Osd: add two fields to pg_pool_t

Signed-off-by: MingXin Liu <mingxinliu@ubuntukylin.com>
Reviewed-by: Li Wang <liwang@ubuntukylin.com>

diff --git a/src/osd/osd_types.cc b/src/osd/osd_types.cc
index 849a70d59cdc1621cce81e31904ecba7a33f459b..3cf41bd2572a5712cdc6bc0d10f4551fd524addd 100644 (file)
--- a/src/osd/osd_types.cc
+++ b/src/osd/osd_types.cc
@@ -889,7 +889,6 @@ void pool_snap_info_t::generate_test_instances(list<pool_snap_info_t*>& o)
   o.back()->name = "foo";
 }
 
-
 // -- pg_pool_t --
 
 void pg_pool_t::dump(Formatter *f) const
@@ -947,6 +946,12 @@ void pg_pool_t::dump(Formatter *f) const
   f->dump_bool("use_gmt_hitset", use_gmt_hitset);
   f->dump_unsigned("min_read_recency_for_promote", min_read_recency_for_promote);
   f->dump_unsigned("min_write_recency_for_promote", min_write_recency_for_promote);
+  f->dump_unsigned("hit_set_grade_decay_rate", hit_set_grade_decay_rate);
+  f->dump_unsigned("hit_set_search_last_n", hit_set_search_last_n);
+  f->open_array_section("grade_table");
+  for (unsigned i = 0; i < hit_set_count; ++i)
+    f->dump_unsigned("value", get_grade(i));
+  f->close_section();
   f->dump_unsigned("stripe_width", get_stripe_width());
   f->dump_unsigned("expected_num_objects", expected_num_objects);
   f->dump_bool("fast_read", fast_read);
@@ -1259,7 +1264,7 @@ void pg_pool_t::encode(bufferlist& bl, uint64_t features) const
     return;
   }
 
-  ENCODE_START(22, 5, bl);
+  ENCODE_START(23, 5, bl);
   ::encode(type, bl);
   ::encode(size, bl);
   ::encode(crush_ruleset, bl);
@@ -1305,12 +1310,14 @@ void pg_pool_t::encode(bufferlist& bl, uint64_t features) const
   ::encode(min_write_recency_for_promote, bl);
   ::encode(use_gmt_hitset, bl);
   ::encode(fast_read, bl);
+  ::encode(hit_set_grade_decay_rate, bl);
+  ::encode(hit_set_search_last_n, bl);
   ENCODE_FINISH(bl);
 }
 
 void pg_pool_t::decode(bufferlist::iterator& bl)
 {
-  DECODE_START_LEGACY_COMPAT_LEN(22, 5, 5, bl);
+  DECODE_START_LEGACY_COMPAT_LEN(23, 5, 5, bl);
   ::decode(type, bl);
   ::decode(size, bl);
   ::decode(crush_ruleset, bl);
@@ -1442,8 +1449,16 @@ void pg_pool_t::decode(bufferlist::iterator& bl)
   } else {
     fast_read = false;
   }
+  if (struct_v >= 23) {
+    ::decode(hit_set_grade_decay_rate, bl);
+    ::decode(hit_set_search_last_n, bl);
+  } else {
+    hit_set_grade_decay_rate = 0;
+    hit_set_search_last_n = 1;
+  }
   DECODE_FINISH(bl);
   calc_pg_masks();
+  calc_grade_table();
 }
 
 void pg_pool_t::generate_test_instances(list<pg_pool_t*>& o)
@@ -1489,6 +1504,9 @@ void pg_pool_t::generate_test_instances(list<pg_pool_t*>& o)
   a.hit_set_count = 8;
   a.min_read_recency_for_promote = 1;
   a.min_write_recency_for_promote = 1;
+  a.hit_set_grade_decay_rate = 50;
+  a.hit_set_search_last_n = 1;
+  a.calc_grade_table();
   a.set_stripe_width(12345);
   a.target_max_bytes = 1238132132;
   a.target_max_objects = 1232132;
@@ -1542,7 +1560,9 @@ ostream& operator<<(ostream& out, const pg_pool_t& p)
   if (p.hit_set_params.get_type() != HitSet::TYPE_NONE) {
     out << " hit_set " << p.hit_set_params
        << " " << p.hit_set_period << "s"
-       << " x" << p.hit_set_count;
+       << " x" << p.hit_set_count << " decay_rate "
+       << p.hit_set_grade_decay_rate
+       << " search_last_n " << p.hit_set_search_last_n;
   }
   if (p.min_read_recency_for_promote)
     out << " min_read_recency_for_promote " << p.min_read_recency_for_promote;

diff --git a/src/osd/osd_types.h b/src/osd/osd_types.h
index adcceee497cf8bf2a7fb107e83b8896cf5fbc6f6..0f127f06dd43b74565d8dbbeebb9262332a3e54c 100644 (file)
--- a/src/osd/osd_types.h
+++ b/src/osd/osd_types.h
@@ -1104,6 +1104,9 @@ public:
     hit_set_params = HitSet::Params();
     hit_set_period = 0;
     hit_set_count = 0;
+    hit_set_grade_decay_rate = 0;
+    hit_set_search_last_n = 0;
+    grade_table.resize(0);
   }
 
   uint64_t target_max_bytes;   ///< tiering: target max pool size
@@ -1122,6 +1125,10 @@ public:
   bool use_gmt_hitset;         ///< use gmt to name the hitset archive object
   uint32_t min_read_recency_for_promote;   ///< minimum number of HitSet to check before promote on read
   uint32_t min_write_recency_for_promote;  ///< minimum number of HitSet to check before promote on write
+  uint32_t hit_set_grade_decay_rate;   ///< current hit_set has highest priority on objects
+                                       ///temperature count,the follow hit_set's priority decay 
+                                       ///by this params than pre hit_set
+  uint32_t hit_set_search_last_n;   ///<accumulate atmost N hit_sets for temperature
 
   uint32_t stripe_width;        ///< erasure coded stripe size in bytes
 
@@ -1129,6 +1136,24 @@ public:
                                  ///< user does not specify any expected value
   bool fast_read;            ///< whether turn on fast read on the pool or not
 
+private:
+  vector<uint32_t> grade_table;
+
+public:
+  uint32_t get_grade(unsigned i) const {
+    if (grade_table.size() <= i)
+      return 0;
+    return grade_table[i];
+  }
+  void calc_grade_table() {
+    unsigned v = 1000000;
+    grade_table.resize(hit_set_count);
+    for (unsigned i = 0; i < hit_set_count; i++) {
+      v = v * (1 - (hit_set_grade_decay_rate / 100.0));
+      grade_table[i] = v;
+    }
+  }
+
   pg_pool_t()
     : flags(0), type(0), size(0), min_size(0),
       crush_ruleset(0), object_hash(0),
@@ -1154,6 +1179,8 @@ public:
       use_gmt_hitset(true),
       min_read_recency_for_promote(0),
       min_write_recency_for_promote(0),
+      hit_set_grade_decay_rate(0),
+      hit_set_search_last_n(0),
       stripe_width(0),
       expected_num_objects(0),
       fast_read(false)