Doom day

[Kaan Soral]

"The 50% discount for frontend instance hours will expire on December 1st. Please check your Billing History page and update your budget if necessary."

Since Python 2.7 is problematic, we have already been paying a lot with %50 discounted prices ... My daily costs went up to 90$ from 40$ before new pricing, now it will be ~160$s ... 4x ...

[Brandon Wirtz]

Don’t think that I don’t feel your pain, and I did rib the guys last night about the time frame for the pricing. (they clearly are sensitive to the matter)

 

But What did you expect?   I went in to the beta expecting release pricing to be 4-5x the beta price.  Sounds like that was spot on for you.

 

I moved to 2.7 I’m very happy with it.  I don’t know what is broken with yours, I’ll believe that it really is broken, but I will bet money, that if you optimized your code you could knock 25% off your price. I’m also guessing if you tuned the performance sliders you could knock another 25% off.

 

-Brandon

--
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/-/xd3yViF2pggJ.
To post to this group, send email to google-a...@googlegroups.com.
To unsubscribe from this group, send email to google-appengi...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/google-appengine?hl=en.

[Andrius A]

We have a form for that, but so far looks like everyone is happy :) but if you are not, please fill in

https://docs.google.com/spreadsheet/viewform?hl=en_US&formkey=dC1DMmpPSU1WZnk0d1FMa3JmNXIwaGc6MQ#gid=0

On 1 December 2011 15:27, Kaan Soral <kaan...@gmail.com> wrote:

"The 50% discount for frontend instance hours will expire on December 1st. Please check your Billing History page and update your budget if necessary."

Since Python 2.7 is problematic, we have already been paying a lot with %50 discounted prices ... My daily costs went up to 90$ from 40$ before new pricing, now it will be ~160$s ... 4x ...

--

[c h]

i too am frustrated by this.  when google announced the new pricing they also told us that they would discount frontend instance hours until python2.7 was supported.  right now python2.7 is in beta, and i am not feeling lucky enough to move production apps on to a beta product (given that i still have to use the beta data migration tool).

i'm disappointed that google has not been able to deliver production supported python2.7, but still raised the price.

[Kaan Soral]

http://code.google.com/p/googleappengine/issues/detail?id=6282
This is the problem

You may be right about the sliders
I am attaching my instances chart, any suggestions?
(I kept everything at automatic-automatic until now, trusting gae, but it seems it is over using instances)

[Gerald Tan]

If you set Max Idle Instance to automatic you will always pay for the total number of instances.
Looking at your chart I'd set Max Idle Instance to something between 20-25

[Alexis]

You can look at this thread if you want more details about instances
settings:

http://groups.google.com/group/google-appengine/browse_thread/thread/260e7cefc6e5d482/bf9b4a4a804a3091

[Alexis]

We are still using Python25 too and don't feel confident moving our
production apps to 2.7 when seeing issues like this:

http://code.google.com/p/googleappengine/issues/detail?id=6401

or this:

http://code.google.com/p/googleappengine/issues/detail?id=6323

or the one Kaan linked to...

[Brandon Wirtz]

Or this.

 

http://www.xyhd.tv/2011/11/industry-news/setting-google-appengine-gae-performance-sliders-for-instance-and-latency/

 

 

From: google-a...@googlegroups.com [mailto:google-a...@googlegroups.com] On Behalf Of Kaan Soral
Sent: Thursday, December 01, 2011 7:27 AM
To: google-a...@googlegroups.com
Subject: [google-appengine] Doom day

 

"The 50% discount for frontend instance hours will expire on December 1st. Please check your Billing History page and update your budget if necessary."

Since Python 2.7 is problematic, we have already been paying a lot with %50 discounted prices ... My daily costs went up to 90$ from 40$ before new pricing, now it will be ~160$s ... 4x ...

--

[Brandon Wirtz]

I run on 2.7. I have been rock solid and my costs are 1/10th what they are
on 2.5

I had very few code changes other than removing CGI handlers.

I would recommend 2.7 without hesitation.

Pointing out 2 bugs that aren't well documented, and (one of which I can't
Repo) Is more of a "I'm too lazy to do the migration" than a real excuse.
If you aren't testing on 2.7 you are weeks from being deployed on it anyway,
and are just griping. Get a test version of your app on it, find out where
it fails and if you have bugs file them. Don't be a wuss unless you have
the same code snippet that is listed in a bug in your source code. Then I
might say you should test on 2.7 but not invest time in adapting the code
beyond removing 2.7 incompatible code.

-Brandon

http://code.google.com/p/googleappengine/issues/detail?id=6401

or this:

http://code.google.com/p/googleappengine/issues/detail?id=6323

--

You received this message because you are subscribed to the Google Groups
"Google App Engine" group.

[Brian Quinlan]

Hi Brandon,

On Sat, Dec 3, 2011 at 8:11 PM, Brandon Wirtz <dra...@digerat.com> wrote:
> I run on 2.7. I have been rock solid and my costs are 1/10th what they are
> on 2.5

Wow, that's awesome!

> I had very few code changes other than removing CGI handlers.

Also good news.

> I would recommend 2.7 without hesitation.
>
> Pointing out 2 bugs that aren't well documented, and (one of which I can't
> Repo) Is more of a "I'm too lazy to do the migration" than a real excuse.
> If you aren't testing on 2.7 you are weeks from being deployed on it anyway,
> and are just griping. Get a test version of your app on it, find out where
> it fails and if you have bugs file them.  Don't be a wuss unless you have
> the same code snippet that is listed in a bug in your source code.  Then I
> might say you should test on 2.7 but not invest time in adapting the code
> beyond removing 2.7 incompatible code.

Both of the bugs that Alexis mentioned are significant and we can
reproduce them - though the probability of being affected by
http://code.google.com/p/googleappengine/issues/detail?id=6401 should
be pretty low.

Python 2.7 is still experimental and your good experiences aren't a
guarantee that others won't encounter serious problems.

Cheers,
Brian

[Brandon Wirtz]

I didn't mean to say that Alexis's bugs aren't real, just that the presence
of 2 bugs shouldn't prevent someone from starting the conversion to the new
system.

You can't make the move without testing. There are bugs in 2.5. Your odds
of hitting a bug is higher on 2.7, but they are far from guaranteed. I would
go so far as to say if your average request is less than 5 seconds, your
move is most likely to be painless.

Most of the issues that are cause by the high computation bug can be avoided
by splitting task across more than one request, or offloading to a back end
instance. Not Ideal, but not insurmountable. Even with the changes
necessary to avoid the scenarios that cause the high computation but 2.7 is
significantly faster than 2.5 and more cost effective. If you don't want to
modify your code setting Min Idle Instance Higher will prevent most issues
as well at the expense of more instance hours.

2.7 does occasionally have long start up times. This can be avoided by
using a warmup in the app.yaml and avoiding splitting your files across too
many .py files (not sure why this matters but it seems to)

2.7 is also much happier if your initialization variables are pulled from
mem-cache not datastore. Part of that whole you have to initialize quickly
or things get really slow. As a result it is not a bad idea to read and
re-write any initialization variables to memcache at the start of each
request. Doing this using serialized data makes this VERY fast. If you
have to use DataStore (which you do) you should also serialize the
initialization variables so that you need only make one call rather than 1
for each value.

2.7 does seem to have some interesting potential security holes, but these
are "by design" and are avoidable if you don't want them, and can be used
for certain performance increase if you know what you are doing. I believe
there are also a few subtle difference in some of the Typing that may impact
you if you are building non-english apps.

[Adrian Scott]

We're simply not going to 2.7 until it's no longer experimental. It's that simple. We're doing enough other stuff that's experimental etc. It would have made sense for Google to extend the 50% discount til one month after 2.7 leaving experimental, imho, and that would have created a nice little incentive to the google team, and also better relationships with developers after all that has gone on recently.

We're doing a bit of optimization as we go along, but putting a hefty priority on improving functionality and ease of use over just minimizing expense.

-A

--
Adrian Scott, Ph.D.
CEO, Founder
CoderBuddy
http://www.coderbuddy.com/ <-- Create a Facebook or Google App Engine app in a minute without installing anything

[Kaan Soral]

Great advice, thanks (to everyone)

I can't believe I was being billed for that much Idle instances, my idiotness for not investigating further and trusting the scheduler and GAE for it ...
I probably paid 1k-2k extra for this mistake

Don't even want to think what I could do with that money ...

[Brandon Wirtz]

The scheduler isn’t all things to all people, or all instances.  That’s why I have been trying to tell people to clamp it, and how to tune it (and complaining about how it is tuned).

 

The Full Auto mode on instances can work if you tune the latency.  Full auto on latency is almost always a bad idea.

 

We ran quite a few test passes to see what QoS users would get at different settings on our app, and I would advise you to do the same thing.  Testing in a Vacuum really lets you get a sense of how many you need of what.

 

Crawling through the apps that appear in Google Search results (yes I crawl results I have a bot) I found that VERY few of them set cache headers.  People keep telling me why those don’t matter, but I’m telling you, even the least cachable apps on the planet have HTTP requests re-polled by ISPs, the Alexa Bot, and other things, so if you aren’t using HTTPS you should enable Cache Headers to save a minimum of 10%.

 

 

From: google-a...@googlegroups.com [mailto:google-a...@googlegroups.com] On Behalf Of Kaan Soral

Sent: Sunday, December 04, 2011 11:26 AM
To: google-a...@googlegroups.com

--

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/-/XddC3fYV9l4J.