zuul
/
nodepool
Code Issues Proposed changes

Don't yield to a provider with unsupported labels 

The optimization introduced in Id04f57ff1e4c28357370729b6383f5119cd616dc
can lead to a starvation of certain requests under the following
conditions:

* node request with a requested provider that doesn't support the
  required node types
* another provider that could technically serve the request yields to
  the requested provider
* requested provider is at quota or high inflow of node request
* requested provider postpones rejecting the yielded request as it
  doesn't support the required labels

To avoid starvation of those request, we only yield to the requested
provider if it is capable of serving the request. This is done by
checking the supported labels of the requested provider.

Change-Id: I0ded785a13d1f955a71d519dc40e5e5c0ec35043
This commit is contained in:
Simon Westphahl 2022-03-24 09:22:12 +01:00
parent b13feb7ae8
commit 4f0628c8c2
2 changed files with 33 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
nodepool/launcher.py
View File

@ -162,8 +162,9 @@ class PoolWorker(threading.Thread, stats.StatsReporter):
if req.provider and req.provider != self.provider_name:
# The request is asking for a specific provider
candidate_launchers = set(
[x.id for x in launchers
if x.provider_name == req.provider])
x.id for x in launchers
if x.provider_name == req.provider
and x.supported_labels.issuperset(req.node_types))
if candidate_launchers:
# There is a launcher online which can satisfy the request
if not candidate_launchers.issubset(set(req.declined_by)):

30
nodepool/tests/unit/test_launcher.py
View File

@ -744,6 +744,36 @@ class TestLauncher(tests.DBTestCase):
self.assertEqual(len(req1.nodes), 1)
self.zk.getNode(req1.nodes[0])
def test_node_request_provider_label_mismatch(self):
"""Test that a node request for a specific provider is only honored
when the requested labels are supported."""
configfile = self.setup_config('node.yaml')
self.useBuilder(configfile)
pool = self.useNodepool(configfile, watermark_sleep=1)
pool.start()
self.waitForImage('fake-provider', 'fake-image')
self.waitForNodes('fake-label', 1)
# Create a dummy launcher with a different set of supported labels
# than what we are going to request.
dummy_launcher = zk.Launcher()
dummy_launcher.provider_name = 'other-provider'
dummy_launcher.supported_labels = {'other-label', }
self.zk.registerLauncher(dummy_launcher)
# Node request for a specific provider that doesn't support the
# requested node type.
req = zk.NodeRequest()
req.state = zk.REQUESTED
req.provider = 'other-provider'
req.node_types.append('fake-label')
self.zk.storeNodeRequest(req)
req = self.waitForNodeRequest(req)
self.assertEqual(req.state, zk.FULFILLED)
self.assertEqual(len(req.nodes), 1)
self.zk.getNode(req.nodes[0])
def test_node_boot_from_volume(self):
"""Test that an image and node are created from a volume"""
configfile = self.setup_config('node_boot_from_volume.yaml')