The client seems to eat more and more memory over time. Anyone else noticed that?
same here, i have to write a small script to restart nxt after it crash
yeah, something's got to be done about this.
Although RAM is cheap and getting cheaper, some leaking is going on for sure

it's not a problem for home users who just launch, make some payments and close, but for public nodes it is an issue.
I never saw Java memory issues with my node, and it served several tens of GB of traffic (public static IP; hallmarked). It did get stuck on some occasions (running and serving requests, but not on the latest block even after browser refresh), and a restart usually fixed that, although sometimes it required reverting to a backup blockchain.
I'd like to look at any Java memory issues you see. If you consistently see out of memory exceptions, please run the client with the following parameters:
-verbose:gc -Xloggc:gc.log -XX:+PrintGCDateStamps -XX:+PrintGCDetails
And then share the resulting gc.log file. NB: this gets overwritten on start, so take a copy first.
You can also run it with this parameter:
-XX:+HeapDumpOnOutOfMemoryError
This will produce a large file which is basically a copy of the entire heap at the point, so it's best to do it on a node with moderate memory settings (e.g. 512MB, not 4GB) and make sure you have sufficient disk space. Zip/gzip this up and upload it somewhere (e.g. mega) and share details. NB: very important, if you do this DO NOT enter your secret key on that node as it is likely to be visible in the heap dump (hallmarks are okay).