Resource graph shows that Hikari connection pool is leaking ~1 thread every day. Eventually, it will hit the roof and remain there when you leave opennms running continuously long enough. At that point, opennms GUI will be unresponsive and the performance will be extremely degraded to the extend that it’s not functional. A full service restart is required to restore the service. This is seen in Meridian and Horizon. This was reported by several users.

Horizon 32.0.6

Open

The sawtooth-like trend corresponds to restarts with which the threads state is restored.

Meridian 2023.1.9

Open

Here it was left running long enough with a restart. The active threads continued to climb up to the roof.

They have one thing in common although I cannot attest this is the root cause. They all have JDBC collector configured to collect PostgreSQL stats.

In the case of Meridian when it’s completely degraded, the connections' state retrieved from pg_stat_activity table showed there are exactly 50 lingering idle queries and they are identical. See attached xslx file for full dump of the table. Most of them are old connection - more than a day old. I would expect the setting the idleTimeoutin the datasource config would remove the old connections but it doesn’t. In this case, both instances are configured with default value of 600s. This deserves an investigation as well.