Memory leaks require periodic restart
Status: Alpha
Brought to you by:
joehopfield
Menu
▾
▴
#16 Memory leaks require periodic restart
The plasmid agent seems to consume much more memory
than I would expect, and the process(es) under linux
continue to grow slowly as long as they live. I
believe this is due at least in part to objectstore's
cached objects. But I can't turn off cross-transaction
caching, because I've got a few places where the lame
odi semantics make things impossible to work with RMI
(like sending over a persistible instance, even though
it has never been persisted).
When we GET RID OF ODI/PSE and replace it with some
more reasonable and RMI friendly persistence
mechanism, this may go away.
Anyway, should run jprobe on the agent to verify my
theory.
Logged In: YES
user_id=4009
This is largely due to the use of deepfetch on the
objectstore objects (required so that RMI doesn't hit
invalid objects while returning persistibles).
My "fix" is just a workaround which restarts all engines
and threads periodically so that objectstore's gc can run.
This sucks. I wish I had something better, but obviously
the objectstore stuff was the wrong way to go. Next version
just use odbc with mysql or cloudbase or...
Anyway, with this workaround the server-agent has been up
for a couple of days (which didn't work before).