Alexander Rukletsov created MESOS-8082:
------------------------------------------
Summary: updateAvailable races with a periodic allocation and
leads to flaky tests.
Key: MESOS-8082
URL: https://issues.apache.org/jira/browse/MESOS-8082
Project: Mesos
Issue Type: Bug
Components: test
Reporter: Alexander Rukletsov
Assignee: Alexander Rukletsov
When an operator requests a resource modification (reserve resources, create a
persitent volume and so on), a corresponding endpoint handler can request
allocator state modification twice: recover resources from rescinded offers and
for update applied operation. These operations should happen atomically, i.e.,
no other allocator change can happen in-between. This is however not the case:
a periodic allocation can kick in. Solutions to this race might be: moving
offer management to the allocator, coupling operations in the allocator,
pausing allocator.
While this race does not necessarily lead to bugs in production—as long as
operators and tooling can handle failures and retry—, it makes some tests using
resource modification flaky, because in tests we do not plan for failures and
retries.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
