Appscale Lite

[Airam Rguez]

Hi appscalers!

I usually run out of memory or CPU and I was wondering if you plan to release a reduced or lite version of Appscale?

Thanks in advance!

[Navraj Chohan]

Hey Airam,

Thanks for your interest in AppScale!

We've been talking about an AppScale lite for a long time, but never get around to implementing it. We would have to create stub implementations that mimic Cassandra and ZooKeeper (the biggest memory consumers, and you could probably get away with just stubbing out cassandra). The interface would be pretty simple to do so but we'd need code to allow for you to configure it in lite mode. And as always, we highly welcome contributions if you are up the challenge of working on this. I'm here to help guide you for such an endeavor =)

For the time being I would suggest using the GAE SDK for development if you cannot get enough memory.

Thanks, 

Raj

--
You received this message because you are subscribed to the Google Groups "AppScale Community" group.
To unsubscribe from this group and stop receiving emails from it, send an email to appscale_commun...@googlegroups.com.
To post to this group, send email to appscale_...@googlegroups.com.
Visit this group at http://groups.google.com/group/appscale_community.
For more options, visit https://groups.google.com/d/optout.

--

Navraj Chohan (Raj)
Co-Founder of AppScale Systems
A Google Cloud Partner,

Gartner 2014 Cool Vendor, 
and a GigaOm Structure/LaunchPad "Best of the Best" Startups

615 State Street
Santa Barbara, CA 93101
r...@appscale.com
http://navrajchohan.com

[graziano obertelli]

Hi Airam,

indeed what Raj mention is correct. Out of curiosity, how much memory
and CPU are you allocating to AppScale, and what service is getting OOM
first?

cheers,
graziano

> --
> You received this message because you are subscribed to the Google
> Groups "AppScale Community" group.
> To unsubscribe from this group and stop receiving emails from it, send
> an email to appscale_commun...@googlegroups.com.
> To post to this group, send email to
> appscale_...@googlegroups.com.
> Visit this group at http://groups.google.com/group/appscale_community.
> For more options, visit https://groups.google.com/d/optout.

--

Graziano Obertelli
AppScale Systems, Inc.

[Airam Rguez]

Hi Graziano,

I've made several apps with Appscale each of them running in different VPSs. One of them is a push notification server. I first started with a 2GB VPS and 1 core. As the system started to grow up I added 2 more GB. I'm currently using an 8GB RAM VPS with 2 cores. I only use the datastore and Taskqueue services for that app (at least explicitly). I've been experiencing OOM in each stage of adding more RAM to the VPS until reaching the amount of 8GB.

The strategy I use to send notifications is to queue them in the Taskqueue. I've set the total_storage_limit to 120M for the Taskqueue in order to avoid too much memory consumption. I've also limited the amount of parallel tasks to 1 (on most of them) or 2 for the 4 available tasks queues. I can ensure that the system does not have a big demand.

Errors from different kinds get raised in the middle of a task and then the system goes down. Sometimes the log shows timeout errors and other times it shows exceptions from somewhere in the Appscale code (cannot remember exactly neither where nor the error message... but if you find this useful I could try to grab some errors from the log).

Thanks for your support.

[Airam Rguez]

Hi Raj,

Thank you for your prompt response.

Stubbing ZooKeeper would mean not having datastore transactions?

El lunes, 11 de mayo de 2015, 22:00:53 (UTC+2), Airam Rguez escribió:

[Navraj Chohan]

Depends how good your stub is, but for the most part you'll lose ACID semantics.

[graziano obertelli]

Hi Airam,

thanks for the details. When the system goes down as you say, do you
know the memory footprint of the different processes is and which
process gets axed? Is there a pattern here, for example it happens after
a certain amount of time, or after certain operations? The logs you
mentioned may be helpful, depending if they happen before or after the
OOM situation.

Also, can you share with us the output of 'free'? If there is high
memory consumption but no need to have it all resident, you may be able
to survive the peaks adding a bit of swap space.

Would you also mind to share how much disk is available? I wonder if you
experienced some issues during automatic Cassandra repairs.

cheers,
graziano

> --
> You received this message because you are subscribed to the Google
> Groups "AppScale Community" group.
> To unsubscribe from this group and stop receiving emails from it, send
> an email to appscale_commun...@googlegroups.com.
> To post to this group, send email to
> appscale_...@googlegroups.com.
> Visit this group at http://groups.google.com/group/appscale_community.
> For more options, visit https://groups.google.com/d/optout.