osd/OSDMap: allow bidirectional swap of pg-upmap-items

This is useful when we also want an even distribution of pg primaries across osds.
For example:
Was:
[0 1 2]

By applying bidirectional swap of pg-upmap-items mapping [[0,1],[1,0]], now:
[1 0 2]

Thus we successfully decrease the number of primaries of osd.0 by 1 without
affecting the current (even) distribution of global pgs.

Real exmaple:
./bin/ceph pg ls-by-pool rbd
PG_STAT OBJECTS MISSING_ON_PRIMARY DEGRADED MISPLACED UNFOUND BYTES LOG DISK_LOG STATE        STATE_STAMP                VERSION REPORTED UP      UP_PRIMARY ACTING  ACTING_PRIMARY LAST_SCRUB SCRUB_STAMP                LAST_DEEP_SCRUB DEEP_SCRUB_STAMP
3.0           0                  0        0         0       0     0   0        0 active+clean 2017-07-12 15:14:45.083441     0'0    29:13 [0,1,3]          0 [0,1,3]              0        0'0 2017-07-12 15:14:14.515989             0'0 2017-07-12 15:14:14.515989

./bin/ceph osd pg-upmap-items 3.0 0 1 1 0 3 5
set 3.0 pg_upmap_items mapping to [0->1,1->0,3->5]

./bin/ceph pg ls-by-pool rbd
PG_STAT OBJECTS MISSING_ON_PRIMARY DEGRADED MISPLACED UNFOUND BYTES LOG DISK_LOG STATE        STATE_STAMP                VERSION REPORTED UP      UP_PRIMARY ACTING  ACTING_PRIMARY LAST_SCRUB SCRUB_STAMP                LAST_DEEP_SCRUB DEEP_SCRUB_STAMP
3.0           0                  0        0         0       0     0   0        0 active+clean 2017-07-12 15:16:22.648424     0'0    33:13 [1,0,5]          1 [1,0,5]              1        0'0 2017-07-12 15:14:14.515989             0'0 2017-07-12 15:14:14.515989

Signed-off-by: xie xingguo <xie.xingguo@zte.com.cn>

diff --git a/src/osd/OSDMap.cc b/src/osd/OSDMap.cc
index ea43ae773228f7a156a18e799a17da291cee7d23..42d896b68ae0dab887836b708e9e3d92c836814b 100644 (file)
--- a/src/osd/OSDMap.cc
+++ b/src/osd/OSDMap.cc
@@ -1936,28 +1936,17 @@ void OSDMap::_apply_upmap(const pg_pool_t& pi, pg_t raw_pg, vector<int> *raw) co
 
   auto q = pg_upmap_items.find(pg);
   if (q != pg_upmap_items.end()) {
-    // NOTE: this approach does not allow a bidirectional swap,
-    // e.g., [[1,2],[2,1]] applied to [0,1,2] -> [0,2,1].
-    for (auto& r : q->second) {
-      // make sure the replacement value doesn't already appear
-      bool exists = false;
-      ssize_t pos = -1;
-      for (unsigned i = 0; i < raw->size(); ++i) {
-       int osd = (*raw)[i];
-       if (osd == r.second) {
-         exists = true;
-         break;
-       }
-       // ignore mapping if target is marked out (or invalid osd id)
-       if (osd == r.first &&
-           pos < 0 &&
-           !(r.second != CRUSH_ITEM_NONE && r.second < max_osd &&
-             osd_weight[r.second] == 0)) {
-         pos = i;
-       }
-      }
-      if (!exists && pos >= 0) {
-       (*raw)[pos] = r.second;
+    for (auto& i : *raw) {
+      for (auto& r : q->second) {
+        if (r.first != i) {
+          continue;
+        }
+        if (!(r.second != CRUSH_ITEM_NONE &&
+              r.second < max_osd &&
+              osd_weight[r.second] == 0)) {
+          i = r.second;
+        }
+        break;
       }
     }
   }