It turns out that I need to be able to control that. Some of my apps are generating high number of concurrent requests per user, so if only few clients connects to previously idle app, they can spawn a lot of new workers and it's better to keep them longer. In any other case I want to cheap those workers faster, so another setting is needed. Patch attached.
2012/9/5 Łukasz Mierzwa <[email protected]>: > I've made few tweaks that allows to cheap workers quicker if they were > spawned because backlog queue was to high. > I've noticed that sometimes (mostly during app upgrades) one node gets > hit with big load spike and it spawns few emergency workers, load goes > away quickly, and it takes a lot of time to cheap those extra workers. > >> * NEW PLUGIN: cheaper_busyness (by Łukasz Mierzwa) >> >> This is an extremely advanced cheaper algorithm (cheaper algorithms are >> used for adaptive process spawning) targeted at big deployments where you >> want always the maximum possibile performances while having the minimum >> amount of required workers. >> >> We are lucky, as Łukasz wrote documentation for it: >> >> http://projects.unbit.it/uwsgi/wiki/CheaperBusyness > > -- > Łukasz Mierzwa -- Łukasz Mierzwa
busyness_emergency_v2.diff
Description: Binary data
_______________________________________________ uWSGI mailing list [email protected] http://lists.unbit.it/cgi-bin/mailman/listinfo/uwsgi
