Things for 2.5: [09:27] we also need to audit all of these schema defs [09:27] we're supposed to have official, non-experimental OIDs for released schema [09:28] accesslog is still using 666, experimental arc [09:29] I think this means we should polish up the logschema draft, Informational status, and publish it again as final [13:57] there's a nagging problem though, pcache's entry_release function needs to distinguish between its backend actually freeing the entry, or being a no-op [13:57] so it can decide whether to return success or continue [13:58] the patch to translucent sidesteps the question, by avoiding other overlays [13:58] but we need to revisit this in 2.5 ITS#7920 fix for slapacl This needs to be streamlined in 2.5, current tool API is a mess. ITS#7926 - [18:19] this is going to be a WONTFIX for 2.4, I think [18:20] reiniting the listeners array is probably too much of a pain. it will screw with all currently open connections [10:22] and rewriting Bind and Exop result handling [10:22] so that it's no longer a special case for overlays [23:34] ok, so far heap profile shows that memory use during refresh is normal [23:35] not wonderful, but normal. mem usage grows because we're recording the present list while receiving entries in the refresh [23:36] I'm seeing for 1.2GB of data about 235MB of presentlist [23:36] which is pretty awful, considering presentlist is just a list of UUIDs [23:36] being stored in an avl tree [23:37] a btree would have been better here, and we could just use an unsorted segmented array [23:42] for the accumulation phase anyway. we need to be able to lookup records during the delete pphase [00:05] this stuff seriously needs a rewrite [01:13] 2.8M records x 16 bytes per uuid so this should be no more than 48MB of overhead [01:13] and instead it's 3-400MB