X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/b50a3bcb38cf9e79416d8cc9b0d8b66249b3d473..6fe8e52020d421797306e5c6536afbcee761510a:/services/nodemanager/arvnodeman/computenode/driver/__init__.py diff --git a/services/nodemanager/arvnodeman/computenode/driver/__init__.py b/services/nodemanager/arvnodeman/computenode/driver/__init__.py index 29b04845b6..48d19f592b 100644 --- a/services/nodemanager/arvnodeman/computenode/driver/__init__.py +++ b/services/nodemanager/arvnodeman/computenode/driver/__init__.py @@ -1,4 +1,7 @@ #!/usr/bin/env python +# Copyright (C) The Arvados Authors. All rights reserved. +# +# SPDX-License-Identifier: AGPL-3.0 from __future__ import absolute_import, print_function @@ -6,10 +9,10 @@ import logging from operator import attrgetter import libcloud.common.types as cloud_types -from libcloud.common.exceptions import BaseHTTPError from libcloud.compute.base import NodeDriver, NodeAuthSSHKey -from ...config import NETWORK_ERRORS +from ...config import CLOUD_ERRORS +from ...status import tracker from .. import RetryMixin class BaseComputeNodeDriver(RetryMixin): @@ -25,15 +28,17 @@ class BaseComputeNodeDriver(RetryMixin): Subclasses must implement arvados_create_kwargs, sync_node, node_fqdn, and node_start_time. """ - CLOUD_ERRORS = NETWORK_ERRORS + (cloud_types.LibcloudError,) + @RetryMixin._retry() def _create_driver(self, driver_class, **auth_kwargs): return driver_class(**auth_kwargs) @RetryMixin._retry() - def _set_sizes(self): - self.sizes = {sz.id: sz for sz in self.real.list_sizes()} + def sizes(self): + if self._sizes is None: + self._sizes = {sz.id: sz for sz in self.real.list_sizes()} + return self._sizes def __init__(self, auth_kwargs, list_kwargs, create_kwargs, driver_class, retry_wait=1, max_retry_wait=180): @@ -70,7 +75,7 @@ class BaseComputeNodeDriver(RetryMixin): if new_pair is not None: self.create_kwargs[new_pair[0]] = new_pair[1] - self._set_sizes() + self._sizes = None def _init_ping_host(self, ping_host): self.ping_host = ping_host @@ -121,7 +126,11 @@ class BaseComputeNodeDriver(RetryMixin): def list_nodes(self, **kwargs): l = self.list_kwargs.copy() l.update(kwargs) - return self.real.list_nodes(**l) + try: + return self.real.list_nodes(**l) + except CLOUD_ERRORS: + tracker.counter_add('list_nodes_errors') + raise def create_cloud_name(self, arvados_node): """Return a cloud node name for the given Arvados node record. @@ -167,9 +176,9 @@ class BaseComputeNodeDriver(RetryMixin): try: kwargs = self.create_kwargs.copy() kwargs.update(self.arvados_create_kwargs(size, arvados_node)) - kwargs['size'] = size + kwargs['size'] = size.real return self.real.create_node(**kwargs) - except self.CLOUD_ERRORS as create_error: + except CLOUD_ERRORS as create_error: # Workaround for bug #6702: sometimes the create node request # succeeds but times out and raises an exception instead of # returning a result. If this happens, we get stuck in a retry @@ -179,6 +188,7 @@ class BaseComputeNodeDriver(RetryMixin): try: return self.search_for_now(kwargs['name'], 'list_nodes', self._name_key) except ValueError: + tracker.counter_add('create_node_errors') raise create_error def post_create_node(self, cloud_node): @@ -206,23 +216,10 @@ class BaseComputeNodeDriver(RetryMixin): # seconds since the epoch UTC. raise NotImplementedError("BaseComputeNodeDriver.node_start_time") - @classmethod - def is_cloud_exception(cls, exception): - # libcloud compute drivers typically raise bare Exceptions to - # represent API errors. Return True for any exception that is - # exactly an Exception, or a better-known higher-level exception. - if (type(exception) is BaseHTTPError and - exception.message and - (exception.message.startswith("InvalidInstanceID.NotFound") or - exception.message.startswith("InstanceLimitExceeded"))): - return True - return (isinstance(exception, cls.CLOUD_ERRORS) or - type(exception) is Exception) - def destroy_node(self, cloud_node): try: return self.real.destroy_node(cloud_node) - except self.CLOUD_ERRORS as destroy_error: + except CLOUD_ERRORS: # Sometimes the destroy node request succeeds but times out and # raises an exception instead of returning success. If this # happens, we get a noisy stack trace. Check if the node is still @@ -234,6 +231,7 @@ class BaseComputeNodeDriver(RetryMixin): # it, which means destroy_node actually succeeded. return True # The node is still on the list. Re-raise. + tracker.counter_add('destroy_node_errors') raise # Now that we've defined all our own methods, delegate generic, public