Incorrect resource alarm handling in RabbitMQ

[Alexey Lebedeff]

There are a couple RabbitMQ bugs that can lead to:

- Blocking publishing throughout whole cluster, if node in resource starvation mode is brutally killed ( `kill -9' from ocf script, OOM killer) - https://github.com/rabbitmq/rabbitmq-server/issues/362
- Unblocking publishing too early, when there is still nodes in resource starvation mode - it will be the last straw to break the starved node's back - https://github.com/rabbitmq/rabbitmq-server/issues/379

With current code the only fix for both issues is restart of all rabbitmq nodes.

There is a proposed upstream fix for this bugs, which probably will be included in RabbitMQ 3.6.0.

[Davanum Srinivas (DIMS) (dims-v)]

Alexey, Can you please add a pointer to the upstream fix here?

[Davanum Srinivas (DIMS) (dims-v)]

Ah, never mind the issues themselves have link to the pull requests with fixes

[Dmitry Mescheryakov (dmitrymex)]

Moving to 9.0 since it is a medium bug and we've already passed SCF

[Alexey Lebedeff (alebedev-a)]

Fixed in upstream rabbit 3.6.1

[Dmitry Mescheryakov (dmitrymex)]

Alexey's fixes were consumed in 9.0 during upgrade of RabbitMQ to 3.6.1

[Alexey Galkin (agalkin)]

Verification on:
fuel_build_id - 97
fuel_release - 9.0
RabbitMQ:
{rabbitmq_management,"RabbitMQ Management Console","3.6.1"}
,
{amqp_client,"RabbitMQ AMQP Client","3.6.1"}
,
{rabbitmq_management_agent,"RabbitMQ Management Agent","3.6.1"}
,
{rabbit,"RabbitMQ","3.6.1"}
,
{rabbitmq_web_dispatch,"RabbitMQ Web Dispatcher","3.6.1"}