In a web application I'm using a pool of servers with Ketama hashing. I'm using Memcached::libmemcached, and I'm persisting the "memc" handle between requests, per process.
If a get() or set() call fails due to a memcached server failure I want to pull that server out of the server list so that new keys will map to the remaining active servers. The Perl module only provides a "walk_stats()" call which uses memcached_stat to return stats for every server in the current list. If a memcached server fails then "walk_stats()" returns nothing -- so It doesn't tell me which server failed. So, here's the approach I'm considering, but I'd like to hear other suggestions (or reasons why this approach is insane): I create a new handle (memcached_create) and then add the servers. As I call memcached_server_add() for each server I also call walk_stats() to make sure that I have stats for that newly added server. If that stat fails I flag the server as bad and start over. In the end I end up with just a list of servers that are alive (well, returned stats). Then in my get() and set() calls if I get an error I destroy my "memc" handle which will trigger reloading the server list as above. Is there a better way to do this? I can see one issue. If a server is intermittent might be possible to end up with the same key in multiple servers -- so as the server is toggled in and out of the pool the app could get different values for the same key at different times. But, that's a problem not really related to the method above. Thanks, -- Bill Moseley [EMAIL PROTECTED] Sent from my iMutt
