14

I have a very weird problem, when storing my session in Memcached. From time to time some users takes the session of others. E.g. John, becomes logged in as Maria, Maria as Chris and so on.

I use Rails 2.3.4, but the same problem has been happening with earlier versions of Rails. I use only one Memcache server and it's running on the same machine. The problem with debugging this is that I can not reproduce it.

I'll be very glad if anybody can guide me how to solve this problem or debug it. I'll be also happy if you are using Memcached for sessions and you share your example confgs.

These are my configurations:

# memcache options
memcache_options = {
  :c_threshold => 10_000,
  :compression => false,
  :debug => false,
  :namespace => 'app_prod',
  :readonly => false,
  :urlencode => false,
}
memcache_servers = ['localhost:11211']

CACHE = MemCache.new(memcache_options)
CACHE.servers = memcache_servers

config.cache_store = :mem_cache_store, memcache_servers, memcache_options
config.action_controller.session_store = :mem_cache_store
config.action_controller.session = {
  :session_key => '_appname',
  :cache => CACHE,
#    :expires => 10,
#    :session_expires => 10,
  :secret      => '5391aaaaaaaaaa56f8e8234beb638b97b32bbbbbbbbbbcc9dcae2beccccccccc89e8b508328def001a368da0678b061eb0e9d5a82a5ac94c8d35bd31a9a49e1'
}

Thank you in advance, Stan

Jeremy Stein
  • 19,171
  • 16
  • 68
  • 83
Stan Bright
  • 1,355
  • 1
  • 15
  • 22

5 Answers5

5

I've seen this and found it very difficult to debug.

If you're using passenger, you may want to look at using the conservative method for spawning new servers.

The default method has servers sharing a single socket to memcache.

The docs discuss it in more detail. http://www.modrails.com/documentation/Users%20guide%20Apache.html#_example_1_memcached_connection_sharing_harmful

AndyWatts
  • 346
  • 2
  • 5
  • Yes, I use Passenger. I'll test it on both my applications and will share the results here. Thank you a lot for your help. I can "smell" that the problem is here. – Stan Bright Oct 15 '09 at 11:07
  • I envy your nose but I'd not flag the answer as correct until the result is verified – Andriy Volkov Oct 27 '09 at 20:01
3

This could be a problem with the session cookie flipping between two values. For example, you might have one assigned to example.com and another to www.example.com, a common situation with some sites that respond to both without redirecting to make one canonical.

The behavior of some browsers is to send the cookie matching the longest subdomain, whereas others actually send through both values, and they may differ. This could lead to a session toggling between two different values at unpredictable times.

One way to fix this is to lock your cookies to .domain.com instead of letting it assume the www or www-less version, if this is the case, or redirecting to force the use of one only.

Another way to diagnose the nature of the session situation is to have a debugging page that displays the session ID, or embed it in the page output somehow so someone who encounters the problem can help in diagnosing it. Something like /session_info is easy to create.

tadman
  • 208,517
  • 23
  • 234
  • 262
  • OK, I have this config option: ActionController::Base.session_options[:session_domain] = '.mysite.com' I think this config option means that the application will have to use one session for all subdomains including the main (mysite.com) . Also, can you give a further advice what I could do if I see the session_id of a problematic User? – Stan Bright Oct 07 '09 at 14:17
  • That's the way to set the cookie properly, so that much should be working. If you want to get adventuresome, you can look through the production.log file for which session_id values are used. Usually when one user flips to another it's because sessions are getting mixed up. If that's not the case, at least you can focus your search in other areas. – tadman Oct 07 '09 at 15:15
  • Yes, I think that the sessions are getting mixed up and I'm lookin for a solution of this problem. – Stan Bright Oct 07 '09 at 15:25
3

Here it is the code that resolves the problem for me:

I added these lines at the end of

environment.rb

if defined?(PhusionPassenger)
  PhusionPassenger.on_event(:starting_worker_process) do |forked|
    if forked
      CACHE.reset
      if Rails.cache.class == ActiveSupport::Cache::MemCacheStore
        Rails.cache.instance_variable_get(:@data).reset
      end
    end
  end
end
Stan Bright
  • 1,355
  • 1
  • 15
  • 22
2

I never ran into such a problem before, I just can't imagine that it's even happening. This is my conf:

require 'memcache'

memcache_options = {
  :c_threshold => 10_000,
  :compression => true,
  :debug => false,
  :namespace => "app-me",
  :readonly => false,
  :urlencode => false
}
memcache_servers = [ "#{MEMCACHED_HOST}:#{MEMCACHED_PORT}" ]

CACHE = MemCache.new memcache_options

CACHE.servers = memcache_servers
ActionController::Base.session_options[:expires] = 1800
ActionController::Base.session_options[:cache] = CACHE

# Inside the Rails initializer
config.action_controller.session_store = :mem_cache_store
khelll
  • 23,590
  • 15
  • 91
  • 109
  • Well as you can see, the only difference with my memcache_options is that I don't use :compression. However I don't think it should be a problem. In both apps that have problems with the session I've configured such redirects: http://www.mysite.com => http://mysite.com . Do you have domain/subdomain redirects with yours? – Stan Bright Oct 07 '09 at 14:21
0

The Dalli Gem might help. A recent commit fixed socket sharing, so you could look at their code and see how they did it.

BryanH
  • 5,826
  • 3
  • 34
  • 47
user1648421
  • 1