osd: Create more_underfull with below target that aren't in underfull

Use more_underfull for finding remaps for overfull OSDs

Signed-off-by: David Zafman <dzafman@redhat.com>
(cherry picked from commit fe03383e6691f6bd95189d3ba4d91fbb2ea9e9c8)

diff --git a/src/crush/CrushWrapper.cc b/src/crush/CrushWrapper.cc
index 45ff469a4ac0c5f5286563e4a50f55a71d55bc97..2b11ce9ec131ef0a36125fd916c470686ced546e 100644 (file)
--- a/src/crush/CrushWrapper.cc
+++ b/src/crush/CrushWrapper.cc
@@ -3785,6 +3785,7 @@ int CrushWrapper::_choose_type_stack(
   const vector<pair<int,int>>& stack,
   const set<int>& overfull,
   const vector<int>& underfull,
+  const vector<int>& more_underfull,
   const vector<int>& orig,
   vector<int>::const_iterator& i,
   set<int>& used,
@@ -3897,6 +3898,33 @@ int CrushWrapper::_choose_type_stack(
              ++i;
              break;
            }
+             if (!replaced) {
+             for (auto item : more_underfull) {
+               ldout(cct, 10) << __func__ << " more underfull pos " << pos
+                              << " was " << *i << " considering " << item
+                              << dendl;
+               if (used.count(item)) {
+                 ldout(cct, 20) << __func__ << "   in used " << used << dendl;
+                 continue;
+               }
+               if (!subtree_contains(from, item)) {
+                 ldout(cct, 20) << __func__ << "   not in subtree " << from << dendl;
+                 continue;
+               }
+               if (std::find(orig.begin(), orig.end(), item) != orig.end()) {
+                 ldout(cct, 20) << __func__ << "   in orig " << orig << dendl;
+                 continue;
+               }
+               o.push_back(item);
+               used.insert(item);
+               ldout(cct, 10) << __func__ << " pos " << pos << " replace "
+                              << *i << " -> " << item << dendl;
+               replaced = true;
+                assert(i != orig.end());
+               ++i;
+               break;
+             }
+           }
          }
          if (!replaced) {
            ldout(cct, 10) << __func__ << " pos " << pos << " keep " << *i
@@ -3974,6 +4002,7 @@ int CrushWrapper::try_remap_rule(
   int maxout,
   const set<int>& overfull,
   const vector<int>& underfull,
+  const vector<int>& more_underfull,
   const vector<int>& orig,
   vector<int> *out) const
 {
@@ -3983,7 +4012,9 @@ int CrushWrapper::try_remap_rule(
 
   ldout(cct, 10) << __func__ << " ruleno " << ruleno
                << " numrep " << maxout << " overfull " << overfull
-               << " underfull " << underfull << " orig " << orig
+               << " underfull " << underfull
+               << " more_underfull " << more_underfull
+               << " orig " << orig
                << dendl;
   vector<int> w; // working set
   out->clear();
@@ -4020,7 +4051,7 @@ int CrushWrapper::try_remap_rule(
        type_stack.push_back(make_pair(type, numrep));
         if (type > 0)
          type_stack.push_back(make_pair(0, 1));
-       int r = _choose_type_stack(cct, type_stack, overfull, underfull, orig,
+       int r = _choose_type_stack(cct, type_stack, overfull, underfull, more_underfull, orig,
                                   i, used, &w, root_bucket, ruleno);
        if (r < 0)
          return r;
@@ -4042,7 +4073,7 @@ int CrushWrapper::try_remap_rule(
     case CRUSH_RULE_EMIT:
       ldout(cct, 10) << " emit " << w << dendl;
       if (!type_stack.empty()) {
-       int r = _choose_type_stack(cct, type_stack, overfull, underfull, orig,
+       int r = _choose_type_stack(cct, type_stack, overfull, underfull, more_underfull, orig,
                                   i, used, &w, root_bucket, ruleno);
        if (r < 0)
          return r;

diff --git a/src/crush/CrushWrapper.h b/src/crush/CrushWrapper.h
index 826950cec994a478767d8a441c1472ce367260f5..1796ef69896e542965cc5f778ddca6302618f4a0 100644 (file)
--- a/src/crush/CrushWrapper.h
+++ b/src/crush/CrushWrapper.h
@@ -1580,6 +1580,7 @@ public:
     const vector<pair<int,int>>& stack,
     const set<int>& overfull,
     const vector<int>& underfull,
+    const vector<int>& more_underfull,
     const vector<int>& orig,
     vector<int>::const_iterator& i,
     set<int>& used,
@@ -1593,6 +1594,7 @@ public:
     int maxout,
     const set<int>& overfull,
     const vector<int>& underfull,
+    const vector<int>& more_underfull,
     const vector<int>& orig,
     vector<int> *out) const;
 

diff --git a/src/osd/OSDMap.cc b/src/osd/OSDMap.cc
index a8fa7694ac9481428e12924138337df1056aae2b..cce4b72f8481b63163c847f2c8d190dc7795073c 100644 (file)
--- a/src/osd/OSDMap.cc
+++ b/src/osd/OSDMap.cc
@@ -4398,6 +4398,7 @@ bool OSDMap::try_pg_upmap(
   pg_t pg,                       ///< pg to potentially remap
   const set<int>& overfull,      ///< osds we'd want to evacuate
   const vector<int>& underfull,  ///< osds to move to, in order of preference
+  const vector<int>& more_underfull,  ///< more osds only slightly underfull
   vector<int> *orig,
   vector<int> *out)              ///< resulting alternative mapping
 {
@@ -4426,6 +4427,7 @@ bool OSDMap::try_pg_upmap(
     rule,
     pool->get_size(),
     overfull, underfull,
+    more_underfull,
     *orig,
     out);
   if (r < 0)
@@ -4544,6 +4546,7 @@ int OSDMap::calc_pg_upmaps(
     // build overfull and underfull
     set<int> overfull;
     vector<int> underfull;
+    vector<int> more_underfull;
     for (auto i = deviation_osd.rbegin(); i != deviation_osd.rend(); i++) {
         ldout(cct, 30) << " check " << i->first << " <= " << max_deviation << dendl;
         if (i->first <= max_deviation)
@@ -4558,13 +4561,17 @@ int OSDMap::calc_pg_upmaps(
 
     for (auto i = deviation_osd.begin(); i != deviation_osd.end(); i++) {
         ldout(cct, 30) << " check " << i->first << " >= " << -(int)max_deviation << dendl;
-        if (i->first >= -(int)max_deviation)
+        if (i->first >= 0)
           break;
-       ldout(cct, 30) << " add underfull osd." << i->second << dendl;
-        underfull.push_back(i->second);
+        if (i->first < -(int)max_deviation) {
+         ldout(cct, 30) << " add underfull osd." << i->second << dendl;
+          underfull.push_back(i->second);
+       } else {
+          more_underfull.push_back(i->second);
+       }
     }
-    if (underfull.empty()) {
-      ldout(cct, 20) << __func__ << " failed to build underfull" << dendl;
+    if (underfull.empty() && more_underfull.empty()) {
+      ldout(cct, 20) << __func__ << " failed to build underfull and more_underfull" << dendl;
       break;
     }
 
@@ -4581,7 +4588,7 @@ int OSDMap::calc_pg_upmaps(
     auto temp_pgs_by_osd = pgs_by_osd;
     // always start with fullest, break if we find any changes to make
     for (auto p = deviation_osd.rbegin(); p != deviation_osd.rend(); ++p) {
-      if (skip_overfull) {
+      if (skip_overfull && !underfull.empty()) {
         ldout(cct, 10) << " skipping overfull " << dendl;
         break; // fall through to check underfull
       }
@@ -4693,7 +4700,7 @@ int OSDMap::calc_pg_upmaps(
        ldout(cct, 10) << " trying " << pg << dendl;
         vector<int> raw, orig, out;
         tmp.pg_to_raw_upmap(pg, &raw, &orig); // including existing upmaps too
-       if (!try_pg_upmap(cct, pg, overfull, underfull, &orig, &out)) {
+       if (!try_pg_upmap(cct, pg, overfull, underfull, more_underfull, &orig, &out)) {
          continue;
        }
        ldout(cct, 10) << " " << pg << " " << orig << " -> " << out << dendl;

diff --git a/src/osd/OSDMap.h b/src/osd/OSDMap.h
index 22fc939cbdfb48618063d90879c0c31db3eed7ea..a5414896f473ff3d8cd7228454b0e1d95997141b 100644 (file)
--- a/src/osd/OSDMap.h
+++ b/src/osd/OSDMap.h
@@ -1399,6 +1399,7 @@ public:
     pg_t pg,                       ///< pg to potentially remap
     const set<int>& overfull,      ///< osds we'd want to evacuate
     const vector<int>& underfull,  ///< osds to move to, in order of preference
+    const vector<int>& more_underfull,  ///< less full osds to move to, in order of preference
     vector<int> *orig,
     vector<int> *out);             ///< resulting alternative mapping
 

diff --git a/src/test/crush/CrushWrapper.cc b/src/test/crush/CrushWrapper.cc
index 575af8a01dd4f1035c545ce5f196131e6f62475b..6cfdd3df16cc5e6ed9360c1eeb0812c3cf39d4ad 100644 (file)
--- a/src/test/crush/CrushWrapper.cc
+++ b/src/test/crush/CrushWrapper.cc
@@ -1329,9 +1329,10 @@ TEST(CrushWrapper, try_remap_rule) {
     vector<int> orig = { 0, 3, 9 };
     set<int> overfull = { 3 };
     vector<int> underfull = { 0, 2, 5, 8, 11 };
+    vector<int> more_underfull = {};
     vector<int> out;
     int r = c.try_remap_rule(g_ceph_context, rule, 3,
-                             overfull, underfull,
+                             overfull, underfull, more_underfull,
                              orig, &out);
     cout << orig << " -> r = " << (int)r << " out " << out << std::endl;
     ASSERT_EQ(r, 0);
@@ -1345,7 +1346,7 @@ TEST(CrushWrapper, try_remap_rule) {
     orig = {1, 3, 9};
 
     r = c.try_remap_rule(g_ceph_context, rule, 3,
-                        overfull, underfull,
+                        overfull, underfull, more_underfull,
                         orig, &out);
     cout << orig << " -> r = " << (int)r << " out " << out << std::endl;
     ASSERT_EQ(r, 0);
@@ -1353,6 +1354,21 @@ TEST(CrushWrapper, try_remap_rule) {
     ASSERT_EQ(1, out[0]);
     ASSERT_EQ(0, out[1]);
     ASSERT_EQ(9, out[2]);
+    //
+    // Check that more_underfull is used when underfull runs out
+    orig = { 0, 3, 9 };
+    overfull = { 3, 9 };
+    underfull = { 2 };
+    more_underfull = { 5, 8, 11 };
+    r = c.try_remap_rule(g_ceph_context, rule, 3,
+                             overfull, underfull, more_underfull,
+                             orig, &out);
+    cout << orig << " -> r = " << (int)r << " out " << out << std::endl;
+    ASSERT_EQ(r, 0);
+    ASSERT_EQ(3u, out.size());
+    ASSERT_EQ(0, out[0]);
+    ASSERT_EQ(2, out[1]);
+    ASSERT_EQ(5, out[2]);
   }
 
   // chooseleaf
@@ -1366,9 +1382,10 @@ TEST(CrushWrapper, try_remap_rule) {
     vector<int> orig = { 0, 3, 9 };
     set<int> overfull = { 3 };
     vector<int> underfull = { 0, 2, 5, 8, 11 };
+    vector<int> more_underfull = { };
     vector<int> out;
     int r = c.try_remap_rule(g_ceph_context, rule, 3,
-                             overfull, underfull,
+                             overfull, underfull, more_underfull,
                              orig, &out);
     cout << orig << " -> r = " << (int)r << " out " << out << std::endl;
     ASSERT_EQ(r, 0);
@@ -1392,9 +1409,10 @@ TEST(CrushWrapper, try_remap_rule) {
     vector<int> orig = { 0, 3, 16, 12 };
     set<int> overfull = { 3, 12 };
     vector<int> underfull = { 6, 7, 9, 3, 0, 1, 15, 16, 13, 2, 5, 8, 11 };
+    vector<int> more_underfull = { };
     vector<int> out;
     int r = c.try_remap_rule(g_ceph_context, rule, 3,
-                             overfull, underfull,
+                             overfull, underfull, more_underfull,
                              orig, &out);
     cout << orig << " -> r = " << (int)r << " out " << out << std::endl;
     ASSERT_EQ(r, 0);
@@ -1407,7 +1425,7 @@ TEST(CrushWrapper, try_remap_rule) {
     orig.pop_back();
     out.clear();
     r = c.try_remap_rule(g_ceph_context, rule, 3,
-                        overfull, underfull,
+                        overfull, underfull, more_underfull,
                         orig, &out);
     cout << orig << " -> r = " << (int)r << " out " << out << std::endl;
     ASSERT_EQ(r, 0);