libceph: add support for osd primary affinity
authorIlya Dryomov <ilya.dryomov@inktank.com>
Mon, 24 Mar 2014 15:12:49 +0000 (17:12 +0200)
committerSage Weil <sage@inktank.com>
Sat, 5 Apr 2014 04:08:17 +0000 (21:08 -0700)
Respond to non-default primary_affinity values accordingly.  (Primary
affinity allows the admin to shift 'primary responsibility' away from
specific osds, effectively shifting around the read side of the
workload and whatever overhead is incurred by peering and writes by
virtue of being the primary).

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
net/ceph/osdmap.c

index 20a38a37794cedd554c23f70ca97e02748e3d4a1..ae8f367c529121dbfe3a9d3258348bed9af7460b 100644 (file)
@@ -1596,6 +1596,72 @@ static int raw_to_up_osds(struct ceph_osdmap *osdmap,
        return len;
 }
 
+static void apply_primary_affinity(struct ceph_osdmap *osdmap, u32 pps,
+                                  struct ceph_pg_pool_info *pool,
+                                  int *osds, int len, int *primary)
+{
+       int i;
+       int pos = -1;
+
+       /*
+        * Do we have any non-default primary_affinity values for these
+        * osds?
+        */
+       if (!osdmap->osd_primary_affinity)
+               return;
+
+       for (i = 0; i < len; i++) {
+               if (osds[i] != CRUSH_ITEM_NONE &&
+                   osdmap->osd_primary_affinity[i] !=
+                                       CEPH_OSD_DEFAULT_PRIMARY_AFFINITY) {
+                       break;
+               }
+       }
+       if (i == len)
+               return;
+
+       /*
+        * Pick the primary.  Feed both the seed (for the pg) and the
+        * osd into the hash/rng so that a proportional fraction of an
+        * osd's pgs get rejected as primary.
+        */
+       for (i = 0; i < len; i++) {
+               int osd;
+               u32 aff;
+
+               osd = osds[i];
+               if (osd == CRUSH_ITEM_NONE)
+                       continue;
+
+               aff = osdmap->osd_primary_affinity[osd];
+               if (aff < CEPH_OSD_MAX_PRIMARY_AFFINITY &&
+                   (crush_hash32_2(CRUSH_HASH_RJENKINS1,
+                                   pps, osd) >> 16) >= aff) {
+                       /*
+                        * We chose not to use this primary.  Note it
+                        * anyway as a fallback in case we don't pick
+                        * anyone else, but keep looking.
+                        */
+                       if (pos < 0)
+                               pos = i;
+               } else {
+                       pos = i;
+                       break;
+               }
+       }
+       if (pos < 0)
+               return;
+
+       *primary = osds[pos];
+
+       if (ceph_can_shift_osds(pool) && pos > 0) {
+               /* move the new primary to the front */
+               for (i = pos; i > 0; i--)
+                       osds[i] = osds[i - 1];
+               osds[0] = *primary;
+       }
+}
+
 /*
  * Given up set, apply pg_temp and primary_temp mappings.
  *
@@ -1698,6 +1764,8 @@ int ceph_calc_pg_acting(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
 
        len = raw_to_up_osds(osdmap, pool, osds, len, primary);
 
+       apply_primary_affinity(osdmap, pps, pool, osds, len, primary);
+
        len = apply_temps(osdmap, pool, pgid, osds, len, primary);
 
        return len;