A monitoring table you can read is a report. A monitoring table you can act on is an operations console. This release makes every resource in the observability catalogue declare its actions, and routes every way of invoking them through one service.
Actions
- Declared per resource A pool supports quiesce, online, close and inspect; a connection cancel, kill, close and inspect; a statement cancel, close and inspect; a database or server online, maintenance, quiesce, offline, close and inspect.
- The missing level Deleting a statement row cancels the work in flight and closes the statement, leaving the connection open. It was the one level with no handler.
- One service, one audit All eight catalogue triggers and every JMX operation route through the same service, so preconditions are shared and the audit names who asked, whichever door they used.
JMX and rates
- Beans derived, not written One dynamic bean derives its operations from the resource's declared actions at the moment a console asks, and its lifetime tracks the resource. Connections and statements are deliberately not published as beans, since a busy node has thousands.
- Rates Exponentially weighted one-minute, five-minute and one-hour rates on garbage collection, class loading, allocation and executor outcomes, from a five-second sampler, with process and system CPU load.
The session and script tiers of September stand on this: a row you can delete is only safe when the delete runs through the same gate as everything else.