On Fri, Apr 20, 2018 at 07:23:37PM +0300, Slawa Olhovchenkov wrote:
> On Fri, Apr 20, 2018 at 05:33:34PM +0300, Slawa Olhovchenkov wrote:
>
> > On Fri, Apr 20, 2018 at 04:23:00PM +0200, Willy Tarreau wrote:
> >
> > > > > Thus for you it's better to stick to a single listener, and if you
> > > > > want to
> > > > > increase the fairness between the sockets, you can reduce
> > > > > tune.maxaccept in
> > > > > the global section like below :
> > > > >
> > > > > global
> > > > > tune.maxaccept 8
> > > >
> > > > No significant differenece: average load rised, unequal CPU load still
> > > > at same level
> > >
> > > Then you can try setting it even lower (1) and decreasing maxpollevents :
> > >
> > > global
> > > tune.maxaccept 1
> > > tune.maxpollevents 20
> >
> > No difference, disbalance about same.
> >
>
> Can I got per-thread stats like connections count, bytes transfered?
Not by default, as such information are per-socket, per-frontend etc.
However, by having on listening socket per thread, and "option socket-stats"
you would have this. But in your case it doesn't work since only one socket
receives all the traffic in any case.
The most detailed per-thread info you can get in your situation is by issuing
"show activity" on the CLI, you will then get the information of the thread
that processes the request :
thread_id: 0
date_now: 1524244468.204596
loops: 754
wake_cache: 576
wake_tasks: 0
wake_applets: 14
wake_signal: 0
poll_exp: 590
poll_drop: 9
poll_dead: 0
poll_skip: 0
fd_skip: 0
fd_lock: 0
fd_del: 109905
conn_dead: 0
stream: 109922
empty_rq: 170
long_rq: 414
These information are quite internal, but the following ones will be
useful :
- loops : number of calls to the event loop
- stream : cumulated number of streams processed
(~= connections for HTTP/1, ~= requests for HTTP/2)
- long_rq : number of passes via the scheduler where the runqueue had
more than 200 tasks active
- empty_rq : number of passes via the scheduler where the runqueue was
empty
The rest is probably irrelevant in your case. By doing it a few times
you'll iterate over all threads. Or you can even have multiple stats
sockets as well, but no need to do complex things first.
Willy
