On Wed, Apr 6, 2016, at 09:12 AM, Noel Jones wrote: > > postfix/postscreen[18826]: cache > > btree:/var/lib/postfix/postscreen_cache full cleanup: retained=224 > > dropped=12 entries > > > > It looks like it's happening because they're 'full' at the time. > They are removed because they are expired.
Ok, so it's "full cleanup" of postscreen_cache, NOT a cleanup of "postscreen_cache fuill" > There is no limit on the cache size, and the default > postscreen_*_ttl values should be reasonable for the vast majority > of sites. Makes sense now. > > Is there any advantage/harm in using lmdb here, instead of btree? I'm > > already using lmdb as default DB type for most other !pcre tables. > > Access latency is the important part. Slow access to the cache will > limit the number of connections postfix can service. > btree is suggested because it's fast and supports the features needed. > I don't use lmdb, so I can't really answer if it's suitable for the > postscreen cache. IIUC, lmdb is based on / derived from btree. The in-memory (http://symas.com/mdb/inmem/) & on-disk benchmarks (http://symas.com/mdb/ondisk/)) suggest LMDB's performance, and latency (https://symas.com/getting-down-and-dirty-with-lmdb-qa-with-symas-corporations-howard-chu-about-symass-lightning-memory-mapped-database/) are significantly better than most. Boils down to whether lmdb: can/should be used in this parameter instance. I'll wait to hear from others on it. Thanks Jason
