I waited with my response because I didn't want to jump to conclusions. Yes, I have been experimenting with some settings every now and then, limiting instances, playing with the pending sliders etc. But all it did was change how quickly instances were created or collected (while always increasing latency). It didn't change how the billed instances were calculated.
Above is the chart of my applications' instances for the past month. (The editor didn't allow me to place it at the right location...) It's rather easy to spot when things changed. We used to get billed for roughly one instance on average. Yes, we did have more than one running, but it was (and is!) always one instances that handles 95% to 99% of the load. So it seemed only just that Google would only charge for the main instances. I never asked for an instance that just sits there, but I didn't mind it while it was free :)
So now being charged for two or three instances, when only one is really doing anything, seems like a major change that should be documented. Okay, maybe it's just our application, but our pricing has increased steeply. I just posted another message about the frontend hour calculation. Both issues combined seem to have led to a pricing increase from $5 to $10 before, to now $30 to $50. Our request per second have increased moderately, we may have made some requests slower, we might have slightly different usage patterns. But I cannot see a reason for a price increase this steep.
Any help or insight would be appreciated. Our app ID is small-improvements-hrd
Kind regards, Per
On Friday, April 27, 2012 7:02:26 AM UTC+2, Takashi Matsuo (Google) wrote:
On Fri, Apr 27, 2012 at 10:16 AM, Per <per.fr...@gmail.com> wrote:
> Hi team,
> previously, the scheduler used to spin up 3 instances for our
> application, only used 1 out of the 3, but at least we didn't have to
> pay for the unused instances. There was always a big gap between total
> instances and billed instances.
Wasn't it because that you set Max Idle Instances at that time?
If you set Max Idle Instances to Automatic, the total instances and
billed instances should be the same.
> However, as of yesterday, the scheduler continues to spin up 3
> instances on average, but we have to pay for all of them. We're at
> maybe 1 request per second on average, and it's all handled just fine
> by one instance. We're on F4, so having to pay for 2 mostly unused
> instances hurts.
> Since this seems to coincide with the 1.6.5 release, I'm tempted to
> think that there's something changed in the background, and it would
> be great to learn more about the new suggested way.
> When we tried to limit idle instances last time, all we got was
> instance churn and bad latency. So we're on automatic/automatic these
> days, and I'd prefer not to have to experiment again. Some advice
> would be great.
Again, I'm wondering when you tried to limit the idle instances.
Generally speaking, you can not get everything. If you want a better
performance, you'll need to pay more, and vice versa.
However, if you're really certain that the behavior of our scheduler
has significantly changed, please let me know your app-id and detailed
explanation(it can be off-list), so I can look into it further.