New Relic Java Agent HikariCP Pool Exhaustion Timestamp Granularity Limits

Answer to the three specific questions

1) Timestamp granularity – The New Relic Java agent does not emit a separate event with a sub‑second timestamp when a HikariCP pool reaches its maximum size. It samples the pool metrics at each harvest interval (default 60 s) and stores the aggregated Active, Idle, and Max values for that window. Therefore you can only see the exact moment of exhaustion if the pool stays at or above the limit for an entire harvest period.

2) Metric‑name change impact – Starting with Java agent 8.x the metric namespace shifted from Database/ConnectionPool/* to Datastore/ConnectionPool/*. Historical data collected with agents ≤ 7.x remains under the old prefix, while newer agents use the new prefix. To query both eras in a single NRQL statement you can use a wildcard, e.g.: FROM Metric SELECT max(*) WHERE name LIKE '%ConnectionPool/Active%'. If you query only the old prefix you will miss data from agents 8.x+, and vice‑versa.

3) NRQL alert reliability for sub‑minute spikes – An alert that evaluates max(Database/ConnectionPool/Active) (or the new Datastore/ConnectionPool/Active) over a 60‑second window can only trigger when the elevated value persists for at least one full harvest interval. Shorter bursts that last less than the harvest period are averaged with lower values and may never cross the alert threshold. To detect sub‑minute exhaustion you must either reduce the harvest interval (e.g., to 10 s) or rely on application‑level logging.

Steps to verify or adjust behavior for your environment

1. Ensure pool monitoring is enabled:

   connection_pool_monitoring.enabled=true

   in newrelic.yml.
2. If you need finer granularity, lower the harvest interval (affects all metrics):

   common.structure.metrics.enabled=true
   common.structure.metrics.interval=10

   (values in seconds). Restart the Java agent after changing.
3. Test with a controlled load that exceeds maximumPoolSize for a known short duration (e.g., 15 s). After the test, run an NRQL query like:

   SELECT max(active) FROM Metric WHERE name LIKE '%ConnectionPool/Active%' TIMESERIES 10 seconds

   and observe whether the spike appears in the returned series.
4. To confirm the metric‑name shift, compare two agents: one running 7.x and another 8.x+. In the Metrics UI search for Database/ConnectionPool/Active and Datastore/ConnectionPool/Active; you should see data only from the respective agent versions unless you use a wildcard.
5. If you disable pool monitoring (connection_pool_monitoring.enabled=false), verify that no pool‑related metrics appear in the UI or NRQL (SELECT * FROM Metric WHERE name LIKE '%ConnectionPool%' returns empty).

Note: Changing the harvest interval increases metric volume and may affect ingestion costs; adjust only after evaluating the trade‑off.