The bigger problem was that the sets were simply not working in some
cases. We have downgraded and all has been well for 3 days. 3.0.5 just
does not save data and does not error out when it does not save data.
Brian.
--------
http://brian.moonspot.net/
On 1/4/10 2:43 PM, NICK VERBECK wrote:
The PECL Memcache Client used to have a constant problem with
consistent hashing of keys leading to multi copies or out of date
copies on Memcached Servers. They appear to have fixed some of these
problems in 3.0.4. Which I would guess would lead 2.2.5 to produce
different hashes then 3.0.4+. If you can I would upgrade all machines
to use the same client. Making sure 1st that this doesn't break
anything else with your code. As I've seen problems with caching
un/serializing classes between the different versions of the client.
Otherwise stick with you current 2.2.5 install.
On Sat, Jan 2, 2010 at 3:09 PM, Brian Moon<[email protected]> wrote:
Hi,
We accidentally got 3.0.5 onto a set of our servers. We have seen some
really odd behavior. It would seem we have some failing sets, but the
method never returns false. But, pulling the data back out on the next run
of the script (not immediately, could be minutes apart) yields old data.
Also, it would seem the servers are using a different hashing algorithm than
ones running 2.2.5. No options are set to have it use consistent hashing.
We use traditional hashing. But, getting from a server running 2.2.5 does
not yield the same data as the servers running 3.0.5.
We are downgrading for now as these are production systems. If there is
anything we can do to debug this, let me know.
--
Brian.
--------
http://brian.moonspot.net/
