Re: [Nbd] [PATCH] nbd-server: add -D/-dont-daemonize option
- To: firstname.lastname@example.org
- Subject: Re: [Nbd] [PATCH] nbd-server: add -D/-dont-daemonize option
- From: Goswin von Brederlow <goswin-v-b@...186...>
- Date: Tue, 26 Jun 2012 11:44:49 +0200
- Message-id: <87ipeez9ta.fsf@...860...>
- In-reply-to: <B8A6F0F1B1B6592D21C27562@...873...> (Alex Bligh's message of "Mon, 25 Jun 2012 17:29:45 +0100")
- References: <20120624230358.GA28298@...1147...> <87pq8n1zlb.fsf@...860...> <B8A6F0F1B1B6592D21C27562@...873...>
Alex Bligh <alex@...872...> writes:
> --On 25 June 2012 11:59:44 +0200 Goswin von Brederlow
> <goswin-v-b@...186...> wrote:
>> I think the difference between -D and -d should be made clearer.
>> With -d the server runs in the foreground and accepts a single
>> connection only and outputs debug infos, right?
>> With -D the server runs in the foreground but forks a child for every
>> connect, right?
> Agree re documentation point - must admit it had me confused, but ...
>> What I would like to see is socket activation in nbd-server. That way
>> the service can be started as needed without performance loss (other
>> than the initial startup time once).
> Doesn't this only affect connect time, which given the connections are
> very long lasting isn't of vast importance? In any case, the fact the patch
> doesn't do X extra feature would not be a reason to reject it (not sure
> you were suggesting that).
> Alex Bligh
No, I was just wishing for a further patch.
The time saving of socket activation might not mean much but there is
another benefit. The race condition between the service being asked to
start and actualy being ready to accept incoming connections disapears.
And there is no longer a need to specify wether a service (e.g. kvm/xen
instances being started) depends on nbd-server being started or not.