dhcpcd-discuss

If you run dhcpcd under valgrind, is there a large difference between the number of mallocs made?

On 5 March 2019 03:56:50 GMT, Donald Sharp <sharpd@xxxxxxxxxxxxxxxxxxx> wrote:
>BTW -> I see no difference between using -DRT_FREE_ROUTE_TABLE and not
>using it.
>
>donald
>
>On Mon, Mar 4, 2019 at 10:48 PM Donald Sharp
><sharpd@xxxxxxxxxxxxxxxxxxx> wrote:
>>
>> 1 million routes install/removal:
>> sharpd@janelle ~/dhcpcd> ps -eo rss,pid,euser,args:100 | grep dhcpcd
>>  1892  8413 root     src/dhcpcd
>>   656  8468 sharpd   grep --color=auto dhcpcd
>>
>> This looks awesome, repeated installs/removals and we are at the same
>> level of memory consumption.
>>
>> donald
>>
>> On Mon, Mar 4, 2019 at 7:47 PM Roy Marples <roy@xxxxxxxxxxxx> wrote:
>> >
>> > On 04/03/2019 21:43, Roy Marples wrote:
>> > > Lowering memory usage more we can do this - move if_initrt to the
>top of
>> > > rt_build. This is tricky as from memory some parts of dhcpcd rely
>on it
>> > > being there outside of route building. We'll need to double check
>and
>> > > fix these places.
>> > > Once that's done we can then dispose of kroutes at the end of
>rt_build.
>> > > If that works, kroutes can be removed from the global context.
>> > >
>> > > But that's for another day.
>> >
>> > Well, I was wondering what really needs kroutes around and it turns
>out
>> > I think it can be removed safely.
>> > I've created a new branch, rbtree-memory with the prior patch in
>and the
>> > above done.
>> > It seems to work and valgrind doesn't report any leaks.
>> >
>> > How is the memory usage looking now with your million routes?
>> > If it's lower, could you add -DRT_FREE_ROUTE_TABLE to your
>CPPFLAGS,
>> > clean and compile, and see how that then affects matters?
>> >
>> > Roy

-- 
Sent from my Android device with K-9 Mail. Please excuse my brevity.