2 Resident instances. App was quiet for 10 minutes, no requests being served. 1 request sent. Neither resident instance took it. A cold instance was started without using /ah_/warmup. That user facing request languished for 30 seconds while GAE java spun up. There seems to be no way to prevent these sorts of catastrophes from happening.
App ID = lemurspot Instance ID = 00c61b117cb19763aeea814cbc78e80de2f437 On Friday, January 18, 2013 10:10:25 AM UTC-8, Carl Schroeder wrote: > > There needs to be a checkbox on the dashboard which says "never send user > request to cold instances". Then low traffic apps (or versions) with low > latency handlers will no longer be at the mercy of a load balancing > algorithm making WAGs in an information vacuum. After an app (or version) > has enough traffic for the instance scheduler to make educated guesses, > this box can be unchecked. > > I think that would fix everyone's issues with cold start latency on GAE > java. > > On Friday, January 18, 2013 9:22:15 AM UTC-8, Tom Phillips wrote: >> >> Yep, you need more min idles. At some min idle setting you'll see that >> the blue line no longer goes above green (minus unforeseen bursts, >> which need be you can prepare for with an even higher min idle >> setting). >> >> Once blue doesn't go above green, you'll see warmup request going to / >> _ah/warmup and not to your users. >> >> Basically, you don't get any delta between blue and green for free. >> You pay for it by having your users suffer loading requests. That's >> one reason why very low startup latency apps shine - they can take >> advantage of any delta. Higher startup latency apps (pretty much any >> real java app) are OK at steady state, but only if you are willing to >> pay to have a high enough min idle setting that you are billed for all >> running instances. >> >> /Tom >> >> On Jan 18, 12:07 pm, Francois Masurel <f.masu...@gmail.com> wrote: >> > Here is my dashboard instance graph. >> > >> > First part is with 1 min idle instance, second part with min idle >> instance >> > set to auto. >> > >> > < >> https://lh6.googleusercontent.com/-xVIJBwSd04o/UPmA5FfU4QI/AAAAAAAAz5...> >> >> > I have just set min idle instances to 2 and see how it goes. >> > >> > François >> > >> > >> > >> > >> > >> > >> > >> > On Friday, January 18, 2013 5:11:47 PM UTC+1, Tom Phillips wrote: >> > >> > > Hi Francois, >> > >> > > Don't set min idles to automatic, set it to a high enough number that >> > > over time Total instances (blue) == Billed instances (green). Leave >> > > max idles and the two latency settings at automatic. >> > >> > > If you do have min idles configured, can you post a screencap of your >> > > instances graph? >> > >> > > /Tom >> > >> > > On Jan 18, 8:45 am, Francois Masurel <f.masu...@gmail.com> wrote: >> > > > Things are getting worse : 20 loading requests for the last 23 >> minutes >> > > > (about 20% of all requests). >> > >> > > > Could I have an explanation ? >> > >> > > > What am I doing wrong ? I'm really lost :-( >> > >> > > > App Id : vncts1 >> > >> > > > On Friday, January 18, 2013 2:25:17 PM UTC+1, Francois Masurel >> wrote: >> > >> > > > > I have been testing full automatic mode for the last hours (all >> > > > > application settings set to automatic). >> > >> > > > > Still getting quite a few user-facing loading requests (~20 per >> hour, >> > > 5% >> > > > > of all requests), there is no more warmup requests in the logs. >> > >> > > > > One dynamic instance has been alive since the beginning of the >> test >> > > and is >> > > > > getting most of the traffic. >> > >> > > > > Are these numbers normal ? >> > >> > > > > François >> > >> > > > > App Id: vncts1 >> > >> > > > > On Thursday, January 17, 2013 11:21:16 PM UTC+1, Francois Masurel >> > > wrote: >> > >> > > > >> Hi Tom, >> > >> > > > >> Thanx for your suggestion, that's just the settings I'm testing >> at >> > > the >> > > > >> moment. >> > >> > > > >> Seems to reduce significantly user-facing requests but will >> quite >> > > > >> increase my bill (x4) :-( >> > >> > > > >> May be it will be worth it. >> > >> > > > >> I'll tell you how it goes. >> > >> > > > >> François >> > >> > > > >> On Thursday, January 17, 2013 11:11:56 PM UTC+1, Tom Phillips >> wrote: >> > >> > > > >>> Make sure on your instances graph that the blue "Total" line is >> not >> > > > >>> often, ideally only under unforeseen bursts, going higher than >> the >> > > > >>> green "Billed" line. >> > >> > > > >>> Total(blue) > Billed(green): you aren't charged for the delta >> > > between >> > > > >>> Total and Billed, but lower QOS for users (assuming your >> startup >> > > time >> > > > >>> is high) since they get most loading requests. >> > > > >>> Total(blue) == Billed(green): Your /_ah/warmup gets all loading >> > > > >>> requests, optimal QOS >> > >> > > > >>> This is just what I've observed (leaving max idles and min/max >> > > latency >> > > > >>> at automatic). Unless you have magically low startup latency, >> boost >> > > > >>> min idles until Total == Billed consistently. >> > >> > > > >>> /Tom >> > >> > > > >>> On Jan 17, 11:56 am, Francois Masurel <f.masu...@gmail.com> >> wrote: >> > > > >>> > For my low traffic website I have found that 50 new instances >> were >> > > > >>> started >> > > > >>> > during the last 2 hours and 22 minutes (taken from logs >> searching >> > > for >> > > > >>> "new >> > > > >>> > process"). >> > >> > > > >>> > Seems quite a lot for me as I already have a resident >> instance. >> > >> > > > >>> > But what annoys me the most is that, among those 50 loading >> > > requests, >> > > > >>> 20 >> > > > >>> > were user-facing requests (40%) and 5 were cron requests >> (10%). >> > >> > > > >>> > Is it normal behavior ? Thanx for your help. >> > >> > > > >>> > François >> > -- You received this message because you are subscribed to the Google Groups "Google App Engine" group. To view this discussion on the web visit https://groups.google.com/d/msg/google-appengine/-/MayYh4vJg5YJ. To post to this group, send email to google-appengine@googlegroups.com. To unsubscribe from this group, send email to google-appengine+unsubscr...@googlegroups.com. For more options, visit this group at http://groups.google.com/group/google-appengine?hl=en.
