[nginx] unnormal pid file

[Maxim Dounin]

Hello!

On Sat, Mar 18, 2017 at 03:43:36PM +0800, 洪志道 wrote:

> Hi.
> 
> It's easy to reproduce such problem that
> the master/worker process are running without the pid file.
> 
> Step by step:
> 
> 1. Start master process.
>     > ./objs/nginx
>     nginx.pid successfully create with content (23234)
> 
> 2. Create an active connection, purpose for prevent previous master exit.
>    > telnet 127.1 80
> 
> 3. Quit the previous master, and it still alive for a while.
>    > kill -QUIT 23234
> 
> 4. Start new master process 23361, and the content of nginx.pid is changed
> to 23361
>    > ./objs/nginx
>    It successfully start because of the listening port is closed by the
> step 3.
> 
> 5. Wait for a while, and the nginx.pid is deleted while the old master
> process (23234) completely quit.
> 
> Now the master process (23361) is running without nginx.pid.
> 
> So, there are two key points:
> 1. Master start, I think there is not problem.
> 2. Master Exit, is it better to add the pid check with the pid file?
> 
> Anyway, I think we should guarantee the pid file associates with the
> running process.

By starting an additional nginx instance before the previous nginx 
has been exited you are looking for trouble anyway.  Starting 
additional instance can easily succeed even without using SIGQUIT 
to close listening sockets.  For example, nginx will happily start 
if you've changed listening socket to a different one, or use 
'listen .. reuseport'.

Instead, you should check if there is previous instance of nginx 
running, and don't try to start another one while it is running 
(and this is what init scripts usually do).  Any other approach is 
very likely to result in various problems.

Adding bandaids into nginx to mitigate most obvious of the 
resulting problems is possible, but I don't really think it's 
something needed.

-- 
Maxim Dounin
http://nginx.org/