Dear Gluster devs, the translator performance/quick-read does not have cache-size option the way that performance/io-cache does - and therefore doesn't have anything like the ioc_prune() functionality of io-cache. I've discovered this can cause what is effectively a memory leak. I have a volume of up to 40GB configured with quick-read. When I attempt to do something that reads the contents of every file (e.g. take a tar of it) the glusterfs mount process just continues to grow until it's used all the memory of the client which then becomes unresponsive of course. Setting the timeout value doesn't help this because, without a max cache size and pruning, old files are not flushed out of the cache until they are read or written again - which doesn't happen with a sequential access of all files... This seems like a bug to me %-} Any comments? [aside: where is the gluster bug list held/managed?] Cheers, Ian -- www.ContactClean.com Making changing email address as easy as clicking a mouse. Helping you keep in touch.
