any outputs if this is a rerun and the user chose to continue dependent jobs # This functionality requires tracking jobs in the database. if app.config.track_jobs_in_database and rerun_remap_job_id is not None: try: old_job = trans.sa_session.query( app.model.Job).get(rerun_remap_job_id) assert old_job is not None, '(%s/%s): Old job id is invalid' % ( rerun_remap_job_id, job.id) assert old_job.tool_id == job.tool_id, '(%s/%s): Old tool id (%s) does not match rerun tool id (%s)' % ( old_job.id, job.id, old_job.tool_id, job.tool_id) if trans.user is not None: assert old_job.user_id == trans.user.id, '(%s/%s): Old user id (%s) does not match rerun user id (%s)' % ( old_job.id, job.id, old_job.user_id, trans.user.id) elif trans.user is None and type( galaxy_session) == trans.model.GalaxySession: assert old_job.session_id == galaxy_session.id, '(%s/%s): Old session id (%s) does not match rerun session id (%s)' % ( old_job.id, job.id, old_job.session_id, galaxy_session.id) else: raise Exception( '(%s/%s): Remapping via the API is not (yet) supported' % (old_job.id, job.id)) # Duplicate PJAs before remap. for pjaa in old_job.post_job_actions: job.add_post_job_action(pjaa.post_job_action) for jtod in old_job.output_datasets: for (job_to_remap, jtid) in [(jtid.job, jtid) for jtid in jtod.dataset.dependent_jobs]: if (trans.user is not None and job_to_remap.user_id == trans.user.id) or (trans.user is None and job_to_remap.session_id == galaxy_session.id): if job_to_remap.state == job_to_remap.states.PAUSED: job_to_remap.state = job_to_remap.states.NEW for hda in [ dep_jtod.dataset for dep_jtod in job_to_remap.output_datasets ]: if hda.state == hda.states.PAUSED: hda.state = hda.states.NEW hda.info = None input_values = dict([ (p.name, json.loads(p.value)) for p in job_to_remap.parameters ]) update_param(jtid.name, input_values, str(out_data[jtod.name].id)) for p in job_to_remap.parameters: p.value = json.dumps(input_values[p.name]) jtid.dataset = out_data[jtod.name] jtid.dataset.hid = jtod.dataset.hid log.info( 'Job %s input HDA %s remapped to new HDA %s' % (job_to_remap.id, jtod.dataset.id, jtid.dataset.id)) trans.sa_session.add(job_to_remap) trans.sa_session.add(jtid) jtod.dataset.visible = False trans.sa_session.add(jtod) except Exception: log.exception('Cannot remap rerun dependencies.') log.info("Setup for job %s complete, ready to flush %s" % (job.log_str(), job_setup_timer)) job_flush_timer = ExecutionTimer() trans.sa_session.flush() log.info("Flushed transaction for job %s %s" % (job.log_str(), job_flush_timer)) # Some tools are not really executable, but jobs are still created for them ( for record keeping ). # Examples include tools that redirect to other applications ( epigraph ). These special tools must # include something that can be retrieved from the params ( e.g., REDIRECT_URL ) to keep the job # from being queued. if 'REDIRECT_URL' in incoming: # Get the dataset - there should only be 1 for name in inp_data.keys(): dataset = inp_data[name] redirect_url = tool.parse_redirect_url(dataset, incoming) # GALAXY_URL should be include in the tool params to enable the external application # to send back to the current Galaxy instance GALAXY_URL = incoming.get('GALAXY_URL', None) assert GALAXY_URL is not None, "GALAXY_URL parameter missing in tool config." redirect_url += "&GALAXY_URL=%s" % GALAXY_URL # Job should not be queued, so set state to ok job.set_state(app.model.Job.states.OK) job.info = "Redirected to: %s" % redirect_url trans.sa_session.add(job) trans.sa_session.flush() trans.response.send_redirect( url_for(controller='tool_runner', action='redirect', redirect_url=redirect_url)) else: # Put the job in the queue if tracking in memory app.job_queue.put(job.id, job.tool_id) trans.log_event("Added job to the job queue, id: %s" % str(job.id), tool_id=job.tool_id) return job, out_data
def execute(self, tool, trans, incoming={}, return_job=False, set_output_hid=True, set_output_history=True, history=None, job_params=None, rerun_remap_job_id=None, mapping_over_collection=False): """ Executes a tool, creating job and tool outputs, associating them, and submitting the job to the job queue. If history is not specified, use trans.history as destination for tool's output datasets. """ assert tool.allow_user_access( trans.user), "User (%s) is not allowed to access this tool." % ( trans.user) # Set history. if not history: history = tool.get_default_history_by_trans(trans, create=True) out_data = odict() out_collections = {} out_collection_instances = {} # Track input dataset collections - but replace with simply lists so collect # input datasets can process these normally. inp_dataset_collections = self.collect_input_dataset_collections( tool, incoming) # Collect any input datasets from the incoming parameters inp_data = self.collect_input_datasets(tool, incoming, trans) # Deal with input dataset names, 'dbkey' and types input_names = [] input_ext = 'data' input_dbkey = incoming.get("dbkey", "?") inp_items = inp_data.items() inp_items.reverse() for name, data in inp_items: if not data: data = NoneDataset( datatypes_registry=trans.app.datatypes_registry) continue # Convert LDDA to an HDA. if isinstance(data, LibraryDatasetDatasetAssociation): data = data.to_history_dataset_association(None) inp_data[name] = data else: # HDA if data.hid: input_names.append('data %s' % data.hid) input_ext = data.ext if data.dbkey not in [None, '?']: input_dbkey = data.dbkey identifier = getattr(data, "element_identifier", None) if identifier is not None: incoming["%s|__identifier__" % name] = identifier # Collect chromInfo dataset and add as parameters to incoming (chrom_info, db_dataset) = trans.app.genome_builds.get_chrom_info( input_dbkey, trans=trans, custom_build_hack_get_len_from_fasta_conversion=tool.id != 'CONVERTER_fasta_to_len') if db_dataset: inp_data.update({"chromInfo": db_dataset}) incoming["chromInfo"] = chrom_info # Determine output dataset permission/roles list existing_datasets = [inp for inp in inp_data.values() if inp] if existing_datasets: output_permissions = trans.app.security_agent.guess_derived_permissions_for_datasets( existing_datasets) else: # No valid inputs, we will use history defaults output_permissions = trans.app.security_agent.history_get_default_permissions( history) # Build name for output datasets based on tool name and input names on_text = on_text_for_names(input_names) # Add the dbkey to the incoming parameters incoming["dbkey"] = input_dbkey # wrapped params are used by change_format action and by output.label; only perform this wrapping once, as needed wrapped_params = WrappedParameters(trans, tool, incoming) # Keep track of parent / child relationships, we'll create all the # datasets first, then create the associations parent_to_child_pairs = [] child_dataset_names = set() object_store_populator = ObjectStorePopulator(trans.app) def handle_output(name, output): if output.parent: parent_to_child_pairs.append((output.parent, name)) child_dataset_names.add(name) ## What is the following hack for? Need to document under what ## conditions can the following occur? ([email protected]) # HACK: the output data has already been created # this happens i.e. as a result of the async controller if name in incoming: dataid = incoming[name] data = trans.sa_session.query( trans.app.model.HistoryDatasetAssociation).get(dataid) assert data is not None out_data[name] = data else: ext = determine_output_format(output, wrapped_params.params, inp_data, input_ext) data = trans.app.model.HistoryDatasetAssociation( extension=ext, create_dataset=True, sa_session=trans.sa_session) if output.hidden: data.visible = False # Commit the dataset immediately so it gets database assigned unique id trans.sa_session.add(data) trans.sa_session.flush() trans.app.security_agent.set_all_dataset_permissions( data.dataset, output_permissions) object_store_populator.set_object_store_id(data) # This may not be neccesary with the new parent/child associations data.designation = name # Copy metadata from one of the inputs if requested. # metadata source can be either a string referencing an input # or an actual object to copy. metadata_source = output.metadata_source if metadata_source: if isinstance(metadata_source, basestring): metadata_source = inp_data[metadata_source] if metadata_source is not None: data.init_meta(copy_from=metadata_source) else: data.init_meta() # Take dbkey from LAST input data.dbkey = str(input_dbkey) # Set state # FIXME: shouldn't this be NEW until the job runner changes it? data.state = data.states.QUEUED data.blurb = "queued" # Set output label data.name = self.get_output_name(output, data, tool, on_text, trans, incoming, history, wrapped_params.params, job_params) # Store output out_data[name] = data if output.actions: #Apply pre-job tool-output-dataset actions; e.g. setting metadata, changing format output_action_params = dict(out_data) output_action_params.update(incoming) output.actions.apply_action(data, output_action_params) # Store all changes to database trans.sa_session.flush() return data for name, output in tool.outputs.items(): if not filter_output(output, incoming): if output.collection: collections_manager = trans.app.dataset_collections_service # As far as I can tell - this is always true - but just verify assert set_output_history, "Cannot create dataset collection for this kind of tool." elements = odict() input_collections = dict([ (k, v[0]) for k, v in inp_dataset_collections.iteritems() ]) known_outputs = output.known_outputs( input_collections, collections_manager.type_registry) # Just to echo TODO elsewhere - this should be restructured to allow # nested collections. for output_part_def in known_outputs: effective_output_name = output_part_def.effective_output_name element = handle_output(effective_output_name, output_part_def.output_def) # Following hack causes dataset to no be added to history... child_dataset_names.add(effective_output_name) if set_output_history: history.add_dataset(element, set_hid=set_output_hid) trans.sa_session.add(element) trans.sa_session.flush() elements[output_part_def.element_identifier] = element if output.dynamic_structure: assert not elements # known_outputs must have been empty elements = collections_manager.ELEMENTS_UNINITIALIZED if mapping_over_collection: dc = collections_manager.create_dataset_collection( trans, collection_type=output.structure.collection_type, elements=elements, ) out_collections[name] = dc else: hdca_name = self.get_output_name( output, None, tool, on_text, trans, incoming, history, wrapped_params.params, job_params) hdca = collections_manager.create( trans, history, name=hdca_name, collection_type=output.structure.collection_type, elements=elements, ) # name here is name of the output element - not name # of the hdca. out_collection_instances[name] = hdca else: handle_output_timer = ExecutionTimer() handle_output(name, output) log.info("Handled output %s" % handle_output_timer) # Add all the top-level (non-child) datasets to the history unless otherwise specified for name in out_data.keys(): if name not in child_dataset_names and name not in incoming: # don't add children; or already existing datasets, i.e. async created data = out_data[name] if set_output_history: history.add_dataset(data, set_hid=set_output_hid) trans.sa_session.add(data) trans.sa_session.flush() # Add all the children to their parents for parent_name, child_name in parent_to_child_pairs: parent_dataset = out_data[parent_name] child_dataset = out_data[child_name] parent_dataset.children.append(child_dataset) # Store data after custom code runs trans.sa_session.flush() # Create the job object job = trans.app.model.Job() if hasattr(trans, "get_galaxy_session"): galaxy_session = trans.get_galaxy_session() # If we're submitting from the API, there won't be a session. if type(galaxy_session) == trans.model.GalaxySession: job.session_id = galaxy_session.id if trans.user is not None: job.user_id = trans.user.id job.history_id = history.id job.tool_id = tool.id try: # For backward compatibility, some tools may not have versions yet. job.tool_version = tool.version except: job.tool_version = "1.0.0" # FIXME: Don't need all of incoming here, just the defined parameters # from the tool. We need to deal with tools that pass all post # parameters to the command as a special case. for name, (dataset_collection, reduced) in inp_dataset_collections.iteritems(): # TODO: Does this work if nested in repeat/conditional? if reduced: incoming[ name] = "__collection_reduce__|%s" % dataset_collection.id # Should verify security? We check security of individual # datasets below? job.add_input_dataset_collection(name, dataset_collection) for name, value in tool.params_to_strings(incoming, trans.app).iteritems(): job.add_parameter(name, value) current_user_roles = trans.get_current_user_roles() access_timer = ExecutionTimer() for name, dataset in inp_data.iteritems(): if dataset: if not trans.app.security_agent.can_access_dataset( current_user_roles, dataset.dataset): raise "User does not have permission to use a dataset (%s) provided for input." % data.id job.add_input_dataset(name, dataset) else: job.add_input_dataset(name, None) log.info("Verified access to datasets %s" % access_timer) for name, dataset in out_data.iteritems(): job.add_output_dataset(name, dataset) for name, dataset_collection in out_collections.iteritems(): job.add_implicit_output_dataset_collection(name, dataset_collection) for name, dataset_collection_instance in out_collection_instances.iteritems( ): job.add_output_dataset_collection(name, dataset_collection_instance) job.object_store_id = object_store_populator.object_store_id if job_params: job.params = dumps(job_params) job.set_handler(tool.get_job_handler(job_params)) trans.sa_session.add(job) # Now that we have a job id, we can remap any outputs if this is a rerun and the user chose to continue dependent jobs # This functionality requires tracking jobs in the database. if trans.app.config.track_jobs_in_database and rerun_remap_job_id is not None: try: old_job = trans.sa_session.query( trans.app.model.Job).get(rerun_remap_job_id) assert old_job is not None, '(%s/%s): Old job id is invalid' % ( rerun_remap_job_id, job.id) assert old_job.tool_id == job.tool_id, '(%s/%s): Old tool id (%s) does not match rerun tool id (%s)' % ( old_job.id, job.id, old_job.tool_id, job.tool_id) if trans.user is not None: assert old_job.user_id == trans.user.id, '(%s/%s): Old user id (%s) does not match rerun user id (%s)' % ( old_job.id, job.id, old_job.user_id, trans.user.id) elif trans.user is None and type( galaxy_session) == trans.model.GalaxySession: assert old_job.session_id == galaxy_session.id, '(%s/%s): Old session id (%s) does not match rerun session id (%s)' % ( old_job.id, job.id, old_job.session_id, galaxy_session.id) else: raise Exception( '(%s/%s): Remapping via the API is not (yet) supported' % (old_job.id, job.id)) for jtod in old_job.output_datasets: for (job_to_remap, jtid) in [(jtid.job, jtid) for jtid in jtod.dataset.dependent_jobs]: if (trans.user is not None and job_to_remap.user_id == trans.user.id) or (trans.user is None and job_to_remap.session_id == galaxy_session.id): if job_to_remap.state == job_to_remap.states.PAUSED: job_to_remap.state = job_to_remap.states.NEW for hda in [ dep_jtod.dataset for dep_jtod in job_to_remap.output_datasets ]: if hda.state == hda.states.PAUSED: hda.state = hda.states.NEW hda.info = None for p in job_to_remap.parameters: if p.name == jtid.name and p.value == str( jtod.dataset.id): p.value = str(out_data[jtod.name].id) jtid.dataset = out_data[jtod.name] jtid.dataset.hid = jtod.dataset.hid log.info( 'Job %s input HDA %s remapped to new HDA %s' % (job_to_remap.id, jtod.dataset.id, jtid.dataset.id)) trans.sa_session.add(job_to_remap) trans.sa_session.add(jtid) jtod.dataset.visible = False trans.sa_session.add(jtod) except Exception, e: log.exception('Cannot remap rerun dependencies.')
def execute(self, tool, trans, incoming={}, return_job=False, set_output_hid=True, set_output_history=True, history=None, job_params=None ): """ Executes a tool, creating job and tool outputs, associating them, and submitting the job to the job queue. If history is not specified, use trans.history as destination for tool's output datasets. """ def make_dict_copy( from_dict ): """ Makes a copy of input dictionary from_dict such that all values that are dictionaries result in creation of a new dictionary ( a sort of deepcopy ). We may need to handle other complex types ( e.g., lists, etc ), but not sure... Yes, we need to handle lists (and now are)... """ copy_from_dict = {} for key, value in from_dict.items(): if type( value ).__name__ == 'dict': copy_from_dict[ key ] = make_dict_copy( value ) elif isinstance( value, list ): copy_from_dict[ key ] = make_list_copy( value ) else: copy_from_dict[ key ] = value return copy_from_dict def make_list_copy( from_list ): new_list = [] for value in from_list: if isinstance( value, dict ): new_list.append( make_dict_copy( value ) ) elif isinstance( value, list ): new_list.append( make_list_copy( value ) ) else: new_list.append( value ) return new_list def wrap_values( inputs, input_values, skip_missing_values = False ): # Wrap tool inputs as necessary for input in inputs.itervalues(): if input.name not in input_values and skip_missing_values: continue if isinstance( input, Repeat ): for d in input_values[ input.name ]: wrap_values( input.inputs, d, skip_missing_values = skip_missing_values ) elif isinstance( input, Conditional ): values = input_values[ input.name ] current = values[ "__current_case__" ] wrap_values( input.cases[current].inputs, values, skip_missing_values = skip_missing_values ) elif isinstance( input, DataToolParameter ) and input.multiple: input_values[ input.name ] = \ galaxy.tools.DatasetListWrapper( input_values[ input.name ], datatypes_registry = trans.app.datatypes_registry, tool = tool, name = input.name ) elif isinstance( input, DataToolParameter ): input_values[ input.name ] = \ galaxy.tools.DatasetFilenameWrapper( input_values[ input.name ], datatypes_registry = trans.app.datatypes_registry, tool = tool, name = input.name ) elif isinstance( input, SelectToolParameter ): input_values[ input.name ] = galaxy.tools.SelectToolParameterWrapper( input, input_values[ input.name ], tool.app, other_values = incoming ) else: input_values[ input.name ] = galaxy.tools.InputValueWrapper( input, input_values[ input.name ], incoming ) # Set history. if not history: history = trans.history out_data = odict() # Collect any input datasets from the incoming parameters inp_data = self.collect_input_datasets( tool, incoming, trans ) # Deal with input dataset names, 'dbkey' and types input_names = [] input_ext = 'data' input_dbkey = incoming.get( "dbkey", "?" ) for name, data in inp_data.items(): if not data: data = NoneDataset( datatypes_registry = trans.app.datatypes_registry ) continue # Convert LDDA to an HDA. if isinstance(data, LibraryDatasetDatasetAssociation): data = data.to_history_dataset_association( None ) inp_data[name] = data else: # HDA if data.hid: input_names.append( 'data %s' % data.hid ) input_ext = data.ext if data.dbkey not in [None, '?']: input_dbkey = data.dbkey # Collect chromInfo dataset and add as parameters to incoming db_datasets = {} db_dataset = trans.db_dataset_for( input_dbkey ) if db_dataset: db_datasets[ "chromInfo" ] = db_dataset incoming[ "chromInfo" ] = db_dataset.file_name else: # For custom builds, chrom info resides in converted dataset; for built-in builds, chrom info resides in tool-data/shared. chrom_info = None if trans.user and ( 'dbkeys' in trans.user.preferences ) and ( input_dbkey in from_json_string( trans.user.preferences[ 'dbkeys' ] ) ): # Custom build. custom_build_dict = from_json_string( trans.user.preferences[ 'dbkeys' ] )[ input_dbkey ] if 'fasta' in custom_build_dict: build_fasta_dataset = trans.app.model.HistoryDatasetAssociation.get( custom_build_dict[ 'fasta' ] ) chrom_info = build_fasta_dataset.get_converted_dataset( trans, 'len' ).file_name if not chrom_info: # Default to built-in build. chrom_info = os.path.join( trans.app.config.tool_data_path, 'shared','ucsc','chrom', "%s.len" % input_dbkey ) incoming[ "chromInfo" ] = chrom_info inp_data.update( db_datasets ) # Determine output dataset permission/roles list existing_datasets = [ inp for inp in inp_data.values() if inp ] if existing_datasets: output_permissions = trans.app.security_agent.guess_derived_permissions_for_datasets( existing_datasets ) else: # No valid inputs, we will use history defaults output_permissions = trans.app.security_agent.history_get_default_permissions( history ) # Build name for output datasets based on tool name and input names if len( input_names ) == 1: on_text = input_names[0] elif len( input_names ) == 2: on_text = '%s and %s' % tuple(input_names[0:2]) elif len( input_names ) == 3: on_text = '%s, %s, and %s' % tuple(input_names[0:3]) elif len( input_names ) > 3: on_text = '%s, %s, and others' % tuple(input_names[0:2]) else: on_text = "" # Add the dbkey to the incoming parameters incoming[ "dbkey" ] = input_dbkey params = None #wrapped params are used by change_format action and by output.label; only perform this wrapping once, as needed # Keep track of parent / child relationships, we'll create all the # datasets first, then create the associations parent_to_child_pairs = [] child_dataset_names = set() object_store_id = None for name, output in tool.outputs.items(): for filter in output.filters: try: if not eval( filter.text.strip(), globals(), incoming ): break #do not create this dataset except Exception, e: log.debug( 'Dataset output filter failed: %s' % e ) else: #all filters passed if output.parent: parent_to_child_pairs.append( ( output.parent, name ) ) child_dataset_names.add( name ) ## What is the following hack for? Need to document under what ## conditions can the following occur? ([email protected]) # HACK: the output data has already been created # this happens i.e. as a result of the async controller if name in incoming: dataid = incoming[name] data = trans.sa_session.query( trans.app.model.HistoryDatasetAssociation ).get( dataid ) assert data != None out_data[name] = data else: # the type should match the input ext = output.format if ext == "input": ext = input_ext if output.format_source is not None and output.format_source in inp_data: try: ext = inp_data[output.format_source].ext except Exception, e: pass #process change_format tags if output.change_format: if params is None: params = make_dict_copy( incoming ) wrap_values( tool.inputs, params, skip_missing_values = not tool.check_values ) for change_elem in output.change_format: for when_elem in change_elem.findall( 'when' ): check = when_elem.get( 'input', None ) if check is not None: try: if '$' not in check: #allow a simple name or more complex specifications check = '${%s}' % check if str( fill_template( check, context = params ) ) == when_elem.get( 'value', None ): ext = when_elem.get( 'format', ext ) except: #bad tag input value; possibly referencing a param within a different conditional when block or other nonexistent grouping construct continue else: check = when_elem.get( 'input_dataset', None ) if check is not None: check = inp_data.get( check, None ) if check is not None: if str( getattr( check, when_elem.get( 'attribute' ) ) ) == when_elem.get( 'value', None ): ext = when_elem.get( 'format', ext ) data = trans.app.model.HistoryDatasetAssociation( extension=ext, create_dataset=True, sa_session=trans.sa_session ) if output.hidden: data.visible = False # Commit the dataset immediately so it gets database assigned unique id trans.sa_session.add( data ) trans.sa_session.flush() trans.app.security_agent.set_all_dataset_permissions( data.dataset, output_permissions ) # Create an empty file immediately. The first dataset will be # created in the "default" store, all others will be created in # the same store as the first. data.dataset.object_store_id = object_store_id try: trans.app.object_store.create( data.dataset ) except ObjectInvalid: raise Exception('Unable to create output dataset: object store is full') object_store_id = data.dataset.object_store_id # these will be the same thing after the first output # This may not be neccesary with the new parent/child associations data.designation = name # Copy metadata from one of the inputs if requested. if output.metadata_source: data.init_meta( copy_from=inp_data[output.metadata_source] ) else: data.init_meta() # Take dbkey from LAST input data.dbkey = str(input_dbkey) # Set state # FIXME: shouldn't this be NEW until the job runner changes it? data.state = data.states.QUEUED data.blurb = "queued" # Set output label if output.label: if params is None: params = make_dict_copy( incoming ) # wrapping the params allows the tool config to contain things like # <outputs> # <data format="input" name="output" label="Blat on ${<input_param>.name}" /> # </outputs> wrap_values( tool.inputs, params, skip_missing_values = not tool.check_values ) #tool (only needing to be set once) and on_string (set differently for each label) are overwritten for each output dataset label being determined params['tool'] = tool params['on_string'] = on_text params['time']=logging.time.strftime('%X %x', logging.time.gmtime()) data.name = fill_template( output.label, context=params ) else: data.name = tool.name if on_text: data.name += ( " on " + on_text ) # Store output out_data[ name ] = data if output.actions: #Apply pre-job tool-output-dataset actions; e.g. setting metadata, changing format output_action_params = dict( out_data ) output_action_params.update( incoming ) output.actions.apply_action( data, output_action_params ) # Store all changes to database trans.sa_session.flush()
def execute(self, tool, trans, incoming={}, return_job=False, set_output_hid=True, set_output_history=True, history=None, job_params=None, rerun_remap_job_id=None): """ Executes a tool, creating job and tool outputs, associating them, and submitting the job to the job queue. If history is not specified, use trans.history as destination for tool's output datasets. """ # Set history. if not history: history = tool.get_default_history_by_trans(trans, create=True) out_data = odict() # Track input dataset collections - but replace with simply lists so collect # input datasets can process these normally. inp_dataset_collections = self.collect_input_dataset_collections( tool, incoming, trans) # Collect any input datasets from the incoming parameters inp_data = self.collect_input_datasets(tool, incoming, trans) # Deal with input dataset names, 'dbkey' and types input_names = [] input_ext = 'data' input_dbkey = incoming.get("dbkey", "?") for name, data in inp_data.items(): if not data: data = NoneDataset( datatypes_registry=trans.app.datatypes_registry) continue # Convert LDDA to an HDA. if isinstance(data, LibraryDatasetDatasetAssociation): data = data.to_history_dataset_association(None) inp_data[name] = data else: # HDA if data.hid: input_names.append('data %s' % data.hid) input_ext = data.ext if data.dbkey not in [None, '?']: input_dbkey = data.dbkey # Collect chromInfo dataset and add as parameters to incoming db_datasets = {} (chrom_info, db_dataset) = trans.app.genome_builds.get_chrom_info( input_dbkey, trans=trans, custom_build_hack_get_len_from_fasta_conversion=tool.id != 'CONVERTER_fasta_to_len') if db_dataset: inp_data.update({"chromInfo": db_dataset}) incoming["chromInfo"] = chrom_info # Determine output dataset permission/roles list existing_datasets = [inp for inp in inp_data.values() if inp] if existing_datasets: output_permissions = trans.app.security_agent.guess_derived_permissions_for_datasets( existing_datasets) else: # No valid inputs, we will use history defaults output_permissions = trans.app.security_agent.history_get_default_permissions( history) # Build name for output datasets based on tool name and input names on_text = on_text_for_names(input_names) # Add the dbkey to the incoming parameters incoming["dbkey"] = input_dbkey # wrapped params are used by change_format action and by output.label; only perform this wrapping once, as needed wrapped_params = WrappedParameters(trans, tool, incoming) # Keep track of parent / child relationships, we'll create all the # datasets first, then create the associations parent_to_child_pairs = [] child_dataset_names = set() object_store_id = None for name, output in tool.outputs.items(): for filter in output.filters: try: if not eval(filter.text.strip(), globals(), incoming): break # do not create this dataset except Exception, e: log.debug('Dataset output filter failed: %s' % e) else: # all filters passed if output.parent: parent_to_child_pairs.append((output.parent, name)) child_dataset_names.add(name) ## What is the following hack for? Need to document under what ## conditions can the following occur? ([email protected]) # HACK: the output data has already been created # this happens i.e. as a result of the async controller if name in incoming: dataid = incoming[name] data = trans.sa_session.query( trans.app.model.HistoryDatasetAssociation).get(dataid) assert data != None out_data[name] = data else: # the type should match the input ext = output.format if ext == "input": ext = input_ext if output.format_source is not None and output.format_source in inp_data: try: input_dataset = inp_data[output.format_source] input_extension = input_dataset.ext ext = input_extension except Exception, e: pass #process change_format tags if output.change_format: for change_elem in output.change_format: for when_elem in change_elem.findall('when'): check = when_elem.get('input', None) if check is not None: try: if '$' not in check: #allow a simple name or more complex specifications check = '${%s}' % check if str( fill_template( check, context=wrapped_params. params)) == when_elem.get( 'value', None): ext = when_elem.get('format', ext) except: # bad tag input value; possibly referencing a param within a different conditional when block or other nonexistent grouping construct continue else: check = when_elem.get( 'input_dataset', None) if check is not None: check = inp_data.get(check, None) if check is not None: if str( getattr( check, when_elem.get( 'attribute')) ) == when_elem.get('value', None): ext = when_elem.get( 'format', ext) data = trans.app.model.HistoryDatasetAssociation( extension=ext, create_dataset=True, sa_session=trans.sa_session) if output.hidden: data.visible = False # Commit the dataset immediately so it gets database assigned unique id trans.sa_session.add(data) trans.sa_session.flush() trans.app.security_agent.set_all_dataset_permissions( data.dataset, output_permissions) # Create an empty file immediately. The first dataset will be # created in the "default" store, all others will be created in # the same store as the first. data.dataset.object_store_id = object_store_id try: trans.app.object_store.create(data.dataset) except ObjectInvalid: raise Exception( 'Unable to create output dataset: object store is full' ) object_store_id = data.dataset.object_store_id # these will be the same thing after the first output # This may not be neccesary with the new parent/child associations data.designation = name # Copy metadata from one of the inputs if requested. if output.metadata_source: data.init_meta(copy_from=inp_data[output.metadata_source]) else: data.init_meta() # Take dbkey from LAST input data.dbkey = str(input_dbkey) # Set state # FIXME: shouldn't this be NEW until the job runner changes it? data.state = data.states.QUEUED data.blurb = "queued" # Set output label data.name = self.get_output_name(output, data, tool, on_text, trans, incoming, history, wrapped_params.params, job_params) # Store output out_data[name] = data if output.actions: #Apply pre-job tool-output-dataset actions; e.g. setting metadata, changing format output_action_params = dict(out_data) output_action_params.update(incoming) output.actions.apply_action(data, output_action_params) # Store all changes to database trans.sa_session.flush()
def execute(self, tool, trans, incoming={}, return_job=False, set_output_hid=True, set_output_history=True, history=None, job_params=None, rerun_remap_job_id=None, mapping_over_collection=False, execution_cache=None ): """ Executes a tool, creating job and tool outputs, associating them, and submitting the job to the job queue. If history is not specified, use trans.history as destination for tool's output datasets. """ app = trans.app if execution_cache is None: execution_cache = ToolExecutionCache(trans) current_user_roles = execution_cache.current_user_roles assert tool.allow_user_access( trans.user ), "User (%s) is not allowed to access this tool." % ( trans.user ) # Set history. if not history: history = tool.get_default_history_by_trans( trans, create=True ) if history not in trans.sa_session: history = trans.sa_session.query( app.model.History ).get( history.id ) out_data = odict() out_collections = {} out_collection_instances = {} # Track input dataset collections - but replace with simply lists so collect # input datasets can process these normally. inp_dataset_collections = self.collect_input_dataset_collections( tool, incoming ) # Collect any input datasets from the incoming parameters inp_data = self.collect_input_datasets( tool, incoming, trans, current_user_roles=current_user_roles ) # Deal with input dataset names, 'dbkey' and types input_names = [] input_ext = 'data' input_dbkey = incoming.get( "dbkey", "?" ) for name, data in reversed(inp_data.items()): if not data: data = NoneDataset( datatypes_registry=app.datatypes_registry ) continue # Convert LDDA to an HDA. if isinstance(data, LibraryDatasetDatasetAssociation): data = data.to_history_dataset_association( None ) inp_data[name] = data else: # HDA if data.hid: input_names.append( 'data %s' % data.hid ) input_ext = data.ext if data.dbkey not in [None, '?']: input_dbkey = data.dbkey identifier = getattr( data, "element_identifier", None ) if identifier is not None: incoming[ "%s|__identifier__" % name ] = identifier # Collect chromInfo dataset and add as parameters to incoming ( chrom_info, db_dataset ) = app.genome_builds.get_chrom_info( input_dbkey, trans=trans, custom_build_hack_get_len_from_fasta_conversion=tool.id != 'CONVERTER_fasta_to_len' ) if db_dataset: inp_data.update( { "chromInfo": db_dataset } ) incoming[ "chromInfo" ] = chrom_info # Determine output dataset permission/roles list existing_datasets = [ inp for inp in inp_data.values() if inp ] if existing_datasets: output_permissions = app.security_agent.guess_derived_permissions_for_datasets( existing_datasets ) else: # No valid inputs, we will use history defaults output_permissions = app.security_agent.history_get_default_permissions( history ) # Build name for output datasets based on tool name and input names on_text = on_text_for_names( input_names ) # Add the dbkey to the incoming parameters incoming[ "dbkey" ] = input_dbkey # wrapped params are used by change_format action and by output.label; only perform this wrapping once, as needed wrapped_params = WrappedParameters( trans, tool, incoming ) # Keep track of parent / child relationships, we'll create all the # datasets first, then create the associations parent_to_child_pairs = [] child_dataset_names = set() object_store_populator = ObjectStorePopulator( app ) def handle_output( name, output, hidden=None ): if output.parent: parent_to_child_pairs.append( ( output.parent, name ) ) child_dataset_names.add( name ) # What is the following hack for? Need to document under what # conditions can the following occur? ([email protected]) # HACK: the output data has already been created # this happens i.e. as a result of the async controller if name in incoming: dataid = incoming[name] data = trans.sa_session.query( app.model.HistoryDatasetAssociation ).get( dataid ) assert data is not None out_data[name] = data else:
def execute(self, tool, trans, incoming={}, return_job=False, set_output_hid=True, set_output_history=True, history=None, job_params=None, rerun_remap_job_id=None): """ Executes a tool, creating job and tool outputs, associating them, and submitting the job to the job queue. If history is not specified, use trans.history as destination for tool's output datasets. """ # Set history. if not history: history = tool.get_default_history_by_trans( trans, create=True ) out_data = odict() # Track input dataset collections - but replace with simply lists so collect # input datasets can process these normally. inp_dataset_collections = self.collect_input_dataset_collections( tool, incoming ) # Collect any input datasets from the incoming parameters inp_data = self.collect_input_datasets( tool, incoming, trans ) # Deal with input dataset names, 'dbkey' and types input_names = [] input_ext = 'data' input_dbkey = incoming.get( "dbkey", "?" ) for name, data in inp_data.items(): if not data: data = NoneDataset( datatypes_registry=trans.app.datatypes_registry ) continue # Convert LDDA to an HDA. if isinstance(data, LibraryDatasetDatasetAssociation): data = data.to_history_dataset_association( None ) inp_data[name] = data else: # HDA if data.hid: input_names.append( 'data %s' % data.hid ) input_ext = data.ext if data.dbkey not in [None, '?']: input_dbkey = data.dbkey # Collect chromInfo dataset and add as parameters to incoming ( chrom_info, db_dataset ) = trans.app.genome_builds.get_chrom_info( input_dbkey, trans=trans, custom_build_hack_get_len_from_fasta_conversion=tool.id != 'CONVERTER_fasta_to_len' ) if db_dataset: inp_data.update( { "chromInfo": db_dataset } ) incoming[ "chromInfo" ] = chrom_info # Determine output dataset permission/roles list existing_datasets = [ inp for inp in inp_data.values() if inp ] if existing_datasets: output_permissions = trans.app.security_agent.guess_derived_permissions_for_datasets( existing_datasets ) else: # No valid inputs, we will use history defaults output_permissions = trans.app.security_agent.history_get_default_permissions( history ) # Build name for output datasets based on tool name and input names on_text = on_text_for_names( input_names ) # Add the dbkey to the incoming parameters incoming[ "dbkey" ] = input_dbkey # wrapped params are used by change_format action and by output.label; only perform this wrapping once, as needed wrapped_params = WrappedParameters( trans, tool, incoming ) # Keep track of parent / child relationships, we'll create all the # datasets first, then create the associations parent_to_child_pairs = [] child_dataset_names = set() object_store_populator = ObjectStorePopulator( trans.app ) def handle_output( name, output ): if output.parent: parent_to_child_pairs.append( ( output.parent, name ) ) child_dataset_names.add( name ) ## What is the following hack for? Need to document under what ## conditions can the following occur? ([email protected]) # HACK: the output data has already been created # this happens i.e. as a result of the async controller if name in incoming: dataid = incoming[name] data = trans.sa_session.query( trans.app.model.HistoryDatasetAssociation ).get( dataid ) assert data is not None out_data[name] = data else: ext = determine_output_format( output, wrapped_params.params, inp_data, input_ext ) data = trans.app.model.HistoryDatasetAssociation( extension=ext, create_dataset=True, sa_session=trans.sa_session ) if output.hidden: data.visible = False # Commit the dataset immediately so it gets database assigned unique id trans.sa_session.add( data ) trans.sa_session.flush() trans.app.security_agent.set_all_dataset_permissions( data.dataset, output_permissions ) object_store_populator.set_object_store_id( data ) # This may not be neccesary with the new parent/child associations data.designation = name # Copy metadata from one of the inputs if requested. if output.metadata_source: data.init_meta( copy_from=inp_data[output.metadata_source] ) else: data.init_meta() # Take dbkey from LAST input data.dbkey = str(input_dbkey) # Set state # FIXME: shouldn't this be NEW until the job runner changes it? data.state = data.states.QUEUED data.blurb = "queued" # Set output label data.name = self.get_output_name( output, data, tool, on_text, trans, incoming, history, wrapped_params.params, job_params ) # Store output out_data[ name ] = data if output.actions: #Apply pre-job tool-output-dataset actions; e.g. setting metadata, changing format output_action_params = dict( out_data ) output_action_params.update( incoming ) output.actions.apply_action( data, output_action_params ) # Store all changes to database trans.sa_session.flush() for name, output in tool.outputs.items(): if not filter_output(output, incoming): handle_output( name, output ) # Add all the top-level (non-child) datasets to the history unless otherwise specified for name in out_data.keys(): if name not in child_dataset_names and name not in incoming: # don't add children; or already existing datasets, i.e. async created data = out_data[ name ] if set_output_history: history.add_dataset( data, set_hid=set_output_hid ) trans.sa_session.add( data ) trans.sa_session.flush() # Add all the children to their parents for parent_name, child_name in parent_to_child_pairs: parent_dataset = out_data[ parent_name ] child_dataset = out_data[ child_name ] parent_dataset.children.append( child_dataset ) # Store data after custom code runs trans.sa_session.flush() # Create the job object job = trans.app.model.Job() if hasattr( trans, "get_galaxy_session" ): galaxy_session = trans.get_galaxy_session() # If we're submitting from the API, there won't be a session. if type( galaxy_session ) == trans.model.GalaxySession: job.session_id = galaxy_session.id if trans.user is not None: job.user_id = trans.user.id job.history_id = history.id job.tool_id = tool.id try: # For backward compatibility, some tools may not have versions yet. job.tool_version = tool.version except: job.tool_version = "1.0.0" # FIXME: Don't need all of incoming here, just the defined parameters # from the tool. We need to deal with tools that pass all post # parameters to the command as a special case. for name, ( dataset_collection, reduced ) in inp_dataset_collections.iteritems(): # TODO: Does this work if nested in repeat/conditional? if reduced: incoming[ name ] = "__collection_reduce__|%s" % dataset_collection.id # Should verify security? We check security of individual # datasets below? job.add_input_dataset_collection( name, dataset_collection ) for name, value in tool.params_to_strings( incoming, trans.app ).iteritems(): job.add_parameter( name, value ) current_user_roles = trans.get_current_user_roles() for name, dataset in inp_data.iteritems(): if dataset: if not trans.app.security_agent.can_access_dataset( current_user_roles, dataset.dataset ): raise "User does not have permission to use a dataset (%s) provided for input." % data.id job.add_input_dataset( name, dataset ) else: job.add_input_dataset( name, None ) for name, dataset in out_data.iteritems(): job.add_output_dataset( name, dataset ) job.object_store_id = object_store_populator.object_store_id if job_params: job.params = dumps( job_params ) job.set_handler(tool.get_job_handler(job_params)) trans.sa_session.add( job ) # Now that we have a job id, we can remap any outputs if this is a rerun and the user chose to continue dependent jobs # This functionality requires tracking jobs in the database. if trans.app.config.track_jobs_in_database and rerun_remap_job_id is not None: try: old_job = trans.sa_session.query( trans.app.model.Job ).get(rerun_remap_job_id) assert old_job is not None, '(%s/%s): Old job id is invalid' % (rerun_remap_job_id, job.id) assert old_job.tool_id == job.tool_id, '(%s/%s): Old tool id (%s) does not match rerun tool id (%s)' % (old_job.id, job.id, old_job.tool_id, job.tool_id) if trans.user is not None: assert old_job.user_id == trans.user.id, '(%s/%s): Old user id (%s) does not match rerun user id (%s)' % (old_job.id, job.id, old_job.user_id, trans.user.id) elif trans.user is None and type( galaxy_session ) == trans.model.GalaxySession: assert old_job.session_id == galaxy_session.id, '(%s/%s): Old session id (%s) does not match rerun session id (%s)' % (old_job.id, job.id, old_job.session_id, galaxy_session.id) else: raise Exception('(%s/%s): Remapping via the API is not (yet) supported' % (old_job.id, job.id)) for jtod in old_job.output_datasets: for (job_to_remap, jtid) in [(jtid.job, jtid) for jtid in jtod.dataset.dependent_jobs]: if (trans.user is not None and job_to_remap.user_id == trans.user.id) or (trans.user is None and job_to_remap.session_id == galaxy_session.id): if job_to_remap.state == job_to_remap.states.PAUSED: job_to_remap.state = job_to_remap.states.NEW for hda in [ dep_jtod.dataset for dep_jtod in job_to_remap.output_datasets ]: if hda.state == hda.states.PAUSED: hda.state = hda.states.NEW hda.info = None for p in job_to_remap.parameters: if p.name == jtid.name and p.value == str(jtod.dataset.id): p.value = str(out_data[jtod.name].id) jtid.dataset = out_data[jtod.name] jtid.dataset.hid = jtod.dataset.hid log.info('Job %s input HDA %s remapped to new HDA %s' % (job_to_remap.id, jtod.dataset.id, jtid.dataset.id)) trans.sa_session.add(job_to_remap) trans.sa_session.add(jtid) jtod.dataset.visible = False trans.sa_session.add(jtod) except Exception, e: log.exception('Cannot remap rerun dependencies.')
def execute(self, tool, trans, incoming={}, return_job=False, set_output_hid=True, set_output_history=True, history=None, job_params=None, rerun_remap_job_id=None): """ Executes a tool, creating job and tool outputs, associating them, and submitting the job to the job queue. If history is not specified, use trans.history as destination for tool's output datasets. """ # Set history. if not history: history = tool.get_default_history_by_trans( trans, create=True ) out_data = odict() # Track input dataset collections - but replace with simply lists so collect # input datasets can process these normally. inp_dataset_collections = self.collect_input_dataset_collections( tool, incoming ) # Collect any input datasets from the incoming parameters inp_data = self.collect_input_datasets( tool, incoming, trans ) # Deal with input dataset names, 'dbkey' and types input_names = [] input_ext = 'data' input_dbkey = incoming.get( "dbkey", "?" ) for name, data in inp_data.items(): if not data: data = NoneDataset( datatypes_registry=trans.app.datatypes_registry ) continue # Convert LDDA to an HDA. if isinstance(data, LibraryDatasetDatasetAssociation): data = data.to_history_dataset_association( None ) inp_data[name] = data else: # HDA if data.hid: input_names.append( 'data %s' % data.hid ) input_ext = data.ext if data.dbkey not in [None, '?']: input_dbkey = data.dbkey # Collect chromInfo dataset and add as parameters to incoming db_datasets = {} ( chrom_info, db_dataset ) = trans.app.genome_builds.get_chrom_info( input_dbkey, trans=trans, custom_build_hack_get_len_from_fasta_conversion=tool.id!='CONVERTER_fasta_to_len' ) if db_dataset: inp_data.update( { "chromInfo": db_dataset } ) incoming[ "chromInfo" ] = chrom_info # Determine output dataset permission/roles list existing_datasets = [ inp for inp in inp_data.values() if inp ] if existing_datasets: output_permissions = trans.app.security_agent.guess_derived_permissions_for_datasets( existing_datasets ) else: # No valid inputs, we will use history defaults output_permissions = trans.app.security_agent.history_get_default_permissions( history ) # Build name for output datasets based on tool name and input names on_text = on_text_for_names( input_names ) # Add the dbkey to the incoming parameters incoming[ "dbkey" ] = input_dbkey # wrapped params are used by change_format action and by output.label; only perform this wrapping once, as needed wrapped_params = WrappedParameters( trans, tool, incoming ) # Keep track of parent / child relationships, we'll create all the # datasets first, then create the associations parent_to_child_pairs = [] child_dataset_names = set() object_store_id = None for name, output in tool.outputs.items(): for filter in output.filters: try: if not eval( filter.text.strip(), globals(), incoming ): break # do not create this dataset except Exception, e: log.debug( 'Dataset output filter failed: %s' % e ) else: # all filters passed if output.parent: parent_to_child_pairs.append( ( output.parent, name ) ) child_dataset_names.add( name ) ## What is the following hack for? Need to document under what ## conditions can the following occur? ([email protected]) # HACK: the output data has already been created # this happens i.e. as a result of the async controller if name in incoming: dataid = incoming[name] data = trans.sa_session.query( trans.app.model.HistoryDatasetAssociation ).get( dataid ) assert data != None out_data[name] = data else: # the type should match the input ext = output.format if ext == "input": ext = input_ext if output.format_source is not None and output.format_source in inp_data: try: input_dataset = inp_data[output.format_source] input_extension = input_dataset.ext ext = input_extension except Exception, e: pass #process change_format tags if output.change_format: for change_elem in output.change_format: for when_elem in change_elem.findall( 'when' ): check = when_elem.get( 'input', None ) if check is not None: try: if '$' not in check: #allow a simple name or more complex specifications check = '${%s}' % check if str( fill_template( check, context=wrapped_params.params ) ) == when_elem.get( 'value', None ): ext = when_elem.get( 'format', ext ) except: # bad tag input value; possibly referencing a param within a different conditional when block or other nonexistent grouping construct continue else: check = when_elem.get( 'input_dataset', None ) if check is not None: check = inp_data.get( check, None ) if check is not None: if str( getattr( check, when_elem.get( 'attribute' ) ) ) == when_elem.get( 'value', None ): ext = when_elem.get( 'format', ext ) data = trans.app.model.HistoryDatasetAssociation( extension=ext, create_dataset=True, sa_session=trans.sa_session ) if output.hidden: data.visible = False # Commit the dataset immediately so it gets database assigned unique id trans.sa_session.add( data ) trans.sa_session.flush() trans.app.security_agent.set_all_dataset_permissions( data.dataset, output_permissions ) # Create an empty file immediately. The first dataset will be # created in the "default" store, all others will be created in # the same store as the first. data.dataset.object_store_id = object_store_id try: trans.app.object_store.create( data.dataset ) except ObjectInvalid: raise Exception('Unable to create output dataset: object store is full') object_store_id = data.dataset.object_store_id # these will be the same thing after the first output # This may not be neccesary with the new parent/child associations data.designation = name # Copy metadata from one of the inputs if requested. if output.metadata_source: data.init_meta( copy_from=inp_data[output.metadata_source] ) else: data.init_meta() # Take dbkey from LAST input data.dbkey = str(input_dbkey) # Set state # FIXME: shouldn't this be NEW until the job runner changes it? data.state = data.states.QUEUED data.blurb = "queued" # Set output label data.name = self.get_output_name( output, data, tool, on_text, trans, incoming, history, wrapped_params.params, job_params ) # Store output out_data[ name ] = data if output.actions: #Apply pre-job tool-output-dataset actions; e.g. setting metadata, changing format output_action_params = dict( out_data ) output_action_params.update( incoming ) output.actions.apply_action( data, output_action_params ) # Store all changes to database trans.sa_session.flush()
def execute(self, tool, trans, incoming={}, set_output_hid=True ): def make_dict_copy( from_dict ): """ Makes a copy of input dictionary from_dict such that all values that are dictionaries result in creation of a new dictionary ( a sort of deepcopy ). We may need to handle other complex types ( e.g., lists, etc ), but not sure... """ copy_from_dict = {} for key, value in from_dict.items(): if type( value ).__name__ == 'dict': copy_from_dict[ key ] = make_dict_copy( value ) else: copy_from_dict[ key ] = value return copy_from_dict def wrap_values( inputs, input_values ): # Wrap tool inputs as necessary for input in inputs.itervalues(): if isinstance( input, Repeat ): for d in input_values[ input.name ]: wrap_values( input.inputs, d ) elif isinstance( input, Conditional ): values = input_values[ input.name ] current = values[ "__current_case__" ] wrap_values( input.cases[current].inputs, values ) elif isinstance( input, DataToolParameter ): input_values[ input.name ] = \ galaxy.tools.DatasetFilenameWrapper( input_values[ input.name ], datatypes_registry = trans.app.datatypes_registry, tool = tool, name = input.name ) elif isinstance( input, SelectToolParameter ): input_values[ input.name ] = galaxy.tools.SelectToolParameterWrapper( input, input_values[ input.name ], tool.app, other_values = incoming ) else: input_values[ input.name ] = galaxy.tools.InputValueWrapper( input, input_values[ input.name ], incoming ) out_data = {} # Collect any input datasets from the incoming parameters inp_data = self.collect_input_datasets( tool, incoming, trans ) # Deal with input dataset names, 'dbkey' and types input_names = [] input_ext = 'data' input_dbkey = incoming.get( "dbkey", "?" ) for name, data in inp_data.items(): if data: input_names.append( 'data %s' % data.hid ) input_ext = data.ext else: data = NoneDataset( datatypes_registry = trans.app.datatypes_registry ) if data.dbkey not in [None, '?']: input_dbkey = data.dbkey # Collect chromInfo dataset and add as parameters to incoming db_datasets = {} db_dataset = trans.db_dataset_for( input_dbkey ) if db_dataset: db_datasets[ "chromInfo" ] = db_dataset incoming[ "chromInfo" ] = db_dataset.file_name else: incoming[ "chromInfo" ] = os.path.join( trans.app.config.tool_data_path, 'shared','ucsc','chrom', "%s.len" % input_dbkey ) inp_data.update( db_datasets ) # Determine output dataset permission/roles list existing_datasets = [ inp for inp in inp_data.values() if inp ] if existing_datasets: output_permissions = trans.app.security_agent.guess_derived_permissions_for_datasets( existing_datasets ) else: # No valid inputs, we will use history defaults output_permissions = trans.app.security_agent.history_get_default_permissions( trans.history ) # Build name for output datasets based on tool name and input names if len( input_names ) == 1: on_text = input_names[0] elif len( input_names ) == 2: on_text = '%s and %s' % tuple(input_names[0:2]) elif len( input_names ) == 3: on_text = '%s, %s, and %s' % tuple(input_names[0:3]) elif len( input_names ) > 3: on_text = '%s, %s, and others' % tuple(input_names[0:2]) else: on_text = "" # Add the dbkey to the incoming parameters incoming[ "dbkey" ] = input_dbkey # Keep track of parent / child relationships, we'll create all the # datasets first, then create the associations parent_to_child_pairs = [] child_dataset_names = set() for name, output in tool.outputs.items(): if output.parent: parent_to_child_pairs.append( ( output.parent, name ) ) child_dataset_names.add( name ) ## What is the following hack for? Need to document under what ## conditions can the following occur? ([email protected]) # HACK: the output data has already been created # this happens i.e. as a result of the async controller if name in incoming: dataid = incoming[name] data = trans.app.model.HistoryDatasetAssociation.get( dataid ) assert data != None out_data[name] = data else: # the type should match the input ext = output.format if ext == "input": ext = input_ext #process change_format tags if output.change_format: for change_elem in output.change_format: for when_elem in change_elem.findall( 'when' ): check = incoming.get( when_elem.get( 'input' ), None ) if check is not None: if check == when_elem.get( 'value', None ): ext = when_elem.get( 'format', ext ) else: check = when_elem.get( 'input_dataset', None ) if check is not None: check = inp_data.get( check, None ) if check is not None: if str( getattr( check, when_elem.get( 'attribute' ) ) ) == when_elem.get( 'value', None ): ext = when_elem.get( 'format', ext ) data = trans.app.model.HistoryDatasetAssociation( extension=ext, create_dataset=True ) # Commit the dataset immediately so it gets database assigned unique id data.flush() trans.app.security_agent.set_all_dataset_permissions( data.dataset, output_permissions ) # Create an empty file immediately open( data.file_name, "w" ).close() # This may not be neccesary with the new parent/child associations data.designation = name # Copy metadata from one of the inputs if requested. if output.metadata_source: data.init_meta( copy_from=inp_data[output.metadata_source] ) else: data.init_meta() # Take dbkey from LAST input data.dbkey = str(input_dbkey) # Set state # FIXME: shouldn't this be NEW until the job runner changes it? data.state = data.states.QUEUED data.blurb = "queued" # Set output label if output.label: params = make_dict_copy( incoming ) # wrapping the params allows the tool config to contain things like # <outputs> # <data format="input" name="output" label="Blat on ${<input_param>.name}" /> # </outputs> wrap_values( tool.inputs, params ) params['tool'] = tool params['on_string'] = on_text data.name = fill_template( output.label, context=params ) else: data.name = tool.name if on_text: data.name += ( " on " + on_text ) # Store output out_data[ name ] = data # Store all changes to database trans.app.model.flush() # Add all the top-level (non-child) datasets to the history for name in out_data.keys(): if name not in child_dataset_names and name not in incoming: #don't add children; or already existing datasets, i.e. async created data = out_data[ name ] trans.history.add_dataset( data, set_hid = set_output_hid ) data.flush() # Add all the children to their parents for parent_name, child_name in parent_to_child_pairs: parent_dataset = out_data[ parent_name ] child_dataset = out_data[ child_name ] parent_dataset.children.append( child_dataset ) # Store data after custom code runs trans.app.model.flush() # Create the job object job = trans.app.model.Job() job.session_id = trans.get_galaxy_session().id job.history_id = trans.history.id job.tool_id = tool.id try: # For backward compatibility, some tools may not have versions yet. job.tool_version = tool.version except: job.tool_version = "1.0.0" # FIXME: Don't need all of incoming here, just the defined parameters # from the tool. We need to deal with tools that pass all post # parameters to the command as a special case. for name, value in tool.params_to_strings( incoming, trans.app ).iteritems(): job.add_parameter( name, value ) for name, dataset in inp_data.iteritems(): if dataset: # TODO, Nate: Make sure the permitted actions here are appropriate. if not trans.app.security_agent.allow_action( trans.user, dataset.permitted_actions.DATASET_ACCESS, dataset=dataset ): raise "User does not have permission to use a dataset (%s) provided for input." % data.id job.add_input_dataset( name, dataset ) else: job.add_input_dataset( name, None ) for name, dataset in out_data.iteritems(): job.add_output_dataset( name, dataset ) trans.app.model.flush() # Some tools are not really executable, but jobs are still created for them ( for record keeping ). # Examples include tools that redirect to other applications ( epigraph ). These special tools must # include something that can be retrieved from the params ( e.g., REDIRECT_URL ) to keep the job # from being queued. if 'REDIRECT_URL' in incoming: # Get the dataset - there should only be 1 for name in inp_data.keys(): dataset = inp_data[ name ] redirect_url = tool.parse_redirect_url( dataset, incoming ) # GALAXY_URL should be include in the tool params to enable the external application # to send back to the current Galaxy instance GALAXY_URL = incoming.get( 'GALAXY_URL', None ) assert GALAXY_URL is not None, "GALAXY_URL parameter missing in tool config." redirect_url += "&GALAXY_URL=%s" % GALAXY_URL # Job should not be queued, so set state to ok job.state = JOB_OK job.info = "Redirected to: %s" % redirect_url job.flush() trans.response.send_redirect( url_for( controller='tool_runner', action='redirect', redirect_url=redirect_url ) ) else: # Queue the job for execution trans.app.job_queue.put( job.id, tool ) trans.log_event( "Added job to the job queue, id: %s" % str(job.id), tool_id=job.tool_id ) return out_data
def execute(self, tool, trans, incoming={}, set_output_hid=True ): def make_dict_copy( from_dict ): """ Makes a copy of input dictionary from_dict such that all values that are dictionaries result in creation of a new dictionary ( a sort of deepcopy ). We may need to handle other complex types ( e.g., lists, etc ), but not sure... """ copy_from_dict = {} for key, value in from_dict.items(): if type( value ).__name__ == 'dict': copy_from_dict[ key ] = make_dict_copy( value ) else: copy_from_dict[ key ] = value return copy_from_dict def wrap_values( inputs, input_values ): # Wrap tool inputs as necessary for input in inputs.itervalues(): if isinstance( input, Repeat ): for d in input_values[ input.name ]: wrap_values( input.inputs, d ) elif isinstance( input, Conditional ): values = input_values[ input.name ] current = values[ "__current_case__" ] wrap_values( input.cases[current].inputs, values ) elif isinstance( input, DataToolParameter ): input_values[ input.name ] = \ galaxy.tools.DatasetFilenameWrapper( input_values[ input.name ], datatypes_registry = trans.app.datatypes_registry, tool = tool, name = input.name ) elif isinstance( input, SelectToolParameter ): input_values[ input.name ] = galaxy.tools.SelectToolParameterWrapper( input, input_values[ input.name ], tool.app, other_values = incoming ) else: input_values[ input.name ] = galaxy.tools.InputValueWrapper( input, input_values[ input.name ], incoming ) out_data = {} # Collect any input datasets from the incoming parameters inp_data = self.collect_input_datasets( tool, incoming, trans ) # Deal with input dataset names, 'dbkey' and types input_names = [] input_ext = 'data' input_dbkey = incoming.get( "dbkey", "?" ) for name, data in inp_data.items(): if data: input_names.append( 'data %s' % data.hid ) input_ext = data.ext else: data = NoneDataset( datatypes_registry = trans.app.datatypes_registry ) if data.dbkey not in [None, '?']: input_dbkey = data.dbkey # Collect chromInfo dataset and add as parameters to incoming db_datasets = {} db_dataset = trans.db_dataset_for( input_dbkey ) if db_dataset: db_datasets[ "chromInfo" ] = db_dataset incoming[ "chromInfo" ] = db_dataset.file_name else: incoming[ "chromInfo" ] = os.path.join( trans.app.config.tool_data_path, 'shared','ucsc','chrom', "%s.len" % input_dbkey ) inp_data.update( db_datasets ) # Determine output dataset permission/roles list existing_datasets = [ inp for inp in inp_data.values() if inp ] if existing_datasets: output_permissions = trans.app.security_agent.guess_derived_permissions_for_datasets( existing_datasets ) else: # No valid inputs, we will use history defaults output_permissions = trans.app.security_agent.history_get_default_permissions( trans.history ) # Build name for output datasets based on tool name and input names if len( input_names ) == 1: on_text = input_names[0] elif len( input_names ) == 2: on_text = '%s and %s' % tuple(input_names[0:2]) elif len( input_names ) == 3: on_text = '%s, %s, and %s' % tuple(input_names[0:3]) elif len( input_names ) > 3: on_text = '%s, %s, and others' % tuple(input_names[0:2]) else: on_text = "" # Add the dbkey to the incoming parameters incoming[ "dbkey" ] = input_dbkey # Keep track of parent / child relationships, we'll create all the # datasets first, then create the associations parent_to_child_pairs = [] child_dataset_names = set() for name, output in tool.outputs.items(): for filter in output.filters: try: if not eval( filter.text, globals(), incoming ): break #do not create this dataset except Exception, e: log.debug( 'Dataset output filter failed: %s' % e ) else: #all filters passed if output.parent: parent_to_child_pairs.append( ( output.parent, name ) ) child_dataset_names.add( name ) ## What is the following hack for? Need to document under what ## conditions can the following occur? ([email protected]) # HACK: the output data has already been created # this happens i.e. as a result of the async controller if name in incoming: dataid = incoming[name] data = trans.sa_session.query( trans.app.model.HistoryDatasetAssociation ).get( dataid ) assert data != None out_data[name] = data else: # the type should match the input ext = output.format if ext == "input": ext = input_ext #process change_format tags if output.change_format: for change_elem in output.change_format: for when_elem in change_elem.findall( 'when' ): check = incoming.get( when_elem.get( 'input' ), None ) if check is not None: if check == when_elem.get( 'value', None ): ext = when_elem.get( 'format', ext ) else: check = when_elem.get( 'input_dataset', None ) if check is not None: check = inp_data.get( check, None ) if check is not None: if str( getattr( check, when_elem.get( 'attribute' ) ) ) == when_elem.get( 'value', None ): ext = when_elem.get( 'format', ext ) data = trans.app.model.HistoryDatasetAssociation( extension=ext, create_dataset=True, sa_session=trans.sa_session ) # Commit the dataset immediately so it gets database assigned unique id trans.sa_session.add( data ) trans.sa_session.flush() trans.app.security_agent.set_all_dataset_permissions( data.dataset, output_permissions ) # Create an empty file immediately open( data.file_name, "w" ).close() # Fix permissions util.umask_fix_perms( data.file_name, trans.app.config.umask, 0666 ) # This may not be neccesary with the new parent/child associations data.designation = name # Copy metadata from one of the inputs if requested. if output.metadata_source: data.init_meta( copy_from=inp_data[output.metadata_source] ) else: data.init_meta() # Take dbkey from LAST input data.dbkey = str(input_dbkey) # Set state # FIXME: shouldn't this be NEW until the job runner changes it? data.state = data.states.QUEUED data.blurb = "queued" # Set output label if output.label: params = make_dict_copy( incoming ) # wrapping the params allows the tool config to contain things like # <outputs> # <data format="input" name="output" label="Blat on ${<input_param>.name}" /> # </outputs> wrap_values( tool.inputs, params ) params['tool'] = tool params['on_string'] = on_text data.name = fill_template( output.label, context=params ) else: data.name = tool.name if on_text: data.name += ( " on " + on_text ) # Store output out_data[ name ] = data # Store all changes to database trans.sa_session.flush()
def execute(self, tool, trans, incoming={}, return_job=False, set_output_hid=True, set_output_history=True, history=None, job_params=None, rerun_remap_job_id=None): """ Executes a tool, creating job and tool outputs, associating them, and submitting the job to the job queue. If history is not specified, use trans.history as destination for tool's output datasets. """ def make_dict_copy(from_dict): """ Makes a copy of input dictionary from_dict such that all values that are dictionaries result in creation of a new dictionary ( a sort of deepcopy ). We may need to handle other complex types ( e.g., lists, etc ), but not sure... Yes, we need to handle lists (and now are)... """ copy_from_dict = {} for key, value in from_dict.items(): if type(value).__name__ == 'dict': copy_from_dict[key] = make_dict_copy(value) elif isinstance(value, list): copy_from_dict[key] = make_list_copy(value) else: copy_from_dict[key] = value return copy_from_dict def make_list_copy(from_list): new_list = [] for value in from_list: if isinstance(value, dict): new_list.append(make_dict_copy(value)) elif isinstance(value, list): new_list.append(make_list_copy(value)) else: new_list.append(value) return new_list def wrap_values(inputs, input_values, skip_missing_values=False): # Wrap tool inputs as necessary for input in inputs.itervalues(): if input.name not in input_values and skip_missing_values: continue if isinstance(input, Repeat): for d in input_values[input.name]: wrap_values(input.inputs, d, skip_missing_values=skip_missing_values) elif isinstance(input, Conditional): values = input_values[input.name] current = values["__current_case__"] wrap_values(input.cases[current].inputs, values, skip_missing_values=skip_missing_values) elif isinstance(input, DataToolParameter) and input.multiple: input_values[ input.name ] = \ galaxy.tools.DatasetListWrapper( input_values[ input.name ], datatypes_registry = trans.app.datatypes_registry, tool = tool, name = input.name ) elif isinstance(input, DataToolParameter): input_values[ input.name ] = \ galaxy.tools.DatasetFilenameWrapper( input_values[ input.name ], datatypes_registry = trans.app.datatypes_registry, tool = tool, name = input.name ) elif isinstance(input, SelectToolParameter): input_values[ input.name] = galaxy.tools.SelectToolParameterWrapper( input, input_values[input.name], tool.app, other_values=incoming) else: input_values[input.name] = galaxy.tools.InputValueWrapper( input, input_values[input.name], incoming) # Set history. if not history: history = tool.get_default_history_by_trans(trans, create=True) out_data = odict() # Collect any input datasets from the incoming parameters inp_data = self.collect_input_datasets(tool, incoming, trans) # Deal with input dataset names, 'dbkey' and types input_names = [] input_ext = 'data' input_dbkey = incoming.get("dbkey", "?") for name, data in inp_data.items(): if not data: data = NoneDataset( datatypes_registry=trans.app.datatypes_registry) continue # Convert LDDA to an HDA. if isinstance(data, LibraryDatasetDatasetAssociation): data = data.to_history_dataset_association(None) inp_data[name] = data else: # HDA if data.hid: input_names.append('data %s' % data.hid) input_ext = data.ext if data.dbkey not in [None, '?']: input_dbkey = data.dbkey # Collect chromInfo dataset and add as parameters to incoming db_datasets = {} db_dataset = trans.db_dataset_for(input_dbkey) if db_dataset: db_datasets["chromInfo"] = db_dataset incoming["chromInfo"] = db_dataset.file_name else: # For custom builds, chrom info resides in converted dataset; for built-in builds, chrom info resides in tool-data/shared. chrom_info = None if trans.user and ('dbkeys' in trans.user.preferences) and ( input_dbkey in from_json_string( trans.user.preferences['dbkeys'])): # Custom build. custom_build_dict = from_json_string( trans.user.preferences['dbkeys'])[input_dbkey] if 'fasta' in custom_build_dict: build_fasta_dataset = trans.sa_session.query( trans.app.model.HistoryDatasetAssociation).get( custom_build_dict['fasta']) chrom_info = build_fasta_dataset.get_converted_dataset( trans, 'len').file_name if not chrom_info: # Default to built-in build. chrom_info = os.path.join(trans.app.config.len_file_path, "%s.len" % input_dbkey) incoming["chromInfo"] = chrom_info inp_data.update(db_datasets) # Determine output dataset permission/roles list existing_datasets = [inp for inp in inp_data.values() if inp] if existing_datasets: output_permissions = trans.app.security_agent.guess_derived_permissions_for_datasets( existing_datasets) else: # No valid inputs, we will use history defaults output_permissions = trans.app.security_agent.history_get_default_permissions( history) # Build name for output datasets based on tool name and input names if len(input_names) == 1: on_text = input_names[0] elif len(input_names) == 2: on_text = '%s and %s' % tuple(input_names[0:2]) elif len(input_names) == 3: on_text = '%s, %s, and %s' % tuple(input_names[0:3]) elif len(input_names) > 3: on_text = '%s, %s, and others' % tuple(input_names[0:2]) else: on_text = "" # Add the dbkey to the incoming parameters incoming["dbkey"] = input_dbkey params = None #wrapped params are used by change_format action and by output.label; only perform this wrapping once, as needed # Keep track of parent / child relationships, we'll create all the # datasets first, then create the associations parent_to_child_pairs = [] child_dataset_names = set() object_store_id = None for name, output in tool.outputs.items(): for filter in output.filters: try: if not eval(filter.text.strip(), globals(), incoming): break #do not create this dataset except Exception, e: log.debug('Dataset output filter failed: %s' % e) else: #all filters passed if output.parent: parent_to_child_pairs.append((output.parent, name)) child_dataset_names.add(name) ## What is the following hack for? Need to document under what ## conditions can the following occur? ([email protected]) # HACK: the output data has already been created # this happens i.e. as a result of the async controller if name in incoming: dataid = incoming[name] data = trans.sa_session.query( trans.app.model.HistoryDatasetAssociation).get(dataid) assert data != None out_data[name] = data else: # the type should match the input ext = output.format if ext == "input": ext = input_ext if output.format_source is not None and output.format_source in inp_data: try: input_dataset = inp_data[output.format_source] input_extension = input_dataset.ext ext = input_extension except Exception, e: pass #process change_format tags if output.change_format: if params is None: params = make_dict_copy(incoming) wrap_values( tool.inputs, params, skip_missing_values=not tool.check_values) for change_elem in output.change_format: for when_elem in change_elem.findall('when'): check = when_elem.get('input', None) if check is not None: try: if '$' not in check: #allow a simple name or more complex specifications check = '${%s}' % check if str( fill_template(check, context=params) ) == when_elem.get('value', None): ext = when_elem.get('format', ext) except: #bad tag input value; possibly referencing a param within a different conditional when block or other nonexistent grouping construct continue else: check = when_elem.get( 'input_dataset', None) if check is not None: check = inp_data.get(check, None) if check is not None: if str( getattr( check, when_elem.get( 'attribute')) ) == when_elem.get('value', None): ext = when_elem.get( 'format', ext) data = trans.app.model.HistoryDatasetAssociation( extension=ext, create_dataset=True, sa_session=trans.sa_session) if output.hidden: data.visible = False # Commit the dataset immediately so it gets database assigned unique id trans.sa_session.add(data) trans.sa_session.flush() trans.app.security_agent.set_all_dataset_permissions( data.dataset, output_permissions) # Create an empty file immediately. The first dataset will be # created in the "default" store, all others will be created in # the same store as the first. data.dataset.object_store_id = object_store_id try: trans.app.object_store.create(data.dataset) except ObjectInvalid: raise Exception( 'Unable to create output dataset: object store is full' ) object_store_id = data.dataset.object_store_id # these will be the same thing after the first output # This may not be neccesary with the new parent/child associations data.designation = name # Copy metadata from one of the inputs if requested. if output.metadata_source: data.init_meta(copy_from=inp_data[output.metadata_source]) else: data.init_meta() # Take dbkey from LAST input data.dbkey = str(input_dbkey) # Set state # FIXME: shouldn't this be NEW until the job runner changes it? data.state = data.states.QUEUED data.blurb = "queued" # Set output label if output.label: if params is None: params = make_dict_copy(incoming) # wrapping the params allows the tool config to contain things like # <outputs> # <data format="input" name="output" label="Blat on ${<input_param>.name}" /> # </outputs> wrap_values(tool.inputs, params, skip_missing_values=not tool.check_values) #tool (only needing to be set once) and on_string (set differently for each label) are overwritten for each output dataset label being determined params['tool'] = tool params['on_string'] = on_text data.name = fill_template(output.label, context=params) else: if params is None: params = make_dict_copy(incoming) wrap_values(tool.inputs, params, skip_missing_values=not tool.check_values) data.name = self._get_default_data_name( data, tool, on_text=on_text, trans=trans, incoming=incoming, history=history, params=params, job_params=job_params) # Store output out_data[name] = data if output.actions: #Apply pre-job tool-output-dataset actions; e.g. setting metadata, changing format output_action_params = dict(out_data) output_action_params.update(incoming) output.actions.apply_action(data, output_action_params) # Store all changes to database trans.sa_session.flush()
def execute(self, tool, trans, incoming={}, return_job=False, set_output_hid=True, history=None, job_params=None, rerun_remap_job_id=None, mapping_over_collection=False, execution_cache=None ): """ Executes a tool, creating job and tool outputs, associating them, and submitting the job to the job queue. If history is not specified, use trans.history as destination for tool's output datasets. """ self._check_access( tool, trans ) app = trans.app if execution_cache is None: execution_cache = ToolExecutionCache(trans) current_user_roles = execution_cache.current_user_roles history, inp_data, inp_dataset_collections = self._collect_inputs(tool, trans, incoming, history, current_user_roles) # Build name for output datasets based on tool name and input names on_text = self._get_on_text( inp_data ) # format='input" previously would give you a random extension from # the input extensions, now it should just give "input" as the output # format. input_ext = 'data' if tool.profile < 16.04 else "input" input_dbkey = incoming.get( "dbkey", "?" ) preserved_tags = [] for name, data in reversed(inp_data.items()): if not data: data = NoneDataset( datatypes_registry=app.datatypes_registry ) continue # Convert LDDA to an HDA. if isinstance(data, LibraryDatasetDatasetAssociation): data = data.to_history_dataset_association( None ) inp_data[name] = data if tool.profile < 16.04: input_ext = data.ext if data.dbkey not in [None, '?']: input_dbkey = data.dbkey identifier = getattr( data, "element_identifier", None ) if identifier is not None: incoming[ "%s|__identifier__" % name ] = identifier for tag in data.tags: if tag.user_tname == 'name': preserved_tags.append(tag) # Collect chromInfo dataset and add as parameters to incoming ( chrom_info, db_dataset ) = app.genome_builds.get_chrom_info( input_dbkey, trans=trans, custom_build_hack_get_len_from_fasta_conversion=tool.id != 'CONVERTER_fasta_to_len' ) if db_dataset: inp_data.update( { "chromInfo": db_dataset } ) incoming[ "chromInfo" ] = chrom_info # Determine output dataset permission/roles list existing_datasets = [ inp for inp in inp_data.values() if inp ] if existing_datasets: output_permissions = app.security_agent.guess_derived_permissions_for_datasets( existing_datasets ) else: # No valid inputs, we will use history defaults output_permissions = app.security_agent.history_get_default_permissions( history ) # Add the dbkey to the incoming parameters incoming[ "dbkey" ] = input_dbkey # wrapped params are used by change_format action and by output.label; only perform this wrapping once, as needed wrapped_params = self._wrapped_params( trans, tool, incoming ) out_data = odict() input_collections = dict( (k, v[0][0]) for k, v in inp_dataset_collections.items() ) output_collections = OutputCollections( trans, history, tool=tool, tool_action=self, input_collections=input_collections, mapping_over_collection=mapping_over_collection, on_text=on_text, incoming=incoming, params=wrapped_params.params, job_params=job_params, ) # Keep track of parent / child relationships, we'll create all the # datasets first, then create the associations parent_to_child_pairs = [] child_dataset_names = set() object_store_populator = ObjectStorePopulator( app ) def handle_output( name, output, hidden=None ): if output.parent: parent_to_child_pairs.append( ( output.parent, name ) ) child_dataset_names.add( name ) # What is the following hack for? Need to document under what # conditions can the following occur? ([email protected]) # HACK: the output data has already been created # this happens i.e. as a result of the async controller if name in incoming: dataid = incoming[name] data = trans.sa_session.query( app.model.HistoryDatasetAssociation ).get( dataid ) assert data is not None out_data[name] = data else: ext = determine_output_format( output, wrapped_params.params, inp_data, inp_dataset_collections, input_ext ) data = app.model.HistoryDatasetAssociation( extension=ext, create_dataset=True, flush=False ) if hidden is None: hidden = output.hidden if hidden: data.visible = False trans.sa_session.add( data ) trans.app.security_agent.set_all_dataset_permissions( data.dataset, output_permissions, new=True ) for tag in preserved_tags: data.tags.append(tag.copy()) # Must flush before setting object store id currently. # TODO: optimize this. trans.sa_session.flush() object_store_populator.set_object_store_id( data ) # This may not be neccesary with the new parent/child associations data.designation = name # Copy metadata from one of the inputs if requested. # metadata source can be either a string referencing an input # or an actual object to copy. metadata_source = output.metadata_source if metadata_source: if isinstance( metadata_source, string_types ): metadata_source = inp_data.get( metadata_source ) if metadata_source is not None: data.init_meta( copy_from=metadata_source ) else: data.init_meta() # Take dbkey from LAST input data.dbkey = str(input_dbkey) # Set state data.blurb = "queued" # Set output label data.name = self.get_output_name( output, data, tool, on_text, trans, incoming, history, wrapped_params.params, job_params ) # Store output out_data[ name ] = data if output.actions: # Apply pre-job tool-output-dataset actions; e.g. setting metadata, changing format output_action_params = dict( out_data ) output_action_params.update( incoming ) output.actions.apply_action( data, output_action_params ) # Also set the default values of actions of type metadata self.set_metadata_defaults( output, data, tool, on_text, trans, incoming, history, wrapped_params.params, job_params ) # Flush all datasets at once. return data for name, output in tool.outputs.items(): if not filter_output(output, incoming): if output.collection: collections_manager = app.dataset_collections_service element_identifiers = [] known_outputs = output.known_outputs( input_collections, collections_manager.type_registry ) # Just to echo TODO elsewhere - this should be restructured to allow # nested collections. for output_part_def in known_outputs: # Add elements to top-level collection, unless nested... current_element_identifiers = element_identifiers current_collection_type = output.structure.collection_type for parent_id in (output_part_def.parent_ids or []): # TODO: replace following line with formal abstractions for doing this. current_collection_type = ":".join(current_collection_type.split(":")[1:]) name_to_index = dict((value["name"], index) for (index, value) in enumerate(current_element_identifiers)) if parent_id not in name_to_index: if parent_id not in current_element_identifiers: index = len(current_element_identifiers) current_element_identifiers.append(dict( name=parent_id, collection_type=current_collection_type, src="new_collection", element_identifiers=[], )) else: index = name_to_index[parent_id] current_element_identifiers = current_element_identifiers[ index ][ "element_identifiers" ] effective_output_name = output_part_def.effective_output_name element = handle_output( effective_output_name, output_part_def.output_def, hidden=True ) # TODO: this shouldn't exist in the top-level of the history at all # but for now we are still working around that by hiding the contents # there. # Following hack causes dataset to no be added to history... child_dataset_names.add( effective_output_name ) history.add_dataset( element, set_hid=set_output_hid, quota=False ) trans.sa_session.add( element ) trans.sa_session.flush() current_element_identifiers.append({ "__object__": element, "name": output_part_def.element_identifier, }) log.info(element_identifiers) if output.dynamic_structure: assert not element_identifiers # known_outputs must have been empty element_kwds = dict(elements=collections_manager.ELEMENTS_UNINITIALIZED) else: element_kwds = dict(element_identifiers=element_identifiers) output_collections.create_collection( output=output, name=name, tags=preserved_tags, **element_kwds ) else: handle_output_timer = ExecutionTimer() handle_output( name, output ) log.info("Handled output named %s for tool %s %s" % (name, tool.id, handle_output_timer)) add_datasets_timer = ExecutionTimer() # Add all the top-level (non-child) datasets to the history unless otherwise specified datasets_to_persist = [] for name in out_data.keys(): if name not in child_dataset_names and name not in incoming: # don't add children; or already existing datasets, i.e. async created data = out_data[ name ] datasets_to_persist.append( data ) # Set HID and add to history. # This is brand new and certainly empty so don't worry about quota. # TOOL OPTIMIZATION NOTE - from above loop to the job create below 99%+ # of execution time happens within in history.add_datasets. history.add_datasets( trans.sa_session, datasets_to_persist, set_hid=set_output_hid, quota=False, flush=False ) # Add all the children to their parents for parent_name, child_name in parent_to_child_pairs: parent_dataset = out_data[ parent_name ] child_dataset = out_data[ child_name ] parent_dataset.children.append( child_dataset ) log.info("Added output datasets to history %s" % add_datasets_timer) job_setup_timer = ExecutionTimer() # Create the job object job, galaxy_session = self._new_job_for_session( trans, tool, history ) self._record_inputs( trans, tool, job, incoming, inp_data, inp_dataset_collections, current_user_roles ) self._record_outputs( job, out_data, output_collections ) job.object_store_id = object_store_populator.object_store_id if job_params: job.params = dumps( job_params ) job.set_handler(tool.get_job_handler(job_params)) trans.sa_session.add( job ) # Now that we have a job id, we can remap any outputs if this is a rerun and the user chose to continue dependent jobs # This functionality requires tracking jobs in the database. if app.config.track_jobs_in_database and rerun_remap_job_id is not None: try: old_job = trans.sa_session.query( app.model.Job ).get(rerun_remap_job_id) assert old_job is not None, '(%s/%s): Old job id is invalid' % (rerun_remap_job_id, job.id) assert old_job.tool_id == job.tool_id, '(%s/%s): Old tool id (%s) does not match rerun tool id (%s)' % (old_job.id, job.id, old_job.tool_id, job.tool_id) if trans.user is not None: assert old_job.user_id == trans.user.id, '(%s/%s): Old user id (%s) does not match rerun user id (%s)' % (old_job.id, job.id, old_job.user_id, trans.user.id) elif trans.user is None and type( galaxy_session ) == trans.model.GalaxySession: assert old_job.session_id == galaxy_session.id, '(%s/%s): Old session id (%s) does not match rerun session id (%s)' % (old_job.id, job.id, old_job.session_id, galaxy_session.id) else: raise Exception('(%s/%s): Remapping via the API is not (yet) supported' % (old_job.id, job.id)) # Duplicate PJAs before remap. for pjaa in old_job.post_job_actions: job.add_post_job_action(pjaa.post_job_action) for jtod in old_job.output_datasets: for (job_to_remap, jtid) in [(jtid.job, jtid) for jtid in jtod.dataset.dependent_jobs]: if (trans.user is not None and job_to_remap.user_id == trans.user.id) or (trans.user is None and job_to_remap.session_id == galaxy_session.id): if job_to_remap.state == job_to_remap.states.PAUSED: job_to_remap.state = job_to_remap.states.NEW for hda in [ dep_jtod.dataset for dep_jtod in job_to_remap.output_datasets ]: if hda.state == hda.states.PAUSED: hda.state = hda.states.NEW hda.info = None input_values = dict( [ ( p.name, json.loads( p.value ) ) for p in job_to_remap.parameters ] ) update_param( jtid.name, input_values, str( out_data[ jtod.name ].id ) ) for p in job_to_remap.parameters: p.value = json.dumps( input_values[ p.name ] ) jtid.dataset = out_data[jtod.name] jtid.dataset.hid = jtod.dataset.hid log.info('Job %s input HDA %s remapped to new HDA %s' % (job_to_remap.id, jtod.dataset.id, jtid.dataset.id)) trans.sa_session.add(job_to_remap) trans.sa_session.add(jtid) jtod.dataset.visible = False trans.sa_session.add(jtod) except Exception: log.exception('Cannot remap rerun dependencies.') log.info("Setup for job %s complete, ready to flush %s" % (job.log_str(), job_setup_timer)) job_flush_timer = ExecutionTimer() trans.sa_session.flush() log.info("Flushed transaction for job %s %s" % (job.log_str(), job_flush_timer)) # Some tools are not really executable, but jobs are still created for them ( for record keeping ). # Examples include tools that redirect to other applications ( epigraph ). These special tools must # include something that can be retrieved from the params ( e.g., REDIRECT_URL ) to keep the job # from being queued. if 'REDIRECT_URL' in incoming: # Get the dataset - there should only be 1 for name in inp_data.keys(): dataset = inp_data[ name ] redirect_url = tool.parse_redirect_url( dataset, incoming ) # GALAXY_URL should be include in the tool params to enable the external application # to send back to the current Galaxy instance GALAXY_URL = incoming.get( 'GALAXY_URL', None ) assert GALAXY_URL is not None, "GALAXY_URL parameter missing in tool config." redirect_url += "&GALAXY_URL=%s" % GALAXY_URL # Job should not be queued, so set state to ok job.set_state( app.model.Job.states.OK ) job.info = "Redirected to: %s" % redirect_url trans.sa_session.add( job ) trans.sa_session.flush() trans.response.send_redirect( url_for( controller='tool_runner', action='redirect', redirect_url=redirect_url ) ) else: # Put the job in the queue if tracking in memory app.job_queue.put( job.id, job.tool_id ) trans.log_event( "Added job to the job queue, id: %s" % str(job.id), tool_id=job.tool_id ) return job, out_data
def execute(self, tool, trans, incoming={}, return_job=False, set_output_hid=True, set_output_history=True, history=None, job_params=None, rerun_remap_job_id=None, mapping_over_collection=False): """ Executes a tool, creating job and tool outputs, associating them, and submitting the job to the job queue. If history is not specified, use trans.history as destination for tool's output datasets. """ assert tool.allow_user_access( trans.user ), "User (%s) is not allowed to access this tool." % ( trans.user ) # Set history. if not history: history = tool.get_default_history_by_trans( trans, create=True ) out_data = odict() out_collections = {} out_collection_instances = {} # Track input dataset collections - but replace with simply lists so collect # input datasets can process these normally. inp_dataset_collections = self.collect_input_dataset_collections( tool, incoming ) # Collect any input datasets from the incoming parameters inp_data = self.collect_input_datasets( tool, incoming, trans ) # Deal with input dataset names, 'dbkey' and types input_names = [] input_ext = 'data' input_dbkey = incoming.get( "dbkey", "?" ) inp_items = inp_data.items() inp_items.reverse() for name, data in inp_items: if not data: data = NoneDataset( datatypes_registry=trans.app.datatypes_registry ) continue # Convert LDDA to an HDA. if isinstance(data, LibraryDatasetDatasetAssociation): data = data.to_history_dataset_association( None ) inp_data[name] = data else: # HDA if data.hid: input_names.append( 'data %s' % data.hid ) input_ext = data.ext if data.dbkey not in [None, '?']: input_dbkey = data.dbkey identifier = getattr( data, "element_identifier", None ) if identifier is not None: incoming[ "%s|__identifier__" % name ] = identifier # Collect chromInfo dataset and add as parameters to incoming ( chrom_info, db_dataset ) = trans.app.genome_builds.get_chrom_info( input_dbkey, trans=trans, custom_build_hack_get_len_from_fasta_conversion=tool.id != 'CONVERTER_fasta_to_len' ) if db_dataset: inp_data.update( { "chromInfo": db_dataset } ) incoming[ "chromInfo" ] = chrom_info # Determine output dataset permission/roles list existing_datasets = [ inp for inp in inp_data.values() if inp ] if existing_datasets: output_permissions = trans.app.security_agent.guess_derived_permissions_for_datasets( existing_datasets ) else: # No valid inputs, we will use history defaults output_permissions = trans.app.security_agent.history_get_default_permissions( history ) # Build name for output datasets based on tool name and input names on_text = on_text_for_names( input_names ) # Add the dbkey to the incoming parameters incoming[ "dbkey" ] = input_dbkey # wrapped params are used by change_format action and by output.label; only perform this wrapping once, as needed wrapped_params = WrappedParameters( trans, tool, incoming ) # Keep track of parent / child relationships, we'll create all the # datasets first, then create the associations parent_to_child_pairs = [] child_dataset_names = set() object_store_populator = ObjectStorePopulator( trans.app ) def handle_output( name, output ): if output.parent: parent_to_child_pairs.append( ( output.parent, name ) ) child_dataset_names.add( name ) # What is the following hack for? Need to document under what # conditions can the following occur? ([email protected]) # HACK: the output data has already been created # this happens i.e. as a result of the async controller if name in incoming: dataid = incoming[name] data = trans.sa_session.query( trans.app.model.HistoryDatasetAssociation ).get( dataid ) assert data is not None out_data[name] = data else: ext = determine_output_format( output, wrapped_params.params, inp_data, input_ext ) data = trans.app.model.HistoryDatasetAssociation( extension=ext, create_dataset=True, sa_session=trans.sa_session ) if output.hidden: data.visible = False # Commit the dataset immediately so it gets database assigned unique id trans.sa_session.add( data ) trans.sa_session.flush() trans.app.security_agent.set_all_dataset_permissions( data.dataset, output_permissions ) object_store_populator.set_object_store_id( data ) # This may not be neccesary with the new parent/child associations data.designation = name # Copy metadata from one of the inputs if requested. # metadata source can be either a string referencing an input # or an actual object to copy. metadata_source = output.metadata_source if metadata_source: if isinstance( metadata_source, basestring ): metadata_source = inp_data[metadata_source] if metadata_source is not None: data.init_meta( copy_from=metadata_source ) else: data.init_meta() # Take dbkey from LAST input data.dbkey = str(input_dbkey) # Set state # FIXME: shouldn't this be NEW until the job runner changes it? data.state = data.states.QUEUED data.blurb = "queued" # Set output label data.name = self.get_output_name( output, data, tool, on_text, trans, incoming, history, wrapped_params.params, job_params ) # Store output out_data[ name ] = data if output.actions: # Apply pre-job tool-output-dataset actions; e.g. setting metadata, changing format output_action_params = dict( out_data ) output_action_params.update( incoming ) output.actions.apply_action( data, output_action_params ) # Store all changes to database trans.sa_session.flush() return data for name, output in tool.outputs.items(): if not filter_output(output, incoming): if output.collection: collections_manager = trans.app.dataset_collections_service # As far as I can tell - this is always true - but just verify assert set_output_history, "Cannot create dataset collection for this kind of tool." element_identifiers = [] input_collections = dict( [ (k, v[0]) for k, v in inp_dataset_collections.iteritems() ] ) known_outputs = output.known_outputs( input_collections, collections_manager.type_registry ) # Just to echo TODO elsewhere - this should be restructured to allow # nested collections. for output_part_def in known_outputs: # Add elements to top-level collection, unless nested... current_element_identifiers = element_identifiers current_collection_type = output.structure.collection_type for parent_id in (output_part_def.parent_ids or []): # TODO: replace following line with formal abstractions for doing this. current_collection_type = ":".join(current_collection_type.split(":")[1:]) name_to_index = dict(map(lambda (index, value): (value["name"], index), enumerate(current_element_identifiers))) if parent_id not in name_to_index: if parent_id not in current_element_identifiers: index = len(current_element_identifiers) current_element_identifiers.append(dict( name=parent_id, collection_type=current_collection_type, src="new_collection", element_identifiers=[], )) else: index = name_to_index[parent_id] current_element_identifiers = current_element_identifiers[ index ][ "element_identifiers" ] effective_output_name = output_part_def.effective_output_name element = handle_output( effective_output_name, output_part_def.output_def ) # Following hack causes dataset to no be added to history... child_dataset_names.add( effective_output_name ) if set_output_history: history.add_dataset( element, set_hid=set_output_hid ) trans.sa_session.add( element ) trans.sa_session.flush() current_element_identifiers.append({ "__object__": element, "name": output_part_def.element_identifier, }) log.info(element_identifiers) if output.dynamic_structure: assert not element_identifiers # known_outputs must have been empty element_kwds = dict(elements=collections_manager.ELEMENTS_UNINITIALIZED) else: element_kwds = dict(element_identifiers=element_identifiers) if mapping_over_collection: dc = collections_manager.create_dataset_collection( trans, collection_type=output.structure.collection_type, **element_kwds ) out_collections[ name ] = dc else: hdca_name = self.get_output_name( output, None, tool, on_text, trans, incoming, history, wrapped_params.params, job_params ) hdca = collections_manager.create( trans, history, name=hdca_name, collection_type=output.structure.collection_type, trusted_identifiers=True, **element_kwds ) # name here is name of the output element - not name # of the hdca. out_collection_instances[ name ] = hdca else: handle_output_timer = ExecutionTimer() handle_output( name, output ) log.info("Handled output %s" % handle_output_timer) # Add all the top-level (non-child) datasets to the history unless otherwise specified for name in out_data.keys(): if name not in child_dataset_names and name not in incoming: # don't add children; or already existing datasets, i.e. async created data = out_data[ name ] if set_output_history: history.add_dataset( data, set_hid=set_output_hid ) trans.sa_session.add( data ) trans.sa_session.flush() # Add all the children to their parents for parent_name, child_name in parent_to_child_pairs: parent_dataset = out_data[ parent_name ] child_dataset = out_data[ child_name ] parent_dataset.children.append( child_dataset ) # Store data after custom code runs trans.sa_session.flush() # Create the job object job = trans.app.model.Job() if hasattr( trans, "get_galaxy_session" ): galaxy_session = trans.get_galaxy_session() # If we're submitting from the API, there won't be a session. if type( galaxy_session ) == trans.model.GalaxySession: job.session_id = galaxy_session.id if trans.user is not None: job.user_id = trans.user.id job.history_id = history.id job.tool_id = tool.id try: # For backward compatibility, some tools may not have versions yet. job.tool_version = tool.version except: job.tool_version = "1.0.0" # FIXME: Don't need all of incoming here, just the defined parameters # from the tool. We need to deal with tools that pass all post # parameters to the command as a special case. for name, ( dataset_collection, reduced ) in inp_dataset_collections.iteritems(): # TODO: Does this work if nested in repeat/conditional? if reduced: incoming[ name ] = "__collection_reduce__|%s" % dataset_collection.id # Should verify security? We check security of individual # datasets below? job.add_input_dataset_collection( name, dataset_collection ) for name, value in tool.params_to_strings( incoming, trans.app ).iteritems(): job.add_parameter( name, value ) current_user_roles = trans.get_current_user_roles() access_timer = ExecutionTimer() for name, dataset in inp_data.iteritems(): if dataset: if not trans.app.security_agent.can_access_dataset( current_user_roles, dataset.dataset ): raise Exception("User does not have permission to use a dataset (%s) provided for input." % data.id) job.add_input_dataset( name, dataset ) else: job.add_input_dataset( name, None ) log.info("Verified access to datasets %s" % access_timer) for name, dataset in out_data.iteritems(): job.add_output_dataset( name, dataset ) for name, dataset_collection in out_collections.iteritems(): job.add_implicit_output_dataset_collection( name, dataset_collection ) for name, dataset_collection_instance in out_collection_instances.iteritems(): job.add_output_dataset_collection( name, dataset_collection_instance ) job.object_store_id = object_store_populator.object_store_id if job_params: job.params = dumps( job_params ) job.set_handler(tool.get_job_handler(job_params)) trans.sa_session.add( job ) # Now that we have a job id, we can remap any outputs if this is a rerun and the user chose to continue dependent jobs # This functionality requires tracking jobs in the database. if trans.app.config.track_jobs_in_database and rerun_remap_job_id is not None: try: old_job = trans.sa_session.query( trans.app.model.Job ).get(rerun_remap_job_id) assert old_job is not None, '(%s/%s): Old job id is invalid' % (rerun_remap_job_id, job.id) assert old_job.tool_id == job.tool_id, '(%s/%s): Old tool id (%s) does not match rerun tool id (%s)' % (old_job.id, job.id, old_job.tool_id, job.tool_id) if trans.user is not None: assert old_job.user_id == trans.user.id, '(%s/%s): Old user id (%s) does not match rerun user id (%s)' % (old_job.id, job.id, old_job.user_id, trans.user.id) elif trans.user is None and type( galaxy_session ) == trans.model.GalaxySession: assert old_job.session_id == galaxy_session.id, '(%s/%s): Old session id (%s) does not match rerun session id (%s)' % (old_job.id, job.id, old_job.session_id, galaxy_session.id) else: raise Exception('(%s/%s): Remapping via the API is not (yet) supported' % (old_job.id, job.id)) for jtod in old_job.output_datasets: for (job_to_remap, jtid) in [(jtid.job, jtid) for jtid in jtod.dataset.dependent_jobs]: if (trans.user is not None and job_to_remap.user_id == trans.user.id) or (trans.user is None and job_to_remap.session_id == galaxy_session.id): if job_to_remap.state == job_to_remap.states.PAUSED: job_to_remap.state = job_to_remap.states.NEW for hda in [ dep_jtod.dataset for dep_jtod in job_to_remap.output_datasets ]: if hda.state == hda.states.PAUSED: hda.state = hda.states.NEW hda.info = None for p in job_to_remap.parameters: if p.name == jtid.name and p.value == str(jtod.dataset.id): p.value = str(out_data[jtod.name].id) jtid.dataset = out_data[jtod.name] jtid.dataset.hid = jtod.dataset.hid log.info('Job %s input HDA %s remapped to new HDA %s' % (job_to_remap.id, jtod.dataset.id, jtid.dataset.id)) trans.sa_session.add(job_to_remap) trans.sa_session.add(jtid) jtod.dataset.visible = False trans.sa_session.add(jtod) except Exception: log.exception('Cannot remap rerun dependencies.') trans.sa_session.flush() # Some tools are not really executable, but jobs are still created for them ( for record keeping ). # Examples include tools that redirect to other applications ( epigraph ). These special tools must # include something that can be retrieved from the params ( e.g., REDIRECT_URL ) to keep the job # from being queued. if 'REDIRECT_URL' in incoming: # Get the dataset - there should only be 1 for name in inp_data.keys(): dataset = inp_data[ name ] redirect_url = tool.parse_redirect_url( dataset, incoming ) # GALAXY_URL should be include in the tool params to enable the external application # to send back to the current Galaxy instance GALAXY_URL = incoming.get( 'GALAXY_URL', None ) assert GALAXY_URL is not None, "GALAXY_URL parameter missing in tool config." redirect_url += "&GALAXY_URL=%s" % GALAXY_URL # Job should not be queued, so set state to ok job.set_state( trans.app.model.Job.states.OK ) job.info = "Redirected to: %s" % redirect_url trans.sa_session.add( job ) trans.sa_session.flush() trans.response.send_redirect( url_for( controller='tool_runner', action='redirect', redirect_url=redirect_url ) ) else: # Put the job in the queue if tracking in memory trans.app.job_queue.put( job.id, job.tool_id ) trans.log_event( "Added job to the job queue, id: %s" % str(job.id), tool_id=job.tool_id ) return job, out_data
def execute(self, tool, trans, incoming={}, set_output_hid=True): def make_dict_copy(from_dict): """ Makes a copy of input dictionary from_dict such that all values that are dictionaries result in creation of a new dictionary ( a sort of deepcopy ). We may need to handle other complex types ( e.g., lists, etc ), but not sure... """ copy_from_dict = {} for key, value in from_dict.items(): if type(value).__name__ == 'dict': copy_from_dict[key] = make_dict_copy(value) else: copy_from_dict[key] = value return copy_from_dict def wrap_values(inputs, input_values): # Wrap tool inputs as necessary for input in inputs.itervalues(): if isinstance(input, Repeat): for d in input_values[input.name]: wrap_values(input.inputs, d) elif isinstance(input, Conditional): values = input_values[input.name] current = values["__current_case__"] wrap_values(input.cases[current].inputs, values) elif isinstance(input, DataToolParameter): input_values[ input.name ] = \ galaxy.tools.DatasetFilenameWrapper( input_values[ input.name ], datatypes_registry = trans.app.datatypes_registry, tool = tool, name = input.name ) elif isinstance(input, SelectToolParameter): input_values[ input.name] = galaxy.tools.SelectToolParameterWrapper( input, input_values[input.name], tool.app, other_values=incoming) else: input_values[input.name] = galaxy.tools.InputValueWrapper( input, input_values[input.name], incoming) out_data = {} # Collect any input datasets from the incoming parameters inp_data = self.collect_input_datasets(tool, incoming, trans) # Deal with input dataset names, 'dbkey' and types input_names = [] input_ext = 'data' input_dbkey = incoming.get("dbkey", "?") for name, data in inp_data.items(): if data: input_names.append('data %s' % data.hid) input_ext = data.ext else: data = NoneDataset( datatypes_registry=trans.app.datatypes_registry) if data.dbkey not in [None, '?']: input_dbkey = data.dbkey # Collect chromInfo dataset and add as parameters to incoming db_datasets = {} db_dataset = trans.db_dataset_for(input_dbkey) if db_dataset: db_datasets["chromInfo"] = db_dataset incoming["chromInfo"] = db_dataset.file_name else: incoming["chromInfo"] = os.path.join( trans.app.config.tool_data_path, 'shared', 'ucsc', 'chrom', "%s.len" % input_dbkey) inp_data.update(db_datasets) # Determine output dataset permission/roles list existing_datasets = [inp for inp in inp_data.values() if inp] if existing_datasets: output_permissions = trans.app.security_agent.guess_derived_permissions_for_datasets( existing_datasets) else: # No valid inputs, we will use history defaults output_permissions = trans.app.security_agent.history_get_default_permissions( trans.history) # Build name for output datasets based on tool name and input names if len(input_names) == 1: on_text = input_names[0] elif len(input_names) == 2: on_text = '%s and %s' % tuple(input_names[0:2]) elif len(input_names) == 3: on_text = '%s, %s, and %s' % tuple(input_names[0:3]) elif len(input_names) > 3: on_text = '%s, %s, and others' % tuple(input_names[0:2]) else: on_text = "" # Add the dbkey to the incoming parameters incoming["dbkey"] = input_dbkey # Keep track of parent / child relationships, we'll create all the # datasets first, then create the associations parent_to_child_pairs = [] child_dataset_names = set() for name, output in tool.outputs.items(): for filter in output.filters: try: if not eval(filter.text, globals(), incoming): break #do not create this dataset except Exception, e: log.debug('Dataset output filter failed: %s' % e) else: #all filters passed if output.parent: parent_to_child_pairs.append((output.parent, name)) child_dataset_names.add(name) ## What is the following hack for? Need to document under what ## conditions can the following occur? ([email protected]) # HACK: the output data has already been created # this happens i.e. as a result of the async controller if name in incoming: dataid = incoming[name] data = trans.sa_session.query( trans.app.model.HistoryDatasetAssociation).get(dataid) assert data != None out_data[name] = data else: # the type should match the input ext = output.format if ext == "input": ext = input_ext #process change_format tags if output.change_format: for change_elem in output.change_format: for when_elem in change_elem.findall('when'): check = incoming.get(when_elem.get('input'), None) if check is not None: if check == when_elem.get('value', None): ext = when_elem.get('format', ext) else: check = when_elem.get( 'input_dataset', None) if check is not None: check = inp_data.get(check, None) if check is not None: if str( getattr( check, when_elem.get( 'attribute')) ) == when_elem.get('value', None): ext = when_elem.get( 'format', ext) data = trans.app.model.HistoryDatasetAssociation( extension=ext, create_dataset=True, sa_session=trans.sa_session) # Commit the dataset immediately so it gets database assigned unique id trans.sa_session.add(data) trans.sa_session.flush() trans.app.security_agent.set_all_dataset_permissions( data.dataset, output_permissions) # Create an empty file immediately open(data.file_name, "w").close() # Fix permissions util.umask_fix_perms(data.file_name, trans.app.config.umask, 0666) # This may not be neccesary with the new parent/child associations data.designation = name # Copy metadata from one of the inputs if requested. if output.metadata_source: data.init_meta(copy_from=inp_data[output.metadata_source]) else: data.init_meta() # Take dbkey from LAST input data.dbkey = str(input_dbkey) # Set state # FIXME: shouldn't this be NEW until the job runner changes it? data.state = data.states.QUEUED data.blurb = "queued" # Set output label if output.label: params = make_dict_copy(incoming) # wrapping the params allows the tool config to contain things like # <outputs> # <data format="input" name="output" label="Blat on ${<input_param>.name}" /> # </outputs> wrap_values(tool.inputs, params) params['tool'] = tool params['on_string'] = on_text data.name = fill_template(output.label, context=params) else: data.name = tool.name if on_text: data.name += (" on " + on_text) # Store output out_data[name] = data # Store all changes to database trans.sa_session.flush()