How much of the platform an installation uses is a question for capacity planning and for licensing alike. This release measures it inside the server, at a cost small enough to leave on, and keeps the result as ordinary rows an administrator can query.
What is measured
- Connection time Each connection pool sums the time its connections were open and the time each lease was busy, plus how often and how long a caller waited for one. Its counters moved to 64-bit so they no longer overflow on a busy node.
- Statements and commits Per database server and logical database, from the totals the pool already keeps.
- Threads Platform and virtual threads alive, started and running a task, counted per component with one atomic add per event and no lookups.
How it is kept
- Collected every five seconds Gauges are summed with their peak, counters as increments; once its rows exist a tick allocates nothing.
- Written every five minutes The server adds what it counted into the hour's rows in one transaction. A figure not measured is never written as zero, and a write that fails is handed back to the collector whole.
- Started at boot The writer starts with the application server, and the usage tables are writable by the manager only.
Where it is read
- Licensing, Usage Database threads by month and by hour, a row per application server over its own time up, and JVM threads.
- Nodes Each application server with the processors and memory it reported at its last start.
The reports are SQL over sums and maximums only, so the same figures can be checked by hand.