From 5a13b1834b3dfdeeeabb4569af5def7baa7c6248 Mon Sep 17 00:00:00 2001 From: w00t Date: Thu, 1 Nov 2007 15:15:52 +0000 Subject: Update docs on how XLine system will be working git-svn-id: http://svn.inspircd.org/repository/trunk/inspircd@8452 e03df62e-2008-0410-955e-edbf42e46eb7 --- src/xline.cpp | 32 +++++++++----------------------- 1 file changed, 9 insertions(+), 23 deletions(-) diff --git a/src/xline.cpp b/src/xline.cpp index f3a7b28b7..8fbb73280 100644 --- a/src/xline.cpp +++ b/src/xline.cpp @@ -36,31 +36,17 @@ * than it could have been, something which we address here. * * VERSION 3: - * All lines are (as in v1) stored together -- no seperation of perm and non-perm. Expiry will - * still use a sorted list, and we'll just ignore anything permanent. + * All lines are (as in v1) stored together -- no seperation of perm and non-perm. They are stored in + * a map of maps (first map is line type, second map is for quick lookup on add/delete/etc). * - * Application will be by a list of lines 'pending' application, meaning only the newly added lines - * will be gone over. Much faster. + * Expiry is *no longer* performed on a timer, and no longer uses a sorted list of any variety. This + * is now done by only checking for expiry when a line is accessed, meaning that expiry is no longer + * a resource intensive problem. * - * More of course is to come. - */ - -/* Version two, now with optimized expiry! - * - * Because the old way was horrendously slow, the new way of expiring xlines is very - * very efficient. I have improved the efficiency of the algorithm in two ways: - * - * (1) There are now two lists of items for each linetype. One list holds temporary - * items, and the other list holds permanent items (ones which will expire). - * Items which are on the permanent list are NEVER checked at all by the - * expire_lines() function. - * (2) The temporary xline lists are always kept in strict numerical order, keyed by - * current time + duration. This means that the line which is due to expire the - * soonest is always pointed at by vector::begin(), so a simple while loop can - * very efficiently, very quickly and above all SAFELY pick off the first few - * items in the vector which need zapping. - * - * -- Brain + * Application no longer tries to apply every single line on every single user - instead, now only lines + * added since the previous application are applied. This keeps S2S ADDLINE during burst nice and fast, + * while at the same time not slowing things the fuck down when we try adding a ban with lots of preexisting + * bans. :) */ bool XLine::Matches(User *u) -- cgit v1.2.3