Sam Corbett created BROOKLYN-207:
------------------------------------
Summary: Cluster policy robustness
Key: BROOKLYN-207
URL: https://issues.apache.org/jira/browse/BROOKLYN-207
Project: Brooklyn
Issue Type: Improvement
Reporter: Sam Corbett
Policies applied to clusters and groups must take the state of their members
into account. If they do not they risk taking decisions based on stale data.
See https://github.com/apache/incubator-brooklyn/pull/1109 for more context.
@aledsage said:
{quote}
I think that long-term the right philosophy is that the service-down should
trigger the removal of this node from the calculations. Until that point, we
don't have enough info to know if it's still working away at its previous load
(in which case we don't want to zero the requests-per-second by publishing the
same total-request-count again), or if it is not doing anything.
If it were genuinely not doing anything, then we should mark it as service-down.
This is obviously a fiddly thing to get right, particularly as there are
several failure scenarios and there are different ways the metrics are being
used.
{quote}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
