On a single-node Carbon SR1 setup, we got an OOM over the weekend. Looking at the heap dump shows 1.1GB occupied in the frontend history's closed transactions map, with 33M entries. Analysing the dump further reveals that there's a bug in the controller's transaction handling, which results in transactions being held in memory forever — which guarantees an OOM eventually, regardless of the JVM settings.
The heap dump is on https://www.sk2.org/java_pid2098.hprof.xz (269MB).
This has been fixed in Carbon upstream; it will be part of SR2.
Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.
For information on the advisory, and where to find the updated
files, follow the link below.
If the solution does not work for you, open a new bug report.