[ntp:questions] Proposal: Change the default memlock rlimit to 0
stenn at ntp.org
Sun Mar 15 22:19:50 UTC 2015
If we can RLIMIT_MEMLOCK on a machine we will do so by default. For
most every platform out there, a 32M limit is sufficient (and I think it
used to be smaller). For some Linux admins, they are discovering that
sometimes values as high as 128M are insufficient.
We know better timekeeping happens when ntpd can quickly respond, so by
default we lock ntpd into memory.
This is a good news/bad news tradeoff.
Here's a partial list of ways to address it.
- default it to 0, and force every admin to set it to something if they
want this. This will cause a Lot of work for people.
- leave it alone, and document that if they get these crashes (I don't
think we can trap these crashes but if we can we could offer a helpful
message) they should bump their 'rlimit memlock XX' up or set it to 0.
- add more logic to configure, so if we detect that we're running on an
OS that is prone to these problems we select a bigger default
As I said, there are pros/cons to all of these choices.
Also see http://support.ntp.org/bin/view/Support/LockingNtpdInMemory
Harlan Stenn <stenn at ntp.org>
http://networktimefoundation.org - be a member!
More information about the questions