Retry when tagging EC2 and EBS with Postgres roles. (#418)

Retry if an error happens when setting Role or Name tags for the
EC2 instances or EBS volumes. The maximum retry interval is 15 seconds.
This commit is contained in:
Oleksii Kliukin
2017-03-24 16:51:24 +01:00
committed by GitHub
parent 5bd7d043ae
commit 875e450ff8
2 changed files with 31 additions and 39 deletions
+25 -24
View File
@@ -6,13 +6,18 @@ from requests.exceptions import RequestException
import sys
import boto.ec2
from patroni.utils import Retry, RetryFailedError
logger = logging.getLogger(__name__)
retry_timeout = 15
class AWSConnection(object):
def __init__(self, cluster_name):
self.available = False
self.cluster_name = cluster_name if cluster_name is not None else 'unknown'
self._retry = Retry(deadline=retry_timeout, max_delay=5, max_tries=-1, retry_exceptions=(boto.exception,))
try:
# get the instance id
r = requests.get('http://169.254.169.254/latest/dynamic/instance-identity/document', timeout=0.1)
@@ -29,40 +34,36 @@ class AWSConnection(object):
return
self.available = True
def retry(self, *args, **kwargs):
return self._retry.copy()(*args, **kwargs)
def aws_available(self):
return self.available
def _tag_ebs(self, role):
def _tag_ebs(self, conn, role):
""" set tags, carrying the cluster name, instance role and instance id for the EBS storage """
if not self.available:
return False
tags = {'Name': 'spilo_' + self.cluster_name, 'Role': role, 'Instance': self.instance_id}
try:
conn = boto.ec2.connect_to_region(self.region)
volumes = conn.get_all_volumes(filters={'attachment.instance-id': self.instance_id})
conn.create_tags([v.id for v in volumes], tags)
except Exception as e:
logger.info('could not set tags for EBS storage devices attached: {}'.format(e))
return False
return True
volumes = conn.get_all_volumes(filters={'attachment.instance-id': self.instance_id})
conn.create_tags([v.id for v in volumes], tags)
def _tag_ec2(self, role):
def _tag_ec2(self, conn, role):
""" tag the current EC2 instance with a cluster role """
if not self.available:
return False
tags = {'Role': role}
try:
conn = boto.ec2.connect_to_region(self.region)
conn.create_tags([self.instance_id], tags)
except Exception as e:
logger.info("could not set tags for EC2 instance %s: %s", self.instance_id, e)
return False
return True
conn.create_tags([self.instance_id], tags)
def on_role_change(self, new_role):
ret = self._tag_ec2(new_role)
return self._tag_ebs(new_role) and ret
if not self.available:
return False
try:
conn = self.retry(boto.ec2.connect_to_region, self.region)
self.retry(self._tag_ec2, conn, new_role)
self.retry(self._tag_ebs, conn, new_role)
except RetryFailedError:
logger.warning("Unable to communicate to AWS "
"when setting tags for the EC2 instance {0} "
"and attached EBS volumes".format(self.instance_id))
return False
return True
def main():
+6 -15
View File
@@ -6,6 +6,7 @@ import unittest
from mock import Mock, patch
from collections import namedtuple
from patroni.scripts.aws import AWSConnection, main as _main
from patroni.utils import RetryFailedError
from requests.exceptions import RequestException
@@ -16,13 +17,13 @@ class MockEc2Connection(object):
def get_all_volumes(self, filters):
if self.error:
raise Exception("get_all_volumes")
raise boto.exception("get_all_volumes")
oid = namedtuple('Volume', 'id')
return [oid(id='a'), oid(id='b')]
def create_tags(self, objects, tags):
if self.error or len(objects) == 0:
raise Exception("create_tags")
raise boto.exception("create_tags")
return True
@@ -63,30 +64,20 @@ class TestAWSConnection(unittest.TestCase):
self.assertTrue(self.conn.aws_available())
def test_on_role_change(self):
self.assertTrue(self.conn._tag_ebs('master'))
self.assertTrue(self.conn._tag_ec2('master'))
self.assertTrue(self.conn.on_role_change('master'))
self.conn.retry = Mock(side_effect=RetryFailedError("retry failed"))
self.assertFalse(self.conn.on_role_change('master'))
def test_non_aws(self):
self.error = True
conn = AWSConnection('test')
self.assertFalse(conn.aws_available())
self.assertFalse(conn._tag_ebs('master'))
self.assertFalse(conn._tag_ec2('master'))
self.assertFalse(conn.on_role_change("master"))
def test_aws_bizare_response(self):
self.json_error = True
conn = AWSConnection('test')
self.assertFalse(conn.aws_available())
def test_aws_tag_ebs_error(self):
self.error = True
self.assertFalse(self.conn._tag_ebs("master"))
def test_aws_tag_ec2_error(self):
self.error = True
self.assertFalse(self.conn._tag_ec2("master"))
@patch('sys.exit', Mock())
def test_main(self):
self.assertIsNone(_main())