Hi! On Mon, Jan 03, 2022 at 06:47:04PM +0500, Sohaib Ahmad wrote: > Hi, > > While debugging some timeouts I noticed a very high total time when request > is being served from haproxy's local cache. > > 10.0.17.137:45444 [03/Jan/2022:12:38:46.991] web webnode/<CACHE> > 0/0/0/0/15263 200 308984 - - LR-- 118/118/74/0/0 0/0 "GET /v1/package > HTTP/1.1" > > At the same time there are large responses with total time of few > milliseconds like: > > 10.0.17.39:51256 [03/Jan/2022:12:38:47.144] web webnode/<CACHE> 0/0/0/0/22 > 200 588672 - - LR-- 112/112/83/0/0 0/0 "GET /v1/package HTTP/1.1" > > Any idea what could cause these high total time values when every other > time is 0?
In theory this could be implied by the client's download time. Usually when your system is under load, it's very possible that 300kB will not fit into system buffers, and that client download time can cause this. What is important here is to figure if this only applies to the same client (or same few clients) or if it happens to a specific set of URLs for example. In H2 this could happen when the client downloads many objects at once, because even if we try hard to be fair with all clients, that doesn't magically improve the total bandwidth ;-) As such I think it's important to really figure if it's only/all the cache that does this, or if it's only some clients. Regards, Willy
