zuul/nodepool
Code Issues Proposed changes

Merge "Add provider/pool priority support"

Browse Source
This commit is contained in:
Zuul 2022-05-24 16:02:34 +00:00 committed by Gerrit Code Review
commit ff7dd84aef
18 changed files with 349 additions and 17 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
doc/source/aws.rst
View File

@ -353,6 +353,20 @@ Selecting the ``aws`` driver adds the following options to the
A unique name within the provider for this pool of resources.
.. attr:: priority
:type: int
:default: 100
The priority of this provider pool (a lesser number is a higher
priority). Nodepool launchers will yield requests to other
provider pools with a higher priority as long as they are not
paused. This means that in general, higher priority pools will
reach quota first before lower priority pools begin to be used.
This setting may be specified at the provider level in order
to apply to all pools within that provider, or it can be
overridden here for a specific pool.
.. attr:: node-attributes
:type: dict

14
doc/source/azure.rst
View File

@ -487,6 +487,20 @@ section of the configuration.
A unique name within the provider for this pool of resources.
.. attr:: priority
:type: int
:default: 100
The priority of this provider pool (a lesser number is a higher
priority). Nodepool launchers will yield requests to other
provider pools with a higher priority as long as they are not
paused. This means that in general, higher priority pools will
reach quota first before lower priority pools begin to be used.
This setting may be specified at the provider level in order
to apply to all pools within that provider, or it can be
overridden here for a specific pool.
.. attr:: node-attributes
:type: dict

13
doc/source/configuration.rst
View File

@ -459,6 +459,19 @@ Options
thread, this can be useful for limiting the number of threads
used by the nodepool-launcher daemon.
.. attr:: priority
:type: int
:default: 100
The priority of this provider (a lesser number is a higher
priority). Nodepool launchers will yield requests to other
provider pools with a higher priority as long as they are not
paused. This means that in general, higher priority pools will
reach quota first before lower priority pools begin to be used.
This setting provides the default for each provider pool, but
the value can be overidden in the pool configuration.
.. attr:: driver
:type: string
:default: openstack

14
doc/source/gce.rst
View File

@ -195,6 +195,20 @@ section of the configuration.
A unique name within the provider for this pool of resources.
.. attr:: priority
:type: int
:default: 100
The priority of this provider pool (a lesser number is a higher
priority). Nodepool launchers will yield requests to other
provider pools with a higher priority as long as they are not
paused. This means that in general, higher priority pools will
reach quota first before lower priority pools begin to be used.
This setting may be specified at the provider level in order
to apply to all pools within that provider, or it can be
overridden here for a specific pool.
.. attr:: node-attributes
:type: dict

14
doc/source/ibmvpc.rst
View File

@ -570,6 +570,20 @@ section of the configuration.
A unique name within the provider for this pool of
resources. The name may not contain underscores.
.. attr:: priority
:type: int
:default: 100
The priority of this provider pool (a lesser number is a higher
priority). Nodepool launchers will yield requests to other
provider pools with a higher priority as long as they are not
paused. This means that in general, higher priority pools will
reach quota first before lower priority pools begin to be used.
This setting may be specified at the provider level in order
to apply to all pools within that provider, or it can be
overridden here for a specific pool.
.. attr:: zone
Name of the IBM Cloud region zone to interact with (e.g.,

14
doc/source/kubernetes.rst
View File

@ -76,6 +76,20 @@ Selecting the kubernetes driver adds the following options to the
Namespaces are prefixed with the pool's name.
.. attr:: priority
:type: int
:default: 100
The priority of this provider pool (a lesser number is a higher
priority). Nodepool launchers will yield requests to other
provider pools with a higher priority as long as they are not
paused. This means that in general, higher priority pools will
reach quota first before lower priority pools begin to be used.
This setting may be specified at the provider level in order
to apply to all pools within that provider, or it can be
overridden here for a specific pool.
.. attr:: node-attributes
:type: dict

14
doc/source/metastatic.rst
View File

@ -71,6 +71,20 @@ itself, which is "meta".
A unique name within the provider for this pool of resources.
.. attr:: priority
:type: int
:default: 100
The priority of this provider pool (a lesser number is a higher
priority). Nodepool launchers will yield requests to other
provider pools with a higher priority as long as they are not
paused. This means that in general, higher priority pools will
reach quota first before lower priority pools begin to be used.
This setting may be specified at the provider level in order
to apply to all pools within that provider, or it can be
overridden here for a specific pool.
.. attr:: max-servers
:type: int

14
doc/source/openshift-pods.rst
View File

@ -61,6 +61,20 @@ Selecting the openshift pods driver adds the following options to the
The project's (namespace) name that will be used to create the pods.
.. attr:: priority
:type: int
:default: 100
The priority of this provider pool (a lesser number is a higher
priority). Nodepool launchers will yield requests to other
provider pools with a higher priority as long as they are not
paused. This means that in general, higher priority pools will
reach quota first before lower priority pools begin to be used.
This setting may be specified at the provider level in order
to apply to all pools within that provider, or it can be
overridden here for a specific pool.
.. attr:: node-attributes
:type: dict

14
doc/source/openshift.rst
View File

@ -78,6 +78,20 @@ Selecting the openshift driver adds the following options to the
Project's name are prefixed with the pool's name.
.. attr:: priority
:type: int
:default: 100
The priority of this provider pool (a lesser number is a higher
priority). Nodepool launchers will yield requests to other
provider pools with a higher priority as long as they are not
paused. This means that in general, higher priority pools will
reach quota first before lower priority pools begin to be used.
This setting may be specified at the provider level in order
to apply to all pools within that provider, or it can be
overridden here for a specific pool.
.. attr:: node-attributes
:type: dict

14
doc/source/openstack.rst
View File

@ -419,6 +419,20 @@ Selecting the OpenStack driver adds the following options to the
Pool name
.. attr:: priority
:type: int
:default: 100
The priority of this provider pool (a lesser number is a higher
priority). Nodepool launchers will yield requests to other
provider pools with a higher priority as long as they are not
paused. This means that in general, higher priority pools will
reach quota first before lower priority pools begin to be used.
This setting may be specified at the provider level in order
to apply to all pools within that provider, or it can be
overridden here for a specific pool.
.. attr:: node-attributes
:type: dict

14
doc/source/static.rst
View File

@ -60,6 +60,20 @@ Selecting the static driver adds the following options to the
Pool name
.. attr:: priority
:type: int
:default: 100
The priority of this provider pool (a lesser number is a higher
priority). Nodepool launchers will yield requests to other
provider pools with a higher priority as long as they are not
paused. This means that in general, higher priority pools will
reach quota first before lower priority pools begin to be used.
This setting may be specified at the provider level in order
to apply to all pools within that provider, or it can be
overridden here for a specific pool.
.. attr:: node-attributes
:type: dict

7
nodepool/driver/__init__.py
View File

@ -860,6 +860,7 @@ class ConfigPool(ConfigValue, metaclass=abc.ABCMeta):
self.labels = {}
self.max_servers = math.inf
self.node_attributes = None
self.priority = None
@classmethod
def getCommonSchemaDict(self):
@ -876,6 +877,7 @@ class ConfigPool(ConfigValue, metaclass=abc.ABCMeta):
return {
'max-servers': int,
'node-attributes': dict,
'priority': int,
}
@abc.abstractmethod
@ -894,6 +896,7 @@ class ConfigPool(ConfigValue, metaclass=abc.ABCMeta):
'''
self.max_servers = pool_config.get('max-servers', math.inf)
self.node_attributes = pool_config.get('node-attributes')
self.priority = pool_config.get('priority', None)
class DriverConfig(ConfigValue):
@ -913,13 +916,15 @@ class ProviderConfig(ConfigValue, metaclass=abc.ABCMeta):
self.driver = DriverConfig()
self.driver.name = provider.get('driver', 'openstack')
self.max_concurrency = provider.get('max-concurrency', -1)
self.priority = provider.get('priority', None)
@classmethod
def getCommonSchemaDict(self):
return {
v.Required('name'): str,
'driver': str,
'max-concurrency': int
'max-concurrency': int,
'priority': int,
}
@property

7
nodepool/driver/ibmvpc/config.py
View File

@ -297,12 +297,11 @@ class IBMVPCProviderConfig(ProviderConfig):
'ram': int,
}
provider = {
provider = ProviderConfig.getCommonSchemaDict()
provider.update({
# Normally provided by superclass; we override to add the
# underscore restriction
v.Required('name'): v.Match(r'^[^_]+$'),
'driver': str,
'max-concurrency': int,
# For this class
v.Required('pools'): [pool],
v.Required('vpc'): str,
@ -322,7 +321,7 @@ class IBMVPCProviderConfig(ProviderConfig):
'launch-retries': int,
'credentials-file': str,
'object-storage': object_storage,
}
})
return v.Schema(provider)
def getSupportedLabels(self, pool_name=None):

54
nodepool/launcher.py
View File

@ -81,6 +81,17 @@ class PoolWorker(threading.Thread, stats.StatsReporter):
uuid.uuid4().hex)
stats.StatsReporter.__init__(self)
def getPriority(self):
pool = self.getPoolConfig()
provider = self.getProviderConfig()
if pool.priority is not None:
priority = pool.priority
elif provider.priority is not None:
priority = provider.priority
else:
priority = 100
return priority
# ---------------------------------------------------------------
# Private methods
# ---------------------------------------------------------------
@ -168,21 +179,37 @@ class PoolWorker(threading.Thread, stats.StatsReporter):
log = get_annotated_logger(self.log, event_id=req.event_id,
node_request_id=req.id)
# Get the candidate launchers for these nodes
candidate_launchers = set(
x for x in launchers
if (set(x.supported_labels).issuperset(set(req.node_types)) and
x.id not in req.declined_by)
)
# Skip this request if it is requesting another provider
# which is online
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
launcher_ids_for_provider = set(
x.id for x in candidate_launchers
if x.provider_name == req.provider
and set(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)):
# It has not yet declined the request, so yield to it.
log.debug("Yielding request to provider %s %s",
req.provider, candidate_launchers)
continue
)
if launcher_ids_for_provider:
# There is a launcher online which can satisfy the
# request that has not yet declined the request,
# so yield to it.
log.debug("Yielding request to provider %s %s",
req.provider, launcher_ids_for_provider)
continue
priority = self.getPriority()
launcher_ids_with_higher_priority = set(
x.id for x in candidate_launchers
if x.priority < priority and not x.paused
)
if launcher_ids_with_higher_priority:
log.debug("Yielding request to higher priority providers %s",
launcher_ids_with_higher_priority)
continue
if has_quota_support and not all(label_quota.get(l, math.inf) > 0
for l in req.node_types):
@ -363,6 +390,7 @@ class PoolWorker(threading.Thread, stats.StatsReporter):
'provider_name': self.provider_name,
'supported_labels': list(labels),
'state': self.component_info.RUNNING,
'priority': self.getPriority(),
})
self.component_info.register()
@ -380,8 +408,8 @@ class PoolWorker(threading.Thread, stats.StatsReporter):
labels = set()
for prov_cfg in self.nodepool.config.providers.values():
labels.update(prov_cfg.getSupportedLabels())
if set(self.component_info.supported_labels) != labels:
self.component_info.supported_labels = list(labels)
self.component_info.supported_labels = list(labels)
self.component_info.priority = self.getPriority()
self.updateProviderLimits(
self.nodepool.config.providers.get(self.provider_name))
@ -389,6 +417,7 @@ class PoolWorker(threading.Thread, stats.StatsReporter):
self.nodepool.config.tenant_resource_limits)
if not self.paused_handler:
self.component_info.paused = False
while not self._assignHandlers():
# _assignHandlers can take quite some time on a busy
# system so sprinkle _removeCompletedHandlers in
@ -396,6 +425,7 @@ class PoolWorker(threading.Thread, stats.StatsReporter):
# requests that already have all nodes.
self._removeCompletedHandlers()
else:
self.component_info.paused = True
# If we are paused, one request handler could not
# satisfy its assigned request, so give it
# another shot. Unpause ourselves if it completed.

87
nodepool/tests/fixtures/priority.yaml vendored Normal file
View File

@ -0,0 +1,87 @@
elements-dir: .
images-dir: '{images_dir}'
build-log-dir: '{build_log_dir}'
build-log-retention: 1
zookeeper-servers:
- host: {zookeeper_host}
port: {zookeeper_port}
chroot: {zookeeper_chroot}
zookeeper-tls:
ca: {zookeeper_ca}
cert: {zookeeper_cert}
key: {zookeeper_key}
zookeeper-timeout: 20.0
labels:
- name: fake-label
min-ready: 0
providers:
- name: low-provider
priority: 2
cloud: fake
driver: fake
region-name: fake-region
rate: 0.0001
diskimages:
- name: fake-image
meta:
key: value
key2: value
pools:
- name: main
max-servers: 2
node-attributes:
key1: value1
key2: value2
availability-zones:
- az1
networks:
- net-name
labels:
- name: fake-label
diskimage: fake-image
min-ram: 8192
flavor-name: 'Fake'
- name: high-provider
priority: 1
cloud: fake
driver: fake
region-name: fake-region
rate: 0.0001
diskimages:
- name: fake-image
meta:
key: value
key2: value
pools:
- name: main
max-servers: 1
node-attributes:
key1: value1
key2: value2
availability-zones:
- az1
networks:
- net-name
labels:
- name: fake-label
diskimage: fake-image
min-ram: 8192
flavor-name: 'Fake'
diskimages:
- name: fake-image
elements:
- fedora
- vm
release: 21
dib-cmd: nodepool/tests/fake-image-create
env-vars:
TMPDIR: /opt/dib_tmp
DIB_IMAGE_CACHE: /opt/dib_cache
DIB_CLOUD_IMAGES: http://download.fedoraproject.org/pub/fedora/linux/releases/test/21-Beta/Cloud/Images/x86_64/
BASE_IMAGE_FILE: Fedora-Cloud-Base-20141029-21_Beta.x86_64.qcow2

44
nodepool/tests/unit/test_launcher.py
View File

@ -2503,3 +2503,47 @@ class TestLauncher(tests.DBTestCase):
# Ready for the real delete now
zk.ZooKeeper.deleteRawNode = real_method
self.waitForNodeDeletion(node)
def test_provider_priority(self):
"""Test provider priorities"""
configfile = self.setup_config('priority.yaml')
self.useBuilder(configfile)
pool = self.useNodepool(configfile, watermark_sleep=1)
pool.start()
self.waitForImage('low-provider', 'fake-image')
self.waitForImage('high-provider', 'fake-image')
# The first request should be handled by the highest priority
# provider (high-provider; priority 1)
req1 = zk.NodeRequest()
req1.state = zk.REQUESTED
req1.node_types.append('fake-label')
self.zk.storeNodeRequest(req1)
req1 = self.waitForNodeRequest(req1)
self.assertEqual(req1.state, zk.FULFILLED)
self.assertEqual(len(req1.nodes), 1)
node1 = self.zk.getNode(req1.nodes[0])
self.assertEqual(node1.provider, 'high-provider')
# The second request should also be handled by the highest
# priority provider, but since it has max-servers=1, this
# request should be paused, which will cause the provider to
# be paused.
req2 = zk.NodeRequest()
req2.state = zk.REQUESTED
req2.node_types.append('fake-label')
self.zk.storeNodeRequest(req2)
req2 = self.waitForNodeRequest(req2, (zk.PENDING,))
self.assertEqual(req2.state, zk.PENDING)
# The third request should be handled by the low priority
# provider now that the high provider is paused.
req3 = zk.NodeRequest()
req3.state = zk.REQUESTED
req3.node_types.append('fake-label')
self.zk.storeNodeRequest(req3)
req3 = self.waitForNodeRequest(req3)
self.assertEqual(req3.state, zk.FULFILLED)
self.assertEqual(len(req3.nodes), 1)
node3 = self.zk.getNode(req3.nodes[0])
self.assertEqual(node3.provider, 'low-provider')

5
nodepool/zk/components.py
View File

@ -104,6 +104,9 @@ class BaseComponent(ZooKeeperBase):
if name not in self.content.keys():
return super().__setattr__(name, value)
if self.content[name] == value:
return
# Set the value in the local content dict
self.content[name] = value
@ -214,6 +217,8 @@ class PoolComponent(BaseComponent):
"id": None,
"provider_name": None,
"supported_labels": [],
"priority": 100,
"paused": False,
}
self.content.update(self.initial_state)

9
releasenotes/notes/priority-cb2f0ad1829fbaaf.yaml Normal file
View File

@ -0,0 +1,9 @@
---
features:
- |
Priority may now be set at the provider or provider-pool level.
Providers or pools with the highest priority will fulfill requests
first, until they are at quota, at which point lower-priority
providers will begin to be used.
See :attr:`providers.priority` for details.