#1653 Per-request in memory cache for all communication with the MemCache server(s) - Ruby on Rails

Type	To find
responsible:me	tickets assigned to you
tagged:"@high"	tickets tagged @high
milestone:next	tickets in the upcoming milestone
state:invalid	tickets with the state invalid
created:"last week"	tickets created last week
sort:number, importance, updated	tickets sorted by #, importance or updated
Combine keywords for powerful searching.
Use advanced searching »

This project is archived and is in readonly mode.

#1653 ✓resolved

Per-request in memory cache for all communication with the MemCache server(s)

Reported by Nahum Wild | December 28th, 2008 @ 07:28 AM

In November I released a plugin (http://tinyurl.com/7q5fqh) that enhances MemCacheStore by using a per-request local cache to buffer duplicate memcache reads which can result in halving the number of read requests made. I've been asked to add that functionality to Edge and this is that work.

The description linked to essentially describes two changes, the first to MemCacheStore which this patch applies directly to that class instead of creating a new Store class. And secondly tidy ups to the MemCache Session Store which this patch does not implement. I need to follow up about Joshua's comment on ticket #949 to see if I can help there.

spandex_mem_cache_store.diff 10.3 KB

Comments and changes to this ticket

You flagged this item as spam.
Nahum Wild December 28th, 2008 @ 08:41 PM
Ah, it's not thread safe - working on that now.
You flagged this item as spam.
Jeffrey Hardy January 6th, 2009 @ 05:56 PM
I totally dig the idea of this. It's similar to what Nick Kallen's cache_money plugin does:

http://github.com/nkallen/cache-...

The main thing that bothers me about the implementation in this patch is that it knows about Action Controller. The cache store shouldn't know Action Controller even exists. I'd much rather there be a facility to tell the cache store to use the local cache or not, such that we could declare the setting from AC's side.
You flagged this item as spam.
Nahum Wild January 8th, 2009 @ 09:19 PM
Jeffrey, I totally agree with you on the subject of the class knowing about ActionController. In fact my delay in updating this ticket with a thread safe version is because I've been trying to solve that problem nicely.

I like simply going:

config.action_controller.session_store = :mem_cache_store

to setup memcache in rails. I don't really want people to have to do extra things to get it going when it should just work seamlessly without any effort. At the moment it still depends on AC.

Doing the local caching actually solves a bug in rails, something I've not directly said above. It stops the problem of entries in memcache changing in between the execution of the action and view code. I've had this happen a number of times, pain in the butt to track down. I probably should have tagged this with bug too maybe...

Anyway, I've attached an updated patch with my changes being thread safe now plus a minor fix to the 'exist?' method.
- spandex_mem_cache_store.diff 19.8 KB
You flagged this item as spam.
Nahum Wild January 8th, 2009 @ 09:27 PM
Just to clariy for those reading, the dependency on ActionController is because MemCacheStore hooks into the Dispatcher before_dispatch callback to clear the local cache before each request is processed.
DHH January 12th, 2009 @ 01:49 PM
- Assigned user changed from “DHH” to “josh”
Josh, where should we best put this reset call?
josh January 12th, 2009 @ 02:58 PM
- Milestone cleared.
- State changed from “new” to “open”
I think a middleware hook would be most appropriate.

use ActiveSupport::Cache::MemCacheStore::LocalCache # or Buffering

You could have the middleware initialize and reset the local cache as well. Otherwise the cache store could function standalone without the local buffer.
Repository January 16th, 2009 @ 03:25 AM
- State changed from “open” to “resolved”
(from [a53ad5bba37199047ba20194933e122bf6b0252f]) Added in a local per request cache to MemCacheStore. It acts as a buffer to stop unneccessary requests being sent through to memcache [#1653 state:resolved] Signed-off-by: Joshua Peek josh@joshpeek.com http://github.com/rails/rails/co...
josh January 16th, 2009 @ 03:26 AM
Hey, I just commit it.

I made some changes, it uses a thread local instead of a lock. Should be much less overhead.

Please review and make sure it still functions how you want.
You flagged this item as spam.
Nahum Wild January 16th, 2009 @ 04:09 AM
Will do thanks for doing that. I'll look at it right now. (Clients discovered I had spare time :-(
You flagged this item as spam.
Nahum Wild January 16th, 2009 @ 06:29 AM
Yep, functions as intended. :-)
You flagged this item as spam.
Nahum Wild January 16th, 2009 @ 06:35 AM
Oh and thanks for giving it some polish Joshua.
josh January 16th, 2009 @ 02:40 PM
Interesting comments on github:

http://github.com/rails/rails/co...

Maybe it should be a subclass or extension module?
You flagged this item as spam.
Lourens Naudé January 17th, 2009 @ 03:23 AM
Josh,

Initial backend agnostic local cache strategy attached.Introduced an ActiveSupport::Cache::Strategy::* namespace, which may be frowned upon, but may be useful for similar optimizations.

Original test cases passes.
- backend_agnostic_local_cache_strategy.diff 12.4 KB
josh January 17th, 2009 @ 03:52 AM
- State changed from “resolved” to “open”
@Lourens Thanks

I think it can be implemented cleaner with extend and super. I'll take care of that though.

Thanks for the patch, will commit soon.
Repository January 18th, 2009 @ 12:01 AM
- State changed from “open” to “resolved”
(from [b08c96887538cf53670bb882e79996582375e6c9]) Decouple the local cache strategy from MemCacheStore for reuse with other remote stores [#1653 state:resolved]

Signed-off-by: Joshua Peek josh@joshpeek.com http://github.com/rails/rails/co...
You flagged this item as spam.
Lourens Naudé January 28th, 2009 @ 05:25 PM
@ Josh,

Am aware of two apps where the immutable snippet in #read raises with TypeError when the value being dup'ed is a special case constant eg. Fixnum, TrueClass etc.

Here's the reference to the MRI implementation of Object#clone, Object#dup ( object.c ) && the special constant declaration ( ruby.h ) from a 187 checkout :

http://gist.github.com/54069

Attached is a patch that implements Object#dup? to better handle such cases.

Thoughts ?
- Lourens
- object_conditional_dup.diff 2.8 KB
You flagged this item as spam.
Lourens Naudé January 28th, 2009 @ 05:36 PM
Updated diff for Object#clone?
- object_conditional_dup_and_clone.diff 3.5 KB
josh January 28th, 2009 @ 06:35 PM
Maybe we could just add dup to Integer and just return self?
You flagged this item as spam.
Lourens Naudé January 28th, 2009 @ 08:12 PM
@ Josh,

Interesting, but isn't it dodgy practice to mask the TypeError in non-framework code VS being able to conditionally guard against it in specific framework areas, like the local cache, to not blow up the whole process when that happens ?
You flagged this item as spam.
Lourens Naudé January 29th, 2009 @ 02:38 AM
Josh,

Nevermind, references :

"* object.duplicable? returns true if object.dup is safe. False for nil, true, false, symbols, and numbers; true otherwise. #9333 [sur]"

Patch that piggy backs on that attached.
- local_cache_dup_if_duplicable.diff 885 Bytes
Repository January 29th, 2009 @ 03:18 AM
(from [2e69db18ce2815c25eee64fc2978e7f6b57f6e1f]) Only dup local cache values if duplicable [#1653 state:resolved] http://github.com/rails/rails/co...
Ryan Bigg October 9th, 2010 @ 09:54 PM
- Tag cleared.
- Importance changed from “” to “Low”
Automatic cleanup of spam.