From f293f7ac2f75c58d81da1229b484eb530b7083b5 Mon Sep 17 00:00:00 2001 From: Sunny Kumar Date: Fri, 20 Sep 2019 09:39:12 +0530 Subject: [PATCH 299/302] geo-rep: performance improvement while syncing renames with existing gfid Problem: The bug[1] addresses issue of data inconsistency when handling RENAME with existing destination. This fix requires some performance tuning considering this issue occurs in heavy rename workload. Solution: If distribution count for master volume is one do not verify op's on master and go ahead with rename. The performance improvement with this patch can only be observed if master volume has distribution count one. [1]. https://bugzilla.redhat.com/show_bug.cgi?id=1694820 Backport of: >fixes: bz#1753857 >Change-Id: I8e9bcd575e7e35f40f9f78b7961c92dee642f47b >Signed-off-by: Sunny Kumar Upstream Patch: https://review.gluster.org/#/c/glusterfs/+/23459/ BUG: 1726000 Change-Id: I8e9bcd575e7e35f40f9f78b7961c92dee642f47b Signed-off-by: Sunny Kumar Reviewed-on: https://code.engineering.redhat.com/gerrit/181893 Tested-by: RHGS Build Bot Reviewed-by: Atin Mukherjee --- geo-replication/gsyncd.conf.in | 5 +++++ geo-replication/syncdaemon/gsyncd.py | 2 ++ geo-replication/syncdaemon/monitor.py | 2 ++ geo-replication/syncdaemon/resource.py | 13 +++++++++++-- geo-replication/syncdaemon/syncdutils.py | 11 +++++++++++ 5 files changed, 31 insertions(+), 2 deletions(-) diff --git a/geo-replication/gsyncd.conf.in b/geo-replication/gsyncd.conf.in index 5ebd57a..9155cd8 100644 --- a/geo-replication/gsyncd.conf.in +++ b/geo-replication/gsyncd.conf.in @@ -23,6 +23,11 @@ configurable=false type=int value=1 +[master-distribution-count] +configurable=false +type=int +value=1 + [glusterd-workdir] value = @GLUSTERD_WORKDIR@ diff --git a/geo-replication/syncdaemon/gsyncd.py b/geo-replication/syncdaemon/gsyncd.py index a4c6f32..6ae5269 100644 --- a/geo-replication/syncdaemon/gsyncd.py +++ b/geo-replication/syncdaemon/gsyncd.py @@ -134,6 +134,8 @@ def main(): help="Directory where Gluster binaries exist on slave") p.add_argument("--slave-access-mount", action="store_true", help="Do not lazy umount the slave volume") + p.add_argument("--master-dist-count", type=int, + help="Master Distribution count") # Status p = sp.add_parser("status") diff --git a/geo-replication/syncdaemon/monitor.py b/geo-replication/syncdaemon/monitor.py index 234f3f1..236afe7 100644 --- a/geo-replication/syncdaemon/monitor.py +++ b/geo-replication/syncdaemon/monitor.py @@ -37,6 +37,8 @@ def get_subvol_num(brick_idx, vol, hot): tier = vol.is_tier() disperse_count = vol.disperse_count(tier, hot) replica_count = vol.replica_count(tier, hot) + distribute_count = vol.distribution_count(tier, hot) + gconf.setconfig("master-distribution-count", distribute_count) if (tier and not hot): brick_idx = brick_idx - vol.get_hot_bricks_count(tier) diff --git a/geo-replication/syncdaemon/resource.py b/geo-replication/syncdaemon/resource.py index b16db60..189d8a1 100644 --- a/geo-replication/syncdaemon/resource.py +++ b/geo-replication/syncdaemon/resource.py @@ -377,6 +377,7 @@ class Server(object): def entry_ops(cls, entries): pfx = gauxpfx() logging.debug('entries: %s' % repr(entries)) + dist_count = rconf.args.master_dist_count def entry_purge(op, entry, gfid, e, uid, gid): # This is an extremely racy code and needs to be fixed ASAP. @@ -686,9 +687,15 @@ class Server(object): raise else: raise - elif not matching_disk_gfid(gfid, en): + elif not matching_disk_gfid(gfid, en) and dist_count > 1: collect_failure(e, EEXIST, uid, gid, True) else: + # We are here which means matching_disk_gfid for + # both source and destination has returned false + # and distribution count for master vol is greater + # then one. Which basically says both the source and + # destination exist and not hardlinks. + # So we are safe to go ahead with rename here. rename_with_disk_gfid_confirmation(gfid, entry, en, uid, gid) if blob: @@ -1409,7 +1416,9 @@ class SSH(object): '--slave-gluster-log-level', gconf.get("slave-gluster-log-level"), '--slave-gluster-command-dir', - gconf.get("slave-gluster-command-dir")] + gconf.get("slave-gluster-command-dir"), + '--master-dist-count', + str(gconf.get("master-distribution-count"))] if gconf.get("slave-access-mount"): args_to_slave.append('--slave-access-mount') diff --git a/geo-replication/syncdaemon/syncdutils.py b/geo-replication/syncdaemon/syncdutils.py index 2ee10ac..aadaebd 100644 --- a/geo-replication/syncdaemon/syncdutils.py +++ b/geo-replication/syncdaemon/syncdutils.py @@ -926,6 +926,14 @@ class Volinfo(object): else: return int(self.get('disperseCount')[0].text) + def distribution_count(self, tier, hot): + if (tier and hot): + return int(self.get('hotBricks/hotdistCount')[0].text) + elif (tier and not hot): + return int(self.get('coldBricks/colddistCount')[0].text) + else: + return int(self.get('distCount')[0].text) + @property @memoize def hot_bricks(self): @@ -994,6 +1002,9 @@ class VolinfoFromGconf(object): def disperse_count(self, tier, hot): return gconf.get("master-disperse-count") + def distribution_count(self, tier, hot): + return gconf.get("master-distribution-count") + @property @memoize def hot_bricks(self): -- 1.8.3.1