X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/a50278e3d0e26bb5d513d0af5da2fb559b112388..cac5db66cba0d5dd97c8434853bcbf2ab19fbda5:/services/nodemanager/arvnodeman/computenode/__init__.py

diff --git a/services/nodemanager/arvnodeman/computenode/__init__.py b/services/nodemanager/arvnodeman/computenode/__init__.py
index 6e46bc0f4c..54d6a82bce 100644
--- a/services/nodemanager/arvnodeman/computenode/__init__.py
+++ b/services/nodemanager/arvnodeman/computenode/__init__.py
@@ -3,6 +3,7 @@
 from __future__ import absolute_import, print_function
 
 import calendar
+import functools
 import itertools
 import re
 import time
@@ -43,6 +44,69 @@ def arvados_node_missing(arvados_node, fresh_time):
     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:
+                    try:
+                        ret = orig_func(self, *args, **kwargs)
+                    except Exception as error:
+                        if not (isinstance(error, errors) or
+                                self._cloud.is_cloud_exception(error)):
+                            self.retry_wait = self.min_retry_wait
+                            self._logger.warning(
+                                "Re-raising unknown error (no retry): %s",
+                                error, exc_info=error)
+                            raise
+
+                        self._logger.warning(
+                            "Client error: %s - waiting %s seconds",
+                            error, 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
+                    else:
+                        self.retry_wait = self.min_retry_wait
+                        return ret
+            return retry_wrapper
+        return decorator
+
 class ShutdownTimer(object):
     """Keep track of a cloud node's shutdown windows.