def spawn_updater(self, node_id, init_commands, ray_start_commands,
node_resources, docker_config):
logger.info(f"Creating new (spawn_updater) updater thread for node"
f" {node_id}.")
updater = NodeUpdaterThread(
node_id=node_id,
provider_config=self.config["provider"],
provider=self.provider,
auth_config=self.config["auth"],
cluster_name=self.config["cluster_name"],
file_mounts=self.config["file_mounts"],
initialization_commands=with_head_node_ip(
self._get_node_type_specific_fields(node_id,
"initialization_commands"),
self.head_node_ip),
setup_commands=with_head_node_ip(init_commands, self.head_node_ip),
ray_start_commands=with_head_node_ip(ray_start_commands,
self.head_node_ip),
runtime_hash=self.runtime_hash,
file_mounts_contents_hash=self.file_mounts_contents_hash,
is_head_node=False,
cluster_synced_files=self.config["cluster_synced_files"],
rsync_options={
"rsync_exclude": self.config.get("rsync_exclude"),
"rsync_filter": self.config.get("rsync_filter")
},
process_runner=self.process_runner,
use_internal_ip=True,
docker_config=docker_config,
node_resources=node_resources)
updater.start()
self.updaters[node_id] = updater
def recover_if_needed(self, node_id, now):
if not self.can_update(node_id):
return
key = self.provider.internal_ip(node_id)
if key not in self.load_metrics.last_heartbeat_time_by_ip:
self.load_metrics.last_heartbeat_time_by_ip[key] = now
last_heartbeat_time = self.load_metrics.last_heartbeat_time_by_ip[key]
delta = now - last_heartbeat_time
if delta < AUTOSCALER_HEARTBEAT_TIMEOUT_S:
return
logger.warning("StandardAutoscaler: "
"{}: No heartbeat in {}s, "
"restarting Ray to recover...".format(node_id, delta))
updater = NodeUpdaterThread(
node_id=node_id,
provider_config=self.config["provider"],
provider=self.provider,
auth_config=self.config["auth"],
cluster_name=self.config["cluster_name"],
file_mounts={},
initialization_commands=[],
setup_commands=[],
ray_start_commands=with_head_node_ip(
self.config["worker_start_ray_commands"], self.head_node_ip),
runtime_hash=self.runtime_hash,
file_mounts_contents_hash=self.file_mounts_contents_hash,
process_runner=self.process_runner,
use_internal_ip=True,
is_head_node=False,
docker_config=self.config.get("docker"),
node_resources=self._node_resources(node_id))
updater.start()
self.updaters[node_id] = updater
def recover_if_needed(self, node_id, now):
if not self.can_update(node_id):
return
if self.heartbeat_on_time(node_id, now):
return
logger.warning("StandardAutoscaler: "
"{}: No recent heartbeat, "
"restarting Ray to recover...".format(node_id))
self.event_summarizer.add("Restarting {} nodes of type " +
self._get_node_type(node_id) +
" (lost contact with raylet).",
quantity=1,
aggregate=operator.add)
updater = NodeUpdaterThread(
node_id=node_id,
provider_config=self.config["provider"],
provider=self.provider,
auth_config=self.config["auth"],
cluster_name=self.config["cluster_name"],
file_mounts={},
initialization_commands=[],
setup_commands=[],
ray_start_commands=with_head_node_ip(
self.config["worker_start_ray_commands"], self.head_node_ip),
runtime_hash=self.runtime_hash,
file_mounts_contents_hash=self.file_mounts_contents_hash,
process_runner=self.process_runner,
use_internal_ip=True,
is_head_node=False,
docker_config=self.config.get("docker"),
node_resources=self._node_resources(node_id),
for_recovery=True)
updater.start()
self.updaters[node_id] = updater
