From fe0fc10bd0774909922a8fd71dc45b77dfe132c8 Mon Sep 17 00:00:00 2001
From: Valentin Rigal <rigal@teklia.com>
Date: Tue, 10 Oct 2023 13:05:16 +0000
Subject: [PATCH] Fix ponos tasks assignment failures
---
arkindex/ponos/models.py | 3 ++
arkindex/ponos/tests/test_api.py | 55 ++++++++++++++++++++++++--------
2 files changed, 45 insertions(+), 13 deletions(-)
diff --git a/arkindex/ponos/models.py b/arkindex/ponos/models.py
index 907aee6680..bcf831a13b 100644
--- a/arkindex/ponos/models.py
+++ b/arkindex/ponos/models.py
@@ -159,6 +159,9 @@ class Agent(models.Model):
:param tasks: Number of tasks to estimate the cost for.
:returns: A cost expressed as a percentage. If > 1, the agent would be overloaded.
"""
+ if self.cpu_load is None or self.ram_load is None:
+ # The agent has not shared its state yet
+ return 1
current_tasks_count = getattr(self, 'current_tasks', 0)
if current_tasks_count + AGENT_SLOT["cpu"] >= self.cpu_cores:
return 1
diff --git a/arkindex/ponos/tests/test_api.py b/arkindex/ponos/tests/test_api.py
index 72b96030d1..2c88c3e28b 100644
--- a/arkindex/ponos/tests/test_api.py
+++ b/arkindex/ponos/tests/test_api.py
@@ -56,6 +56,8 @@ class TestAPI(FixtureAPITestCase):
public_key=pubkey,
ram_total=2e9,
last_ping=timezone.now(),
+ cpu_load=.1,
+ ram_load=.1e9,
)
cls.rev = Revision.objects.first()
cls.process = Process.objects.get(mode=ProcessMode.Workers)
@@ -834,7 +836,7 @@ class TestAPI(FixtureAPITestCase):
"agent": {
"cpu_cores": 2,
"cpu_frequency": 1000000000,
- "cpu_load": None,
+ "cpu_load": .1,
"farm": {"id": str(self.agent.farm_id), "name": "Wheat farm"},
"gpus": [
{
@@ -853,7 +855,7 @@ class TestAPI(FixtureAPITestCase):
"hostname": "ghostname",
"id": str(self.agent.id),
"last_ping": str_date(self.agent.last_ping),
- "ram_load": None,
+ "ram_load": 100000000,
"ram_total": 2000000000,
},
"gpu": {
@@ -890,7 +892,7 @@ class TestAPI(FixtureAPITestCase):
"agent": {
"cpu_cores": 2,
"cpu_frequency": 1000000000,
- "cpu_load": None,
+ "cpu_load": .1,
"farm": {"id": str(self.agent.farm_id), "name": "Wheat farm"},
"gpus": [
{
@@ -909,7 +911,7 @@ class TestAPI(FixtureAPITestCase):
"hostname": "ghostname",
"id": str(self.agent.id),
"last_ping": str_date(self.agent.last_ping),
- "ram_load": None,
+ "ram_load": 100000000,
"ram_total": 2000000000,
},
"gpu": {
@@ -1650,7 +1652,7 @@ class TestAPI(FixtureAPITestCase):
{
"cpu_cores": 12,
"cpu_frequency": 1000000000,
- "cpu_load": None,
+ "cpu_load": .1,
"farm": str(self.wheat_farm.id),
"gpus": [
{
@@ -1663,7 +1665,7 @@ class TestAPI(FixtureAPITestCase):
"hostname": "ghostname",
"id": str(self.agent.id),
"last_ping": "2000-01-01T12:00:00Z",
- "ram_load": None,
+ "ram_load": 100000000,
"ram_total": 32000000000,
},
)
@@ -1859,25 +1861,52 @@ class TestAPI(FixtureAPITestCase):
},
)
+ def test_agent_null_state(self):
+ """
+ Agents with unknown CPU or RAM load are excluded by the assignation algorithm
+ """
+ self.agent.cpu_load = None
+ self.agent.ram_load = None
+ self.agent.save()
+ pubkey = build_public_key()
+ second_agent = AgentUser.objects.create(
+ id=hashlib.md5(pubkey.encode("utf-8")).hexdigest(),
+ farm=self.wheat_farm,
+ hostname="new agent",
+ cpu_cores=2,
+ cpu_frequency=1e9,
+ public_key=pubkey,
+ ram_total=1e9,
+ last_ping=timezone.now(),
+ )
+ with self.assertNumQueries(6):
+ resp = self.client.get(
+ reverse("api:agent-actions"),
+ HTTP_AUTHORIZATION=f'Bearer {second_agent.token.access_token}',
+ data={"cpu_load": 1.9, "ram_load": 0.49},
+ )
+ self.assertEqual(resp.status_code, status.HTTP_200_OK)
+ self.assertDictEqual(resp.json(), {"actions": []})
+
def test_agent_non_pending_actions(self):
"""
Only pending tasks may be retrieved as new actions
"""
- self.process.tasks.update(state=State.Error)
+ Task.objects.filter(process__farm=self.agent.farm_id).update(state=State.Error)
with self.assertNumQueries(7):
resp = self.client.get(
reverse("api:agent-actions"),
HTTP_AUTHORIZATION=f'Bearer {self.agent.token.access_token}',
data={"cpu_load": 0.9, "ram_load": 0.49},
)
- self.assertEqual(resp.status_code, status.HTTP_200_OK)
+ self.assertEqual(resp.status_code, status.HTTP_200_OK)
self.assertDictEqual(resp.json(), {"actions": []})
def test_agent_no_stealing(self):
"""
An agent may not take another agent's tasks
"""
- self.process.tasks.update(agent=self.agent, state=State.Pending)
+ Task.objects.filter(process__farm=self.agent.farm_id).update(agent=self.agent, state=State.Pending)
pubkey = build_public_key()
agent2 = AgentUser.objects.create(
id=uuid.UUID(hashlib.md5(pubkey.encode("utf-8")).hexdigest()),
@@ -2377,7 +2406,7 @@ class TestAPI(FixtureAPITestCase):
"active": True,
"cpu_cores": 2,
"cpu_frequency": 1000000000,
- "cpu_load": None,
+ "cpu_load": .1,
"farm": {
"id": str(self.wheat_farm.id),
"name": "Wheat farm",
@@ -2397,7 +2426,7 @@ class TestAPI(FixtureAPITestCase):
},
],
"hostname": "ghostname",
- "ram_load": None,
+ "ram_load": 100000000,
"ram_total": 2000000000,
"running_tasks_count": 1,
},
@@ -2458,7 +2487,7 @@ class TestAPI(FixtureAPITestCase):
"active": True,
"cpu_cores": 2,
"cpu_frequency": 1000000000,
- "cpu_load": None,
+ "cpu_load": .1,
"farm": {
"id": str(self.wheat_farm.id),
"name": "Wheat farm",
@@ -2478,7 +2507,7 @@ class TestAPI(FixtureAPITestCase):
},
],
"hostname": "ghostname",
- "ram_load": None,
+ "ram_load": 100000000,
"ram_total": 2000000000,
"running_tasks": [
{
--
GitLab