Merge pull request #504 from ExaWorks/ci_add_multi_node_queue

Adds a `multi_node_queue_name` ci config option

Author: hategan

testing.conf

@@ -115,6 +115,14 @@ executors = auto
 queue_name =
 
 
+#
+# If a different queue is used for multi-node jobs, specify it here. If empty,
+# queue_name is used for both single-node and multi-node jobs.
+#
+
+multi_node_queue_name =
+
+
 #
 # If you need a project/account for billing purposes, enter it here.
 #

tests/_test_tools.py

@@ -8,7 +8,7 @@
 
 from executor_test_params import ExecutorTestParams
 
-from psij import JobStatus, JobState, Job, JobExecutor, JobAttributes
+from psij import JobStatus, JobState, Job, JobExecutor, JobAttributes, ResourceSpecV1
 
 _QUICK_EXECUTORS = set(['local', 'batch-test'])
 
@@ -51,11 +51,16 @@ def _get_executor_instance(ep: ExecutorTestParams, job: Optional[Job] = None) ->
     if job is not None:
         assert job.spec is not None
         job.spec.launcher = ep.launcher
-        job.spec.attributes = JobAttributes(custom_attributes=ep.custom_attributes)
+        attrs = JobAttributes(custom_attributes=ep.custom_attributes)
+        job.spec.attributes = attrs
         if ep.account is not None:
-            job.spec.attributes.account = ep.account
+            attrs.account = ep.account
+        res = job.spec.resources
         if ep.queue_name is not None:
-            job.spec.attributes.queue_name = ep.queue_name
+            attrs.queue_name = ep.queue_name
+        if (res and isinstance(res, ResourceSpecV1) and res.computed_node_count > 1
+                and ep.multi_node_queue_name is not None):
+            attrs.queue_name = ep.multi_node_queue_name
     return JobExecutor.get_instance(ep.executor, url=ep.url)
 
 

tests/ci_runner.py

@@ -147,11 +147,13 @@ def run_branch_tests(conf: Dict[str, str], dir: Path, run_id: str, clone: bool =
         args.append('--branch-name-override')
         args.append(fake_branch_name)
     for opt in ['maintainer_email', 'executors', 'server_url', 'key', 'max_age',
-                'custom_attributes', 'queue_name', 'project_name', 'account']:
+                'custom_attributes', 'queue_name', 'multi_node_queue_name',
+                'project_name', 'account']:
         try:
             val = get_conf(conf, opt)
-            args.append('--' + opt.replace('_', '-'))
-            args.append(val)
+            if val != '':
+                args.append('--' + opt.replace('_', '-'))
+                args.append(val)
         except KeyError:
             # sometimes options get added; when they do, they could prevent
             # old test cycles from working, if their configs don't contain

tests/executor_test_params.py

@@ -6,6 +6,7 @@ class ExecutorTestParams:
     """A class holding executor, launcher, url pairs."""
 
     def __init__(self, spec: str, queue_name: Optional[str] = None,
+                 multi_node_queue_name: Optional[str] = None,
                  account: Optional[str] = None,
                  custom_attributes_raw: Optional[Dict[str, Dict[str, object]]] = None) \
             -> None:
@@ -19,6 +20,9 @@ def __init__(self, spec: str, queue_name: Optional[str] = None,
             url are specified, the string should be formatted as "<executor>::<url>".
         queue_name
             An optional queue to submit the job to
+        multi_node_queue_name
+            An optional multi-node queue name used for multi-node jobs. If not specified,
+            `queue_name` is used for both single and multi-node jobs.
         account
             An optional account to use for billing purposes.
         custom_attributes
@@ -35,6 +39,10 @@ def __init__(self, spec: str, queue_name: Optional[str] = None,
             self.url = None
 
         self.queue_name = queue_name
+        if multi_node_queue_name is None:
+            self.multi_node_queue_name = queue_name
+        else:
+            self.multi_node_queue_name = multi_node_queue_name
         self.account = account
         self.custom_attributes_raw = custom_attributes_raw
         self.custom_attributes: Dict[str, object] = {}