ceph rgw: increase retry attempts post replication
Sometimes the slave RGW takes a little time to re-sync periods and data from the new master RGW after a promotion. Increase the number of attempts to retrieve the data written to the master RGW from the slave RGW units.
This commit is contained in:
@@ -632,7 +632,7 @@ class CephRGWTest(test_utils.OpenStackBaseTest):
|
||||
)
|
||||
|
||||
@tenacity.retry(wait=tenacity.wait_exponential(multiplier=1, max=60),
|
||||
reraise=True, stop=tenacity.stop_after_attempt(8))
|
||||
reraise=True, stop=tenacity.stop_after_attempt(12))
|
||||
def _target_get_object():
|
||||
return target_client.get_object(_container, 'testfile')
|
||||
_, target_content = _target_get_object()
|
||||
@@ -687,7 +687,7 @@ class CephRGWTest(test_utils.OpenStackBaseTest):
|
||||
)
|
||||
|
||||
@tenacity.retry(wait=tenacity.wait_exponential(multiplier=1, max=60),
|
||||
reraise=True, stop=tenacity.stop_after_attempt(8))
|
||||
reraise=True, stop=tenacity.stop_after_attempt(12))
|
||||
def _source_get_object():
|
||||
return source_client.get_object(_container, 'testfile')
|
||||
_, source_content = _source_get_object()
|
||||
|
||||
Reference in New Issue
Block a user