Gentle shutdown feature
Shawn Routhier
sar at isc.org
Tue Jul 1 20:40:48 UTC 2014
On Jun 27, 2014, at 8:35 AM, John Wobus wrote:
> Here's our situation:
> We run DHCP-4.1-ESV R8 and we stop and restart the
> server every 2 minutes to reconfigure. We have about
> 200,000 addresses in dynamic pools and our log
> grows to 20 million plus lines a day.
>
> There are other similar sites that use the same "stop and
> start frequently" strategy. I would naturally be concerned
> if the daemons had trouble ever reaching normal state.
>
> I am surprised at two things:
> (1) A new feature was put into the ESV version?
> I'm wondering why such a thing exists if it is
> handled like that. Was this considered a bug fix?
Basically - it was on the boundary between a bug and
a new feature. We didn't realize how disruptive it could
be in some cases.
>
> (2) I always thought SIGTERM caused the server
> to finish writing the lease file. I noticed
> significant delays (tenths of a second)
> in dhcpd "term" time a decade ago and attributed
> it to lease file writing. I'm surprised to hear
> otherwise.
>
> John Wobus
> Cornell University IT
>
> On Jun 26, 2014, at 2:13 PM, Shawn Routhier wrote:
>
>> In the 4.3.0 release cycle we added a feature to gently shutdown
>> clients and servers when they received SIGINT and SIGTERM
>> signals. This feature was included in the changes back ported to
>> 4.1-ESV and 4.2 and has two side effects that may affect users.
>> This was done to allow any writes to the lease files to complete
>> before the process stops.
>>
>> When a client receives a SIGINT or SIGTERM signal the new
>> behavior is to send a release message to the server.
>>
>> When a server receives a SIGINT or SIGTERM signal "if" failover
>> is in use then when the server is restarted it may go through recover
>> wait for a number of seconds (determined by the server's MCLT
>> setting) before becoming usable.
>>
>>
>> In both of these cases the previous functionality can be gotten
>> by halting the process with a different signal, for example
>> "kill -9 <pid>"
>>
>> We have received a report objecting to this feature
>> and we are determining the best course of action.
>>
>> 1) Are you already making use of this feature? Or would it
>> not be an issue if we disabled it?
>>
>> 2) Has this feature caused issues for you?
>>
>> 3) Any other useful comments about this issue?
>>
>> Shawn Routhier
>> ISC DHCP
>>
>> _______________________________________________
>> dhcp-users mailing list
>> dhcp-users at lists.isc.org
>> https://lists.isc.org/mailman/listinfo/dhcp-users
>
> _______________________________________________
> dhcp-users mailing list
> dhcp-users at lists.isc.org
> https://lists.isc.org/mailman/listinfo/dhcp-users
More information about the dhcp-users
mailing list