Possible solutions to txtdrive process killing for cpu usage


#1

Hi,

I’ve just deployed a basic ecommerce site on textdrive.com. The
application has been stress tested on our local development machines and
appears to be running fine (no memory leaks).

However, on textdrive the fcgi processes are being killed after just a
few requests apparently because of cpu usage (The log says the processes
exceeds 17% average cpu usage in the last few minutes)

At first it seemed that they we’re being killed because the processes
were exceediing the 48Mb resident memory size limit (RSS seemed to
suddenly increase on certain pages (mostly backoffice functions) though
not consistently) but I was pointed to a process_watchdog.log filr which
points to the cpu usage.

Has anyone come across a similar situation? Any hints suggestions on
what I can do about it?

I’d really appreciate any help as although I’m now doing a optimization
refactoring iteration I’m uncertain how I can avoid this.

Regards,

Saimon M.


#2

I think you should post this to the textdrive forum (http://
forum.textdrive.com/).

On May 18, 2006, at 7:54 PM, Saimon M. wrote:

exceeds 17% average cpu usage in the last few minutes)
what I can do about it?
Posted via http://www.ruby-forum.com/.


Rails mailing list
removed_email_address@domain.invalid
http://lists.rubyonrails.org/mailman/listinfo/rails

Dae San H.
removed_email_address@domain.invalid


#3

I already have but what I’m asking for on this list is help or
suggestions with how I could go about tracking/reducing the cpu usage
problem which is rails related.

Dae San H. wrote:

I think you should post this to the textdrive forum (http://
forum.textdrive.com/).

On May 18, 2006, at 7:54 PM, Saimon M. wrote:

exceeds 17% average cpu usage in the last few minutes)
what I can do about it?
Posted via http://www.ruby-forum.com/.


Rails mailing list
removed_email_address@domain.invalid
http://lists.rubyonrails.org/mailman/listinfo/rails

Dae San H.
removed_email_address@domain.invalid


#4

On May 18, 2006, at 4:17 AM, Saimon M. wrote:

exceeds 17% average cpu usage in the last few minutes)
what I can do about it?
Posted via http://www.ruby-forum.com/.


Rails mailing list
removed_email_address@domain.invalid
http://lists.rubyonrails.org/mailman/listinfo/rails

Dae San H.
removed_email_address@domain.invalid

Well, what does your app do? What parts of it are processor

intensive? Using RMagick? Some details and we might be able to help you.

-Ezra


#5

Hi Ezra,

This is what I find strange. Just to choose one example, I have an edit
product view whose controller action simply finds the product and makes
an instance variable out of it so that the form can be populated (i.e.
it’s not the product update action).

e.g.

@product = Product.find(params[:id])
@song = Song.new
unless @product && @product.sellable
  @artists = String.new
  @label = Label.new
  @genre = Genre.new
  @instrument = Instrument.new
  @supplier = Supplier.new
else
  @artists = @product.sellable.artists.collect {|a| a.name }.join(' 

, ')
@label = @product.sellable.label.name
@genre = @product.sellable.genre.name
@instrument = @product.sellable.instrument.name
@supplier = @product.supplier.name
end

The view is just a basic form (created from a few partials) utilizing
standard rails form tags to edit the product. It has a small section
specific to the kind of sellable and a section specific to product
details.

I have suspected Rmagick as a possible offender but after further tests
I have rejected it as a possibility. (Images are processed via rmagick
just the once and thereafter they are cached on the browser)

The Product model:

has_one :sellable
belongs_to :supplier #where supplier is sublclassed into Album/Song
(STI)
belongs_to :image
has_many :shopping_items, :dependent => :destroy
has_many :supplier_order_items
belongs_to :user

There’s nothing very complex about this action apart from the fact that
the AR model has a few associations.

I have stress tested this action on my development box (set to
production mode) and it appears rock solid. The fcgi process has a
steady 28Mb during and after a heavy attack (ab -kc 50 -t 300). cpu
levels are similarly non-interesting. valgrind memcheck doesn’t report
anything out of the ordinary

On textdrive though it’s a different story, rss starts of at 32Mb and
rockets up after only a few refreshes of this action (and yes I’m
running the app in production mode), the cpu cycles also increase,
though up until I say the process_watchdog.log I wasn’t that worried
about them.

In short I’m stumped as to why this is behaving so differently on
textdrive and I’m not really sure what I can do to optimize this.

I’d appreciate any advice.

Thanks,

Saimon
Ezra Z. wrote:

On May 18, 2006, at 4:17 AM, Saimon M. wrote:

exceeds 17% average cpu usage in the last few minutes)
what I can do about it?
Posted via http://www.ruby-forum.com/.


Rails mailing list
removed_email_address@domain.invalid
http://lists.rubyonrails.org/mailman/listinfo/rails

Dae San H.
removed_email_address@domain.invalid

Well, what does your app do? What parts of it are processor
intensive? Using RMagick? Some details and we might be able to help you.

-Ezra


#6

Yes it’s very strange which is why I’m stumped.

I’m running Gentoo Linux 2.6.16r4.

I have opened the ticket with them but atm they just point me to the
process_watchdog.log.

You mentioned you’re no longer with textdrive. Could I ask who you’re
with?

Thanks again for helping out…

Regards,

Saimon

Ezra Z. wrote:

Hi !

On May 18, 2006, at 9:55 AM, Saimon M. wrote:

a.name }.join(’
details.
(STI)
production mode) and it appears rock solid. The fcgi process has a

In short I’m stumped as to why this is behaving so differently on
textdrive and I’m not really sure what I can do to optimize this.

I’d appreciate any advice.

Thanks,

Saimon

Saimon-

Yeah that sounds a bit fishy when it works fine on your local box
but not on txtdrv server. What OS are you running locally? I don’t
think I have any clear cut answers for you right off the bat here.
Since it seems like a textdrive problem you should probably file a
support ticket there detailing how it works locally but doesn’t work
on their servers.

I no longer use textdrive so I can’t say what the issue might be.
Hope you get it figured out though.

Cheers-
-Ezra


#7

On May 18, 2006, at 11:08 AM, Saimon M. wrote:

Thanks again for helping out…

Regards,

Saimon

Saimon-

I host a bunch of rails stuff on Xen based VPS's with http://

rimuhosting.com It is so much nicer to have your own separate
environment to run your rails apps in without anyone else killing
your processes and such.

I have written a tutorial a while ago about setting up a debian vps

from scratch here: http://brainspl.at/rails_stack.html

If you don't want to be your own sysadmin I also like PlanetArgon

for shared hosting.

-Ezra


#8

Hi !

On May 18, 2006, at 9:55 AM, Saimon M. wrote:

a.name }.join(’
details.
(STI)
production mode) and it appears rock solid. The fcgi process has a

In short I’m stumped as to why this is behaving so differently on
textdrive and I’m not really sure what I can do to optimize this.

I’d appreciate any advice.

Thanks,

Saimon

Saimon-

Yeah that sounds a bit fishy when it works fine on your local box

but not on txtdrv server. What OS are you running locally? I don’t
think I have any clear cut answers for you right off the bat here.
Since it seems like a textdrive problem you should probably file a
support ticket there detailing how it works locally but doesn’t work
on their servers.

I no longer use textdrive so I can't say what the issue might be.

Hope you get it figured out though.

Cheers-
-Ezra


#9

I’ve had a look at rimuhosting.com. Seems quite interesting. As this
application is for a client of mine and at the moment I’m using my own
personal sharing plan to host it, I’m probably going to recommend a
hosting change if things don’t pan out with textdrive (I was going to do
that sooner or later) so I’ll be evaluating it and Planet Argon too.

Nice tutorial…

Thanks for the info…

Ezra Z. wrote:

On May 18, 2006, at 11:08 AM, Saimon M. wrote:

Thanks again for helping out…

Regards,

Saimon

Saimon-

I host a bunch of rails stuff on Xen based VPS’s with http://
rimuhosting.com It is so much nicer to have your own separate
environment to run your rails apps in without anyone else killing
your processes and such.

I have written a tutorial a while ago about setting up a debian vps
from scratch here: http://brainspl.at/rails_stack.html

If you don’t want to be your own sysadmin I also like PlanetArgon
for shared hosting.

-Ezra


#10

On May 18, 2006, at 9:55 AM, Saimon M. wrote:

I have suspected Rmagick as a possible offender but after further
tests
I have rejected it as a possibility. (Images are processed via rmagick
just the once and thereafter they are cached on the browser)

FYI, historically Rmagick has been the source of many issues on
TextDrive because usage can go through the roof. Also, many TxD
servers are 64-bit, which would have different memory usage patterns
than you may expect.


_Deirdre http://deirdre.net