Deliver all alart notification to handle overlapping alarms

connection.blocked requires us to track resources we are
conserving. This means the old logic of determining
edge state transitions for alarms does not work any more.

Instead of using the old strategy of comparing alarmed
node collection sizes, instead of pass around what event
the notification is for and simply deliver it to the
relevant nodes.

This requires that rabbit_alarm event consumers
handle duplicate notifications. They already do so
after earlier changes on branch bug25191.

This makes connection unblocking work correctly
in the following sequence of events:

 * memory alarm set for node A
 * disk alarm set for node A
 * memory alarm cleared for node A
 * disk alarm cleared for node A

as well as other similar scenarios with overlapping
alarms.

This slighly increases internode and intranode message
traffic of alarm notifications. Since alarms occur rarely
in well-monitored systems, this is a reasonable
trade-off.

0 files changed, 0 insertions, 0 deletions