Fix container resource sampling and surface collection failures
This commit is contained in:
+5
-3
@@ -32,7 +32,7 @@ import urllib.parse
|
||||
import urllib.request
|
||||
import uuid
|
||||
|
||||
VERSION = "1.4.5"
|
||||
VERSION = "1.4.6"
|
||||
LABEL = "io.electric-cherry"
|
||||
MAX_FILE = 32 * 1024 * 1024
|
||||
LIBRARY_DOWNLOAD_WORKERS = 4
|
||||
@@ -933,7 +933,7 @@ class Agent:
|
||||
|
||||
def instance_resources(self):
|
||||
"""One bounded Docker stats batch; unavailable samples are never zero load."""
|
||||
ids = self.docker.call("container", "ls", "--filter", LABEL + ".managed=true",
|
||||
ids = self.docker.call("container", "ls", "--filter", "label=" + LABEL + ".managed=true",
|
||||
"--no-trunc", "--format", "{{.ID}}", timeout=10).splitlines()[:1000]
|
||||
if not ids:
|
||||
return []
|
||||
@@ -1084,8 +1084,10 @@ class Agent:
|
||||
images.extend(alias for alias, image_id in json.loads(aliases.read_text()).items() if image_id in image_ids)
|
||||
resources = {}
|
||||
if docker_running:
|
||||
with contextlib.suppress(AgentError, ValueError, TypeError, KeyError):
|
||||
try:
|
||||
resources["instanceResources"] = self.instance_resources()
|
||||
except (AgentError, ValueError, TypeError, KeyError):
|
||||
resources["instanceResourcesError"] = "Container resource sampling failed. Check Docker access and stats availability on this host."
|
||||
os_name = platform.system()
|
||||
with contextlib.suppress(OSError):
|
||||
for line in Path("/etc/os-release").read_text().splitlines():
|
||||
|
||||
@@ -124,6 +124,7 @@ class AgentTests(unittest.TestCase):
|
||||
self.assertEqual(samples[0]["cpuCores"], 1.5)
|
||||
self.assertEqual(samples[0]["memoryBytes"], int(384.5 * 1024 ** 2))
|
||||
self.assertGreater(samples[0]["uptimeSeconds"], 300)
|
||||
self.assertEqual(call.call_args_list[0].args, ("container", "ls", "--filter", "label=io.electric-cherry.managed=true", "--no-trunc", "--format", "{{.ID}}"))
|
||||
self.assertEqual(call.call_args_list[-1].args, ("stats", "--no-stream", "--no-trunc", "--format", "{{json .}}", name))
|
||||
|
||||
def test_instance_resources_ignores_stopped_foreign_and_mismatched_containers(self):
|
||||
@@ -149,7 +150,13 @@ class AgentTests(unittest.TestCase):
|
||||
with patch.object(self.docker, "call", side_effect=["container-id\n", json.dumps([self.resource_container()]), json.dumps(good)]):
|
||||
self.assertEqual(self.agent.instance_resources()[0]["cpuCores"], 0)
|
||||
with patch.object(self.agent, "instance_resources", side_effect=AgentError("timed out")):
|
||||
self.assertNotIn("instanceResources", self.agent.inventory())
|
||||
inventory = self.agent.inventory()
|
||||
self.assertNotIn("instanceResources", inventory)
|
||||
self.assertIn("sampling failed", inventory["instanceResourcesError"])
|
||||
with patch.object(self.agent, "instance_resources", return_value=[]):
|
||||
inventory = self.agent.inventory()
|
||||
self.assertEqual(inventory["instanceResources"], [])
|
||||
self.assertNotIn("instanceResourcesError", inventory)
|
||||
|
||||
def operation(self, kind, payload=None, op_id="operation-1", generation=1):
|
||||
return {"id": op_id, "kind": kind, "instanceId": INSTANCE, "generation": generation, "payload": payload or {}}
|
||||
|
||||
Reference in New Issue
Block a user