[rfc] move activation to a helper process
david at fubar.dk
Tue Oct 17 07:10:14 PDT 2006
On Tue, 2006-10-17 at 00:15 -0400, David Zeuthen wrote:
> My plan for checking OOM handling right now includes extending the
> custom IPC protocol (but only on DBUS_BUILD_TESTS) to ask the helper
> about number of allocations (call it M) and instructing the helper to
> make allocation m=0,1,...,M fail just like in _dbus_test_oom_handling.
> Then the *_try_iterations functions in bus/dispatch.c will just go
> through all N + M combinations where N is the number of allocations in
> the bus process and M is the number of allocations in the helper
> process. Does this sound somewhat sane?
Replying to myself, and after getting some much needed sleep, I think
it's much better to rework the helper bits a bit so it can run in a
thread instead of a separate process. Shouldn't be much work at all.
So we probably want the former for the session bus anyway (to save
kernel resources by not forking) and the OOM handling will fall nicely
out of this as the two threads will share the same address space. Will
have to run some of the tests twice (one for helper in separate process,
one for helper in a thread) but that's OK I suppose.
More information about the dbus