osd/osd_types: add pool pg_autoscale_mode, pg_num_min, target_size_{bytes,ratio}...

These will direct PG autoscaling.

Signed-off-by: Sage Weil <sage@redhat.com>

diff --git a/src/mon/MonCommands.h b/src/mon/MonCommands.h
index 184a5aed11458502eee7d662baacfa7d605d092a..4606ced6cd2d53e59c0075ce38722e9d00afff05 100644 (file)
--- a/src/mon/MonCommands.h
+++ b/src/mon/MonCommands.h
@@ -963,11 +963,11 @@ COMMAND("osd pool rename " \
        "rename <srcpool> to <destpool>", "osd", "rw")
 COMMAND("osd pool get " \
        "name=pool,type=CephPoolname " \
-       "name=var,type=CephChoices,strings=size|min_size|pg_num|pgp_num|crush_rule|hashpspool|nodelete|nopgchange|nosizechange|write_fadvise_dontneed|noscrub|nodeep-scrub|hit_set_type|hit_set_period|hit_set_count|hit_set_fpp|use_gmt_hitset|target_max_objects|target_max_bytes|cache_target_dirty_ratio|cache_target_dirty_high_ratio|cache_target_full_ratio|cache_min_flush_age|cache_min_evict_age|erasure_code_profile|min_read_recency_for_promote|all|min_write_recency_for_promote|fast_read|hit_set_grade_decay_rate|hit_set_search_last_n|scrub_min_interval|scrub_max_interval|deep_scrub_interval|recovery_priority|recovery_op_priority|scrub_priority|compression_mode|compression_algorithm|compression_required_ratio|compression_max_blob_size|compression_min_blob_size|csum_type|csum_min_block|csum_max_block|allow_ec_overwrites|fingerprint_algorithm", \
+       "name=var,type=CephChoices,strings=size|min_size|pg_num|pgp_num|crush_rule|hashpspool|nodelete|nopgchange|nosizechange|write_fadvise_dontneed|noscrub|nodeep-scrub|hit_set_type|hit_set_period|hit_set_count|hit_set_fpp|use_gmt_hitset|target_max_objects|target_max_bytes|cache_target_dirty_ratio|cache_target_dirty_high_ratio|cache_target_full_ratio|cache_min_flush_age|cache_min_evict_age|erasure_code_profile|min_read_recency_for_promote|all|min_write_recency_for_promote|fast_read|hit_set_grade_decay_rate|hit_set_search_last_n|scrub_min_interval|scrub_max_interval|deep_scrub_interval|recovery_priority|recovery_op_priority|scrub_priority|compression_mode|compression_algorithm|compression_required_ratio|compression_max_blob_size|compression_min_blob_size|csum_type|csum_min_block|csum_max_block|allow_ec_overwrites|fingerprint_algorithm|pg_autoscale_mode|pg_num_min|target_size_bytes|target_size_ratio", \
        "get pool parameter <var>", "osd", "r")
 COMMAND("osd pool set " \
        "name=pool,type=CephPoolname " \
-       "name=var,type=CephChoices,strings=size|min_size|pg_num|pgp_num|crush_rule|hashpspool|nodelete|nopgchange|nosizechange|write_fadvise_dontneed|noscrub|nodeep-scrub|hit_set_type|hit_set_period|hit_set_count|hit_set_fpp|use_gmt_hitset|target_max_bytes|target_max_objects|cache_target_dirty_ratio|cache_target_dirty_high_ratio|cache_target_full_ratio|cache_min_flush_age|cache_min_evict_age|min_read_recency_for_promote|min_write_recency_for_promote|fast_read|hit_set_grade_decay_rate|hit_set_search_last_n|scrub_min_interval|scrub_max_interval|deep_scrub_interval|recovery_priority|recovery_op_priority|scrub_priority|compression_mode|compression_algorithm|compression_required_ratio|compression_max_blob_size|compression_min_blob_size|csum_type|csum_min_block|csum_max_block|allow_ec_overwrites|fingerprint_algorithm " \
+       "name=var,type=CephChoices,strings=size|min_size|pg_num|pgp_num|crush_rule|hashpspool|nodelete|nopgchange|nosizechange|write_fadvise_dontneed|noscrub|nodeep-scrub|hit_set_type|hit_set_period|hit_set_count|hit_set_fpp|use_gmt_hitset|target_max_bytes|target_max_objects|cache_target_dirty_ratio|cache_target_dirty_high_ratio|cache_target_full_ratio|cache_min_flush_age|cache_min_evict_age|min_read_recency_for_promote|min_write_recency_for_promote|fast_read|hit_set_grade_decay_rate|hit_set_search_last_n|scrub_min_interval|scrub_max_interval|deep_scrub_interval|recovery_priority|recovery_op_priority|scrub_priority|compression_mode|compression_algorithm|compression_required_ratio|compression_max_blob_size|compression_min_blob_size|csum_type|csum_min_block|csum_max_block|allow_ec_overwrites|fingerprint_algorithm|pg_autoscale_mode|pg_num_min|target_size_bytes|target_size_ratio " \
        "name=val,type=CephString " \
        "name=yes_i_really_mean_it,type=CephBool,req=false", \
        "set pool parameter <var> to <val>", "osd", "rw")

diff --git a/src/mon/OSDMonitor.cc b/src/mon/OSDMonitor.cc
index b9a64849d1bd32030b087aad5445b3bbca7805a2..4cebaecbfb085de72e345a38802654ebbcfa7622 100644 (file)
--- a/src/mon/OSDMonitor.cc
+++ b/src/mon/OSDMonitor.cc
@@ -4499,7 +4499,8 @@ namespace {
     RECOVERY_PRIORITY, RECOVERY_OP_PRIORITY, SCRUB_PRIORITY,
     COMPRESSION_MODE, COMPRESSION_ALGORITHM, COMPRESSION_REQUIRED_RATIO,
     COMPRESSION_MAX_BLOB_SIZE, COMPRESSION_MIN_BLOB_SIZE,
-    CSUM_TYPE, CSUM_MAX_BLOCK, CSUM_MIN_BLOCK, FINGERPRINT_ALGORITHM };
+    CSUM_TYPE, CSUM_MAX_BLOCK, CSUM_MIN_BLOCK, FINGERPRINT_ALGORITHM,
+    PG_AUTOSCALE_MODE, PG_NUM_MIN, TARGET_SIZE_BYTES, TARGET_SIZE_RATIO };
 
   std::set<osd_pool_get_choices>
     subtract_second_from_first(const std::set<osd_pool_get_choices>& first,
@@ -5107,6 +5108,10 @@ bool OSDMonitor::preprocess_command(MonOpRequestRef op)
       {"csum_max_block", CSUM_MAX_BLOCK},
       {"csum_min_block", CSUM_MIN_BLOCK},
       {"fingerprint_algorithm", FINGERPRINT_ALGORITHM},
+      {"pg_autoscale_mode", PG_AUTOSCALE_MODE},
+      {"pg_num_min", PG_NUM_MIN},
+      {"target_size_bytes", TARGET_SIZE_BYTES},
+      {"target_size_ratio", TARGET_SIZE_RATIO},
     };
 
     typedef std::set<osd_pool_get_choices> choices_set_t;
@@ -5209,6 +5214,11 @@ bool OSDMonitor::preprocess_command(MonOpRequestRef op)
            f->dump_bool("allow_ec_overwrites",
                          p->has_flag(pg_pool_t::FLAG_EC_OVERWRITES));
            break;
+         case PG_AUTOSCALE_MODE:
+           f->dump_string("pg_autoscale_mode",
+                          pg_pool_t::get_pg_autoscale_mode_name(
+                            p->pg_autoscale_mode));
+           break;
          case HASHPSPOOL:
          case NODELETE:
          case NOPGCHANGE:
@@ -5314,6 +5324,9 @@ bool OSDMonitor::preprocess_command(MonOpRequestRef op)
          case CSUM_MAX_BLOCK:
          case CSUM_MIN_BLOCK:
          case FINGERPRINT_ALGORITHM:
+         case PG_NUM_MIN:
+         case TARGET_SIZE_BYTES:
+         case TARGET_SIZE_RATIO:
             pool_opts_t::key_t key = pool_opts_t::get_opt_desc(i->first).key;
             if (p->opts.is_set(key)) {
               if(*it == CSUM_TYPE) {
@@ -5354,6 +5367,10 @@ bool OSDMonitor::preprocess_command(MonOpRequestRef op)
              ss << "crush_rule: " << p->get_crush_rule() << "\n";
            }
            break;
+         case PG_AUTOSCALE_MODE:
+           ss << "pg_autoscale_mode: " << pg_pool_t::get_pg_autoscale_mode_name(
+             p->pg_autoscale_mode);
+           break;
          case HIT_SET_PERIOD:
            ss << "hit_set_period: " << p->hit_set_period << "\n";
            break;
@@ -5463,6 +5480,9 @@ bool OSDMonitor::preprocess_command(MonOpRequestRef op)
          case CSUM_MAX_BLOCK:
          case CSUM_MIN_BLOCK:
          case FINGERPRINT_ALGORITHM:
+         case PG_NUM_MIN:
+         case TARGET_SIZE_BYTES:
+         case TARGET_SIZE_RATIO:
            for (i = ALL_CHOICES.begin(); i != ALL_CHOICES.end(); ++i) {
              if (i->second == *it)
                break;
@@ -7080,6 +7100,13 @@ int OSDMonitor::prepare_command_pool_set(const cmdmap_t& cmdmap,
       return -EINVAL;
     }
     p.set_pgp_num_target(n);
+  } else if (var == "pg_autoscale_mode") {
+    n = pg_pool_t::get_pg_autoscale_mode_by_name(val);
+    if (n < 0) {
+      ss << "specified invalid mode " << val;
+      return -EINVAL;
+    }
+    p.pg_autoscale_mode = n;
   } else if (var == "crush_rule") {
     int id = osdmap.crush->get_rule_id(val);
     if (id == -ENOENT) {
@@ -7357,6 +7384,16 @@ int OSDMonitor::prepare_command_pool_set(const cmdmap_t& cmdmap,
          return -EINVAL;
         }
       }
+    } else if (var == "pg_num_min") {
+      if (interr.length()) {
+        ss << "error parsing int value '" << val << "': " << interr;
+        return -EINVAL;
+      }
+      if (n > (int)p.get_pg_num_target()) {
+       ss << "specified pg_num_min " << n
+          << " > pg_num " << p.get_pg_num_target();
+       return -EINVAL;
+      }
     }
 
     pool_opts_t::opt_desc_t desc = pool_opts_t::get_opt_desc(var);

diff --git a/src/osd/osd_types.cc b/src/osd/osd_types.cc
index 2b5485368008eca1a616b3ab405834d3e216126b..102c3a80dff4d7d941c607af34c250b3efb69fca 100644 (file)
--- a/src/osd/osd_types.cc
+++ b/src/osd/osd_types.cc
@@ -1110,7 +1110,13 @@ static opt_mapping_t opt_mapping = boost::assign::map_list_of
            ("csum_min_block", pool_opts_t::opt_desc_t(
             pool_opts_t::CSUM_MIN_BLOCK, pool_opts_t::INT))
            ("fingerprint_algorithm", pool_opts_t::opt_desc_t(
-            pool_opts_t::FINGERPRINT_ALGORITHM, pool_opts_t::STR));
+            pool_opts_t::FINGERPRINT_ALGORITHM, pool_opts_t::STR))
+           ("pg_num_min", pool_opts_t::opt_desc_t(
+            pool_opts_t::PG_NUM_MIN, pool_opts_t::INT))
+           ("target_size_bytes", pool_opts_t::opt_desc_t(
+            pool_opts_t::TARGET_SIZE_BYTES, pool_opts_t::INT))
+           ("target_size_ratio", pool_opts_t::opt_desc_t(
+            pool_opts_t::TARGET_SIZE_RATIO, pool_opts_t::DOUBLE));
 
 bool pool_opts_t::is_opt_name(const std::string& name) {
     return opt_mapping.count(name);
@@ -1273,6 +1279,8 @@ void pg_pool_t::dump(Formatter *f) const
   f->dump_int("min_size", get_min_size());
   f->dump_int("crush_rule", get_crush_rule());
   f->dump_int("object_hash", get_object_hash());
+  f->dump_string("pg_autoscale_mode",
+                get_pg_autoscale_mode_name(pg_autoscale_mode));
   f->dump_unsigned("pg_num", get_pg_num());
   f->dump_unsigned("pg_placement_num", get_pgp_num());
   f->dump_unsigned("pg_placement_num_target", get_pgp_num_target());
@@ -1786,6 +1794,7 @@ void pg_pool_t::encode(bufferlist& bl, uint64_t features) const
     encode(pg_num_dec_last_epoch_started, bl);
     encode(pg_num_dec_last_epoch_clean, bl);
     encode(last_force_op_resend, bl);
+    encode(pg_autoscale_mode, bl);
   }
   ENCODE_FINISH(bl);
 }
@@ -1952,11 +1961,13 @@ void pg_pool_t::decode(bufferlist::const_iterator& bl)
     decode(pg_num_dec_last_epoch_started, bl);
     decode(pg_num_dec_last_epoch_clean, bl);
     decode(last_force_op_resend, bl);
+    decode(pg_autoscale_mode, bl);
   } else {
     pg_num_target = pg_num;
     pgp_num_target = pgp_num;
     pg_num_pending = pg_num;
     last_force_op_resend = last_force_op_resend_prenautilus;
+    pg_autoscale_mode = PG_AUTOSCALE_MODE_WARN;    // default to warn on upgrade
   }
   DECODE_FINISH(bl);
   calc_pg_masks();
@@ -2055,6 +2066,9 @@ ostream& operator<<(ostream& out, const pg_pool_t& p)
       out << " dles/c " << p.get_pg_num_dec_last_epoch_started()
          << "/" << p.get_pg_num_dec_last_epoch_clean();
   }
+  if (p.pg_autoscale_mode) {
+    out << " autoscale_mode " << p.get_pg_autoscale_mode_name(p.pg_autoscale_mode);
+  }
   out << " last_change " << p.get_last_change();
   if (p.get_last_force_op_resend() ||
       p.get_last_force_op_resend_prenautilus() ||

diff --git a/src/osd/osd_types.h b/src/osd/osd_types.h
index 0a82a0ac356d6d0103e671a26f9e9ba6d76e1216..3b48e4d567ef0f06f95eff3d65147dcb93383ae2 100644 (file)
--- a/src/osd/osd_types.h
+++ b/src/osd/osd_types.h
@@ -1007,6 +1007,9 @@ public:
     CSUM_MAX_BLOCK,
     CSUM_MIN_BLOCK,
     FINGERPRINT_ALGORITHM,
+    PG_NUM_MIN,         // min pg_num
+    TARGET_SIZE_BYTES,  // total bytes in pool
+    TARGET_SIZE_RATIO,  // fraction of total cluster
   };
 
   enum type_t {
@@ -1244,12 +1247,39 @@ struct pg_pool_t {
     }
   }
 
+  enum {
+    PG_AUTOSCALE_MODE_OFF = 0,
+    PG_AUTOSCALE_MODE_WARN = 1,
+    PG_AUTOSCALE_MODE_ON = 2,
+  };
+  static const char *get_pg_autoscale_mode_name(int m) {
+    switch (m) {
+    case PG_AUTOSCALE_MODE_OFF: return "off";
+    case PG_AUTOSCALE_MODE_ON: return "on";
+    case PG_AUTOSCALE_MODE_WARN: return "warn";
+    default: return "???";
+    }
+  }
+  static int get_pg_autoscale_mode_by_name(const string& m) {
+    if (m == "off") {
+      return PG_AUTOSCALE_MODE_OFF;
+    }
+    if (m == "warn") {
+      return PG_AUTOSCALE_MODE_WARN;
+    }
+    if (m == "on") {
+      return PG_AUTOSCALE_MODE_ON;
+    }
+    return -1;
+  }
+
   utime_t create_time;
   uint64_t flags;           ///< FLAG_*
   __u8 type;                ///< TYPE_*
   __u8 size, min_size;      ///< number of osds in each pg
   __u8 crush_rule;          ///< crush placement rule
   __u8 object_hash;         ///< hash mapping object name to ps
+  __u8 pg_autoscale_mode;   ///< PG_AUTOSCALE_MODE_
 private:
   __u32 pg_num = 0, pgp_num = 0;  ///< number of pgs
   __u32 pg_num_pending = 0;       ///< pg_num we are about to merge down to