Add executor-specific setup time support

This will allow us to make the setup time TTL for k8s-based builds much lower (on the order of a minute), which means faster timeouts and fallbacks (which is a better user experience).
This commit is contained in:
Joseph Schorr 2016-11-07 15:45:15 -05:00
parent 9aac68fbeb
commit ef41e57aad
3 changed files with 12 additions and 3 deletions

View file

@ -373,7 +373,7 @@ class EphemeralBuilderManager(BaseManager):
# Load components for all realms currently known to the cluster
async(self._register_existing_realms())
def setup_time(self):
def overall_setup_time(self):
return self._ephemeral_setup_timeout
def shutdown(self):
@ -418,7 +418,6 @@ class EphemeralBuilderManager(BaseManager):
realm = str(uuid.uuid4())
token = str(uuid.uuid4())
nonce = str(uuid.uuid4())
setup_time = self.setup_time()
machine_max_expiration = self._manager_config.get('MACHINE_MAX_TIME', 7200)
max_expiration = datetime.utcnow() + timedelta(seconds=machine_max_expiration)
@ -526,6 +525,9 @@ class EphemeralBuilderManager(BaseManager):
})
try:
setup_time = started_with_executor.setup_time or self.overall_setup_time()
logger.debug('Writing job key for job %s using executor %s with ID %s and ttl %s', build_uuid,
started_with_executor.name, execution_id, setup_time)
yield From(self._etcd_client.write(self._etcd_realm_key(realm), realm_spec, prevExist=False,
ttl=setup_time))
except (KeyError, etcd.EtcdKeyError):

View file

@ -55,6 +55,13 @@ class BuilderExecutor(object):
""" Name returns the unique name for this executor. """
return self.executor_config.get('NAME') or self.__class__.__name__
@property
def setup_time(self):
""" Returns the amount of time (in seconds) to wait for the execution to start for the build.
If None, the manager's default will be used.
"""
return self.executor_config.get('SETUP_TIME')
@coroutine
def start_builder(self, realm, token, build_uuid):
""" Create a builder with the specified config. Returns a unique id which can be used to manage

View file

@ -168,7 +168,7 @@ class BuilderServer(object):
logger.debug('Checking for more work for %d active workers',
self._lifecycle_manager.num_workers())
processing_time = self._lifecycle_manager.setup_time() + SETUP_LEEWAY_SECONDS
processing_time = self._lifecycle_manager.overall_setup_time() + SETUP_LEEWAY_SECONDS
job_item = self._queue.get(processing_time=processing_time, ordering_required=True)
if job_item is None:
logger.debug('No additional work found. Going to sleep for %s seconds', WORK_CHECK_TIMEOUT)