From 55886eac960cfc6f311fb4c9812311c420761793 Mon Sep 17 00:00:00 2001
From: Zack Cerza <zack@redhat.com>
Date: Mon, 27 Feb 2023 17:39:25 -0700
Subject: [PATCH] dispatcher: Do not time out when locking machines

Signed-off-by: Zack Cerza <zack@redhat.com>
---
 teuthology/dispatcher/__init__.py |  9 +++++++--
 teuthology/lock/ops.py            | 11 ++++++++---
 teuthology/lock/query.py          |  8 ++++++--
 3 files changed, 21 insertions(+), 7 deletions(-)

diff --git a/teuthology/dispatcher/__init__.py b/teuthology/dispatcher/__init__.py
index 273f747e2..9aea132dd 100644
--- a/teuthology/dispatcher/__init__.py
+++ b/teuthology/dispatcher/__init__.py
@@ -220,8 +220,13 @@ def find_dispatcher_processes(machine_type):
 def lock_machines(job_config):
     report.try_push_job_info(job_config, dict(status='running'))
     fake_ctx = supervisor.create_fake_context(job_config, block=True)
-    block_and_lock_machines(fake_ctx, len(job_config['roles']),
-                            job_config['machine_type'], reimage=False)
+    block_and_lock_machines(
+        fake_ctx,
+        len(job_config['roles']),
+        job_config['machine_type'],
+        tries=-1,
+        reimage=False,
+    )
     job_config = fake_ctx.config
     return job_config
 
diff --git a/teuthology/lock/ops.py b/teuthology/lock/ops.py
index b0c7d8033..5ab995ad7 100644
--- a/teuthology/lock/ops.py
+++ b/teuthology/lock/ops.py
@@ -327,7 +327,7 @@ def reimage_machines(ctx, machines, machine_type):
     return reimaged
 
 
-def block_and_lock_machines(ctx, total_requested, machine_type, reimage=True):
+def block_and_lock_machines(ctx, total_requested, machine_type, reimage=True, tries=10):
     # It's OK for os_type and os_version to be None here.  If we're trying
     # to lock a bare metal machine, we'll take whatever is available.  If
     # we want a vps, defaults will be provided by misc.get_distro and
@@ -347,8 +347,13 @@ def block_and_lock_machines(ctx, total_requested, machine_type, reimage=True):
     requested = total_requested
     while True:
         # get a candidate list of machines
-        machines = query.list_locks(machine_type=machine_type, up=True,
-                                    locked=False, count=requested + reserved)
+        machines = query.list_locks(
+            machine_type=machine_type,
+            up=True,
+            locked=False,
+            count=requested + reserved,
+            tries=tries,
+        )
         if machines is None:
             if ctx.block:
                 log.error('Error listing machines, trying again')
diff --git a/teuthology/lock/query.py b/teuthology/lock/query.py
index bb1044c2b..9fd09d9ab 100644
--- a/teuthology/lock/query.py
+++ b/teuthology/lock/query.py
@@ -51,7 +51,7 @@ def is_vm(name=None, status=None):
     return status.get('is_vm', False)
 
 
-def list_locks(keyed_by_name=False, **kwargs):
+def list_locks(keyed_by_name=False, tries=10, **kwargs):
     uri = os.path.join(config.lock_server, 'nodes', '')
     for key, value in kwargs.items():
         if kwargs[key] is False:
@@ -63,7 +63,11 @@ def list_locks(keyed_by_name=False, **kwargs):
             kwargs['machine_type'] = kwargs['machine_type'].replace(',','|')
         uri += '?' + urlencode(kwargs)
     with safe_while(
-            sleep=1, increment=0.5, action='list_locks') as proceed:
+            sleep=1,
+            increment=0.5,
+            tries=-1,
+            action='list_locks'
+    ) as proceed:
         while proceed():
             try:
                 response = requests.get(uri)
-- 
2.47.3