Wait for primary scheduler to converge

azure-slurm-install/install.py

@@ -1132,6 +1132,7 @@ def main() -> None:
     # create the users
     setup_users(settings)
 
+    set_hostname(settings)
     # create the munge key and/or copy it to /etc/munge/
     munge_key(settings)
 
@@ -1152,8 +1153,10 @@ def main() -> None:
             minute="*/5",
             command=f"{settings.autoscale_dir}/return_to_idle.sh 1>&2 >> {settings.autoscale_dir}/logs/return_to_idle.log",
         )
-
-    set_hostname(settings)
+        if settings.is_primary_scheduler == False:
+            # This is the HA node.
+            logging.info(f"Secondary Scheduler {settings.secondary_scheduler_name} starting wait on primary to finish converging.")
+            ilib.await_node_converge(settings.config, "scheduler", timeout=3600)
 
     if settings.mode == "execute":
         setup_slurmd(settings)

azure-slurm-install/installlib.py

@@ -615,6 +615,29 @@ def await_node_hostname(
     )
 
 
+def await_node_converge(
+    config: Dict,
+    node_name: str,
+    timeout=3600,
+    cluster_status_func: Callable[[Dict], Dict] = cluster_status,
+) -> CCNode:
+    """
+    Waits for node_name to fully converge.
+    """
+    omega = timeout + time()
+    while time() < omega:
+        referenced_node = get_ccnode(config, node_name, cluster_status_func)
+        if referenced_node.status == "Ready":
+            return referenced_node
+        logging.debug(
+            "Waiting for node to converge %s",
+            referenced_node.hostname,
+        )
+        sleep(10)
+    raise RuntimeError(
+        f"Node {node_name} did not converge in {timeout} seconds"
+    )
+
 def is_valid_hostname(config: Dict, node: CCNode) -> bool:
     """
 See await_node_hostname for details.