X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/a0e70cf1da033f7d94e728ab919bd8cfcabf3743..44c95f99098fa6c6acbfa82d4b6cbc6015eb6e39:/services/nodemanager/arvnodeman/computenode/__init__.py

diff --git a/services/nodemanager/arvnodeman/computenode/__init__.py b/services/nodemanager/arvnodeman/computenode/__init__.py
index 4955992faa..8a4e5f312b 100644
--- a/services/nodemanager/arvnodeman/computenode/__init__.py
+++ b/services/nodemanager/arvnodeman/computenode/__init__.py
@@ -1,21 +1,144 @@
 #!/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
 
+import calendar
+import functools
 import itertools
+import re
 import time
 
+from ..config import CLOUD_ERRORS
+from libcloud.common.exceptions import BaseHTTPError
+
+ARVADOS_TIMEFMT = '%Y-%m-%dT%H:%M:%SZ'
+ARVADOS_TIMESUBSEC_RE = re.compile(r'(\.\d+)Z$')
+
 def arvados_node_fqdn(arvados_node, default_hostname='dynamic.compute'):
     hostname = arvados_node.get('hostname') or default_hostname
     return '{}.{}'.format(hostname, arvados_node['domain'])
 
 def arvados_node_mtime(node):
-    return time.mktime(time.strptime(node['modified_at'] + 'UTC',
-                                     '%Y-%m-%dT%H:%M:%SZ%Z')) - time.timezone
+    return arvados_timestamp(node['modified_at'])
+
+def arvados_timestamp(timestr):
+    subsec_match = ARVADOS_TIMESUBSEC_RE.search(timestr)
+    if subsec_match is None:
+        subsecs = .0
+    else:
+        subsecs = float(subsec_match.group(1))
+        timestr = timestr[:subsec_match.start()] + 'Z'
+    return calendar.timegm(time.strptime(timestr + 'UTC',
+                                         ARVADOS_TIMEFMT + '%Z'))
 
 def timestamp_fresh(timestamp, fresh_time):
     return (time.time() - timestamp) < fresh_time
 
+def arvados_node_missing(arvados_node, fresh_time):
+    """Indicate if cloud node corresponding to the arvados
+    node is "missing".
+
+    If True, this means the node has not pinged the API server within the timeout
+    period.  If False, the ping is up to date.  If the node has never pinged,
+    returns None.
+    """
+    if arvados_node["last_ping_at"] is None:
+        return None
+    else:
+        return not timestamp_fresh(arvados_timestamp(arvados_node["last_ping_at"]), fresh_time)
+
+class RetryMixin(object):
+    """Retry decorator for an method that makes remote requests.
+
+    Use this function to decorate method, and pass in a tuple of exceptions to
+    catch.  If the original method raises a known cloud driver error, or any of
+    the given exception types, this decorator will either go into a
+    sleep-and-retry loop with exponential backoff either by sleeping (if
+    self._timer is None) or by scheduling retries of the method (if self._timer
+    is a timer actor.)
+
+    """
+    def __init__(self, retry_wait, max_retry_wait,
+                 logger, cloud, timer=None):
+        self.min_retry_wait = retry_wait
+        self.max_retry_wait = max_retry_wait
+        self.retry_wait = retry_wait
+        self._logger = logger
+        self._cloud = cloud
+        self._timer = timer
+
+    @staticmethod
+    def _retry(errors=()):
+        def decorator(orig_func):
+            @functools.wraps(orig_func)
+            def retry_wrapper(self, *args, **kwargs):
+                while True:
+                    should_retry = False
+                    try:
+                        ret = orig_func(self, *args, **kwargs)
+                    except BaseHTTPError as error:
+                        if error.headers and error.headers.get("retry-after"):
+                            try:
+                                self.retry_wait = int(error.headers["retry-after"])
+                                if self.retry_wait < 0 or self.retry_wait > self.max_retry_wait:
+                                    self.retry_wait = self.max_retry_wait
+                                should_retry = True
+                            except ValueError:
+                                pass
+                        if error.code == 429 or error.code >= 500:
+                            should_retry = True
+                    except CLOUD_ERRORS as error:
+                        should_retry = True
+                    except errors as error:
+                        should_retry = True
+                    except Exception as error:
+                        # As a libcloud workaround for drivers that don't use
+                        # typed exceptions, consider bare Exception() objects
+                        # retryable.
+                        should_retry = type(error) is Exception
+                    else:
+                        # No exception,
+                        self.retry_wait = self.min_retry_wait
+                        return ret
+
+                    # Only got here if an exception was caught.  Now determine what to do about it.
+                    if not should_retry:
+                        self.retry_wait = self.min_retry_wait
+                        self._logger.warning(
+                            "Re-raising error (no retry): %s",
+                            error, exc_info=error)
+                        raise
+
+                    self._logger.warning(
+                        "Client error: %s - %s %s seconds",
+                        error,
+                        "scheduling retry in" if self._timer else "sleeping",
+                        self.retry_wait,
+                        exc_info=error)
+
+                    if self._timer:
+                        start_time = time.time()
+                        # reschedule to be called again
+                        self._timer.schedule(start_time + self.retry_wait,
+                                             getattr(self._later,
+                                                     orig_func.__name__),
+                                             *args, **kwargs)
+                    else:
+                        # sleep on it.
+                        time.sleep(self.retry_wait)
+
+                    self.retry_wait = min(self.retry_wait * 2,
+                                          self.max_retry_wait)
+                    if self._timer:
+                        # expect to be called again by timer so don't loop
+                        return
+
+            return retry_wrapper
+        return decorator
+
 class ShutdownTimer(object):
     """Keep track of a cloud node's shutdown windows.