[Tickets #10123] Re: Use memcache flags to keep track of large objects

[bugs at horde.org]

DO NOT REPLY TO THIS MESSAGE. THIS EMAIL ADDRESS IS NOT MONITORED.

Ticket URL: http://bugs.horde.org/ticket/10123
------------------------------------------------------------------------------
  Ticket             | 10123
  Updated By         | nuno.lopes at portugalmail.net
  Summary            | Use memcache flags to keep track of large objects
  Queue              | Horde Framework Packages
  Version            | Git master
  Type               | Enhancement
  State              | Resolved
  Priority           | 1. Low
  Milestone          |
  Patch              |
  Owners             | Michael Slusarz
------------------------------------------------------------------------------


nuno.lopes at portugalmail.net (2011-05-27 00:33) wrote:

I've been working with Gonçalo on this so I'll answer you.
>>> Your patch did not apply cleanly, so I had to go through by hand and
>>> make the necessary changes.  Could you test and make sure I
>>> transferred the changes correctly?
>>
>> Yep, its ok.
>> I think that it should be sent to dev (or the appropriate place), a
>> message to warn people that starting  to use this code will bring
>> unexpected results if they have large objects on memcache.
>
> How?  If I read the code correctly, this new code will simply cause  
> a cache miss.
>
With the last patch you made you're right. The problem here is that  
every key will cause a cache miss. This returns the platform into a  
cold cache state which isn't an expected behavior from a minor  
upgrade. We now have upwards 10G of cache data so I think you can  
understand how regenerating this data can make a performance hit.

>> We have made (untested yet) for ourselves a temporary patch to
>> convert old large objects to the new flag system. Attached the temp
>> patch
>
> No need to convert these values.  Memcache, as the name itself  
> implies, is a cache.  If data objects suddenly go missing, this is  
> an expected occurrence.

You're very right, but we use memcache to also store sessions and that  
implies logging everyone out. I agree that memcache isn't the right  
solution to store sessions but for our case it seems the best and it's  
an horde conf.

As Gonçalo said the warning would make sysadmins aware of these  
implications and the patch only makes the transition smoother.