The above post notwithstanding, the original problem I was having - which 
caused me to try to set min instances to 1 to begin with - seems to have 
gone away.  I know this is probably going to fall into the PEBCAK bucket, 
but I really don't think I did anything in particular to cause the problem, 
so I will explain in somewhat tedious detail the circumstances surrounding 
the problem, even if some of them aren't relevant.

The original problem was that the client-side latency of requests was 
intermittently very high.  I don't have many details on this because as 
soon as I tried to diagnose it I ran into the problem outlined above, which 
certainly confused my investigation.

However, now that I realize that the min instance slider is basically 
useless to me, I went back to testing against the auto/auto settings which 
I was using to begin with.  I then noticed that the appstats endpoint was 
getting hit at times I didn't expect.  To try to diagnose the problem with 
A/B testing I deleted all references to appstats and re-deployed my app. 
 Sure enough, the mysteriously high latency disappeared.  Confident I was 
on to something, I put appstats back, and the problem failed to reappear.

So now the performance of the app is in the range that I would expect.  The 
client-side latency is around a second, which is not great by many 
standards, but is sort-of OK for my application and is certainly better 
than 5+ seconds.  It's also closer to what I see inside the request 
handling itself - that is, when client latency is higher, so is the 
accumulated latency inside the request handler (from internal data 
manipulation etc).  This gives me the lever of optimizing my own calls to 
avoid occasionally-expensive operations, so it's much less of an issue; 
prior to this, the vast majority of the latency happened outside the 
request handling where I don't have any control.

If the performance of GAE generally is in the range I'm seeing now, that's 
a shippable product in my mind.  I am worried, however, that the problem 
was somewhere else in the system and somewhere down the line I will be hit 
with this problem again, only it will be in production and users will 
rightly complain that their experience is suffering.

So any further comment on this issue would be very welcome.  I'm still 
considering whether a backend is a more appropriate vehicle for this 
functionality.

Thanks,

- Kris

-- 
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/-/bF1zbyVA1BQJ.
To post to this group, send email to [email protected].
To unsubscribe from this group, send email to 
[email protected].
For more options, visit this group at 
http://groups.google.com/group/google-appengine?hl=en.

Reply via email to