From 56332f0b1e67f3c63d13ed38161761f5455719d1 Mon Sep 17 00:00:00 2001 From: Sage Weil Date: Fri, 16 May 2008 14:54:04 -0700 Subject: [PATCH] todos --- src/TODO | 39 +-------------------------------------- 1 file changed, 1 insertion(+), 38 deletions(-) diff --git a/src/TODO b/src/TODO index 8dd60f9114b88..a3a40c07e7d72 100644 --- a/src/TODO +++ b/src/TODO @@ -143,44 +143,10 @@ rados snapshots objecter - maybe_request_map should set a timer event to periodically re-request. -- transaction prepare/commit +- transaction prepare/commit? - read+floor_lockout osd/rados -- fix build_prior_set behavior. needs to not always exclude currently down nodes. e.g., - - 1: A B - 2: B - 3: - 4: A - -> prior_set should be , bc B may have independently applied updates. - - ideas: - - can't activate pg when lone OSD without informing the monitor of alive_thru - - add alive_thru map to osdmap. any lone pg will not have activated if the osd is now down, and alive_thru does not include the given epoch. - - 1: A B C - 2: B C - 3: - 4: A C -> prior_set can be , bc C would carry any epoch 2 updates - - -1: A B -2: C D .. can't have gone active - -1: A B -2: B C B in prior_set, will - - - -> so: we need at least 1 osd from each epoch, IFF we make store sync on osdmap boundaries. - - -> so, use calc_priors_during in build_prior, then make recovery code check for is_up - -- paxos replication (i.e. majority voting)? - -- transaction prepare/commit - - rollback - - rollback logging (to fix slow prepare vs rollback race) - a more general fencing mechanism? per-object granularity isn't usually a good match. @@ -199,10 +165,7 @@ messenger - fix messenger shutdown.. we shouldn't delete messenger, since the caller may be referencing it, etc. simplemessenger -- fix/audit accept() logic to detect reset, do callback - close idle connections -- take a look at RDS? http://oss.oracle.com/projects/rds/ - objectcacher - merge clean bh's -- 2.39.5