Bill Farner created AURORA-848:
----------------------------------
Summary: Handle resourceOffers callback asynchronously
Key: AURORA-848
URL: https://issues.apache.org/jira/browse/AURORA-848
Project: Aurora
Issue Type: Story
Components: Scheduler
Reporter: Bill Farner
Assignee: Zameer Manji
Priority: Critical
Since the mesos driver synchronizes all callbacks, it's very important to
return as quickly as possible to avoid blocking other callbacks. In the case
of {{resourceOffers}}, we start by calling {{storage.write}}, which will stall
if the write lock is held. In very large, busy clusters, we've observed
{{resourceOffers}} routinely taking > 1 second, which can lead to delays in
receiving status updates.
I suggest the entire body of {{resourceOffers}} be asynchronous to prevent it
from backing up status updates.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
