Just switched back to automatic scaling and everything seems to be back to normal, no more crazy instance loading.
There were definitely something buggy on GAE side which seems to have been fixed. On Friday, October 11, 2013 1:23:10 PM UTC+2, Francois Masurel wrote: > > Just missed a point : basic scaling instances are still billed like > backend instance hours and not front end hours even for the default module. > > So we have only 8 hours offered instead of 28 hours for automatic scaling > instances :-( > > Please Google fix this, quick. > > Created an issue for it : > https://code.google.com/p/googleappengine/issues/detail?id=10105 > > François > > > > On Wednesday, October 9, 2013 5:35:25 PM UTC+2, Francois Masurel wrote: >> >> Ok guys, >> >> Looks like I solved the problem by setting *basic scaling* instead of >> *automatic >> scaling *for my default module, it works also with *manual scaling*. >> >> No more crazy instance starts and stops, that's cool :-) >> >> There don't seem to be the 10 concurrent requests maxi limit any more. >> >> Still have one question : how does it scale ? Here is what is said in >> the documentation : >> >> *A basic scaling module version is configured with a maximum number of >> instances using the max_instances basic scaling parameter.* >> >> *The number of live instances scales with the processing volume.* >> >> >> What is "processing volume" exactly? How is it evaluated? There is no >> pending latency or such parameters. >> >> I could start a few more instances by doing some stress tests. Started >> instance are kept alive as defined by the idle_timeout basic scaling >> parameter. But there still seem to have some user facing loading requests. >> >> I guess the automatic scaling mode still needs polish and is not >> production ready yet, unless I missed something. >> >> François >> >> >> On Wednesday, October 9, 2013 2:38:50 PM UTC+2, Francois Masurel wrote: >>> >>> Please, STOP killing instances after 30s of inactivity (average time >>> measured, might be even less). >>> >>> Aren't we paying a minimum 15 minutes startup fee for those started and >>> killed instances ? May be you could increase this delay to at least 2 >>> minutes ? >>> >>> These {censored} loading requests make our app totally unusable. >>> >>> >>> >>> On Wednesday, October 9, 2013 12:33:52 PM UTC+2, Francois Masurel wrote: >>>> >>>> Same problem with F2 instances. This f... scheduler is getting us >>>> crazy. >>>> >>>> There is nothing wrong with starting a new instance but it should wait >>>> for the instance being properly loaded before redirecting traffic to it. >>>> >>>> >>>> >>>> On Wednesday, October 9, 2013 12:09:42 PM UTC+2, Francois Masurel wrote: >>>>> >>>>> It looks like GAE Java instances don't handle more than 10 concurrent >>>>> requests. >>>>> >>>>> As a result, our GWT app home page which makes about 20 requests needs >>>>> at least 2 instances loaded to get displayed. >>>>> >>>>> And as instances are shutdown very agressivelly (a few seconds), our >>>>> users are getting hit by these loading requests pretty often. >>>>> >>>>> This is making our app totally useless (~10-20s to load a new >>>>> instance). >>>>> >>>>> Is there a way to increase this max number of concurrent requests ? >>>>> >>>>> François >>>>> >>>>> >>>>> >>>>> >>>>> -- You received this message because you are subscribed to the Google Groups "Google App Engine" group. To unsubscribe from this group and stop receiving emails from it, send an email to google-appengine+unsubscr...@googlegroups.com. To post to this group, send email to google-appengine@googlegroups.com. Visit this group at http://groups.google.com/group/google-appengine. For more options, visit https://groups.google.com/groups/opt_out.
