Hello Nils, Thanks for your reply. Now I understand better the definition of this parameter (cache efficiency vs response time).
I've checked and I am really graphing the size of the ban list (using the value "MAIN.bans" from varnishstat ). This value is the same that the "varnishadm ban.list | wc -l" output. I've just see that there is a "new" value in varnishstat "MAIN.bans_lurker_obj_killed_cutoff". I will add this value to my monitoring plugin. Olivier 2017-08-30 14:37 GMT+02:00 Nils Goroll <[email protected]>: > Hi Olivier, > > I'm responding to the last two emails from you in one go > > On 30/08/17 08:47, Olivier Hanesse wrote: > > What will happen when the ban list hits the size of bans defined in > ban_cutoff > > value ? > > The ban lurker still works the list of bans as before, but when having > reached > the <ban_cutoff>th ban, we kill all objects hanging off these bans without > testing the ban condition. > > This way, actively used objects (which get tested against the ban list at > request time and will end up hanging off some ban near the top of the ban > list) > will not get killed, but rather only those which were least frequently > accessed > (iow the long tail). > > On 30/08/17 11:44, Olivier Hanesse wrote: > > Last night after your reply, I put a ban_cutoff value of 18500 according > to > > the definition (50ms of latency, 370K/s ban.lurker.tested) (I've > restarted > > varnish, "varnishadm param.show ban_cutoff" shows the right value) > > > > This morning, nothing has changed : ban lists is increasing (well over > > 18500). > > One obvious explanation would be that the lurker had not got to the cutoff > value. > > But I wonder what exactly you are measuring here. In your first email you > wrote > > On 29/08/17 18:19, Olivier Hanesse wrote: > > our ban list keeps increasing to reach 100K objects (and sometimes more). > > This makes me guess that maybe you'd be graphing the number of objects > hanging > off the bans. Quick reminder: > > * the second column in the varnishadm ban.list output is the number of > objects associated with this ban (objects, for which this ban has > last been tested) > > * what the ban_cutoff parameter is limiting is the number of bans > (that would be varnishadm ban.list | wc -l minus 2) > > So can you please double check that you are graphing the latter and not the > former for ban.list? > > If you'd actually be graphing the former, then we don't have a problem as > this > will just be the total number of objects in your cache. > > Thanks, Nils > >
_______________________________________________ varnish-misc mailing list [email protected] https://www.varnish-cache.org/lists/mailman/listinfo/varnish-misc
