Merge pull request ceph#62223 from aainscow/enable-switch-flag

aainscow · web-flow · commit 49225d30fc9a · 2025-04-03T10:21:55.000+01:00
osd:  Add ability to switch EC optimizations on 

Reviewed-by: Bill Scales &lt;bill_scales@uk.ibm.com&gt;|
Reviewed-by: Radoslaw Zarzynski &lt;rzarzyns@redhat.com&gt;
Reviewed-by: Samuel Just &lt;sjust@redhat.com&gt;
diff --git a/src/common/io_exerciser/RadosIo.cc b/src/common/io_exerciser/RadosIo.cc
@@ -41,7 +41,7 @@ RadosIo::RadosIo(librados::Rados& rados, boost::asio::io_context& asio,
                  const std::string& pool, const std::string& oid,
                  const std::optional<std::vector<int>>& cached_shard_order,
                  uint64_t block_size, int seed, int threads, ceph::mutex& lock,
-                 ceph::condition_variable& cond)
+                 ceph::condition_variable& cond, bool ec_optimizations)
     : Model(oid, block_size),
       rados(rados),
       asio(asio),
@@ -58,6 +58,9 @@ RadosIo::RadosIo(librados::Rados& rados, boost::asio::io_context& asio,
   rc = rados.ioctx_create(pool.c_str(), io);
   ceph_assert(rc == 0);
   allow_ec_overwrites(true);
+  if (ec_optimizations) {
+    allow_ec_optimizations();
+  }
 }
 
 RadosIo::~RadosIo() {}
@@ -92,6 +95,17 @@ void RadosIo::allow_ec_overwrites(bool allow) {
   ceph_assert(rc == 0);
 }
 
+void RadosIo::allow_ec_optimizations()
+{
+  int rc;
+  bufferlist inbl, outbl;
+  std::string cmdstr =
+    "{\"prefix\": \"osd pool set\", \"pool\": \"" + pool + "\", \
+      \"var\": \"allow_ec_optimizations\", \"val\": \"true\"}";
+  rc = rados.mon_command(cmdstr, inbl, &outbl, nullptr);
+  ceph_assert(rc == 0);
+}
+
 template <int N>
 RadosIo::AsyncOpInfo<N>::AsyncOpInfo(const std::array<uint64_t, N>& offset,
                                      const std::array<uint64_t, N>& length)
diff --git a/src/common/io_exerciser/RadosIo.h b/src/common/io_exerciser/RadosIo.h
@@ -42,11 +42,12 @@ class RadosIo : public Model {
           const std::string& pool, const std::string& oid,
           const std::optional<std::vector<int>>& cached_shard_order,
           uint64_t block_size, int seed, int threads, ceph::mutex& lock,
-          ceph::condition_variable& cond);
+          ceph::condition_variable& cond, bool ec_optimizations);
 
   ~RadosIo();
 
   void allow_ec_overwrites(bool allow);
+  void allow_ec_optimizations();
 
   template <int N>
   class AsyncOpInfo {
diff --git a/src/mon/MonCommands.h b/src/mon/MonCommands.h
@@ -1170,11 +1170,11 @@ 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|pg_autoscale_mode|pg_autoscale_bias|pg_num_min|pg_num_max|target_size_bytes|target_size_ratio|dedup_tier|dedup_chunk_algorithm|dedup_cdc_chunk_size|eio|bulk|read_ratio|pct_update_delay",
+	"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_autoscale_bias|pg_num_min|pg_num_max|target_size_bytes|target_size_ratio|dedup_tier|dedup_chunk_algorithm|dedup_cdc_chunk_size|eio|bulk|read_ratio|pct_update_delay|allow_ec_optimizations",
 	"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|pgp_num_actual|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_autoscale_bias|pg_num_min|pg_num_max|target_size_bytes|target_size_ratio|dedup_tier|dedup_chunk_algorithm|dedup_cdc_chunk_size|eio|bulk|read_ratio|pct_update_delay "
+	"name=var,type=CephChoices,strings=size|min_size|pg_num|pgp_num|pgp_num_actual|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_autoscale_bias|pg_num_min|pg_num_max|target_size_bytes|target_size_ratio|dedup_tier|dedup_chunk_algorithm|dedup_cdc_chunk_size|eio|bulk|read_ratio|pct_update_delay|allow_ec_optimizations "
 	"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
@@ -5410,7 +5410,8 @@ namespace {
     CSUM_TYPE, CSUM_MAX_BLOCK, CSUM_MIN_BLOCK, FINGERPRINT_ALGORITHM,
     PG_AUTOSCALE_MODE, PG_NUM_MIN, TARGET_SIZE_BYTES, TARGET_SIZE_RATIO,
     PG_AUTOSCALE_BIAS, DEDUP_TIER, DEDUP_CHUNK_ALGORITHM, 
-    DEDUP_CDC_CHUNK_SIZE, POOL_EIO, BULK, PG_NUM_MAX, READ_RATIO };
+    DEDUP_CDC_CHUNK_SIZE, POOL_EIO, BULK, PG_NUM_MAX, READ_RATIO,
+    EC_OPTIMIZATIONS };
 
   std::set<osd_pool_get_choices>
     subtract_second_from_first(const std::set<osd_pool_get_choices>& first,
@@ -6215,7 +6216,8 @@ bool OSDMonitor::preprocess_command(MonOpRequestRef op)
       {"dedup_chunk_algorithm", DEDUP_CHUNK_ALGORITHM},
       {"dedup_cdc_chunk_size", DEDUP_CDC_CHUNK_SIZE},
       {"bulk", BULK},
-      {"read_ratio", READ_RATIO}
+      {"read_ratio", READ_RATIO},
+      {"allow_ec_optimizations", EC_OPTIMIZATIONS}
     };
 
     typedef std::set<osd_pool_get_choices> choices_set_t;
@@ -6230,7 +6232,7 @@ bool OSDMonitor::preprocess_command(MonOpRequestRef op)
       HIT_SET_GRADE_DECAY_RATE, HIT_SET_SEARCH_LAST_N
     };
     const choices_set_t ONLY_ERASURE_CHOICES = {
-      EC_OVERWRITES, ERASURE_CODE_PROFILE
+      EC_OVERWRITES, ERASURE_CODE_PROFILE, EC_OPTIMIZATIONS
     };
     const choices_set_t ONLY_REPLICA_CHOICES = {
       READ_RATIO
@@ -6462,17 +6464,23 @@ bool OSDMonitor::preprocess_command(MonOpRequestRef op)
 	  case DEDUP_CHUNK_ALGORITHM:
 	  case DEDUP_CDC_CHUNK_SIZE:
           case READ_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) {
-                int64_t val;
-                p->opts.get(pool_opts_t::CSUM_TYPE, &val);
-                f->dump_string(i->first.c_str(), Checksummer::get_csum_type_string(val));
-              } else {
-                p->opts.dump(i->first, f.get());
-              }
+	    {
+	      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) {
+		  int64_t val;
+		  p->opts.get(pool_opts_t::CSUM_TYPE, &val);
+		  f->dump_string(i->first.c_str(), Checksummer::get_csum_type_string(val));
+		} else {
+		  p->opts.dump(i->first, f.get());
+		}
+	      }
 	    }
             break;
+	  case EC_OPTIMIZATIONS:
+	    f->dump_bool("allow_ec_optimizations",
+			 p->has_flag(pg_pool_t::FLAG_EC_OPTIMIZATIONS));
+	    break;
 	}
       }
       f->close_section();
@@ -6644,6 +6652,11 @@ bool OSDMonitor::preprocess_command(MonOpRequestRef op)
 	      }
 	    }
 	    break;
+	  case EC_OPTIMIZATIONS:
+	    ss << "allow_ec_optimizations: " <<
+	      (p->has_flag(pg_pool_t::FLAG_EC_OPTIMIZATIONS) ? "true" : "false") <<
+	      "\n";
+	    break;
 	}
 	rdata.append(ss.str());
 	ss.str("");
@@ -8796,8 +8809,31 @@ int OSDMonitor::prepare_command_pool_set(const cmdmap_t& cmdmap,
     if (val == "true" || (interr.empty() && n == 1)) {
 	p.flags |= pg_pool_t::FLAG_EC_OVERWRITES;
     } else if (val == "false" || (interr.empty() && n == 0)) {
-      ss << "ec overwrites cannot be disabled once enabled";
+      if ((p.flags & pg_pool_t::FLAG_EC_OVERWRITES) != 0) {
+	ss << "ec overwrites cannot be disabled once enabled";
+	return -EINVAL;
+      }
+    } else {
+      ss << "expecting value 'true', 'false', '0', or '1'";
       return -EINVAL;
+    }
+  } else if (var == "allow_ec_optimizations") {
+    if (!p.is_erasure()) {
+      ss << "allow_ec_optimizations can only be enabled for an erasure coded pool";
+      return -EINVAL;
+    }
+    if (osdmap.require_osd_release < ceph_release_t::tentacle) {
+      ss << "All OSDs must be upgraded to tentacle or "
+           << "later before setting allow_ec_optimizations";
+        return -EINVAL;
+      }
+    if (val == "true" || (interr.empty() && n == 1)) {
+      p.flags |= pg_pool_t::FLAG_EC_OPTIMIZATIONS;
+    } else if (val == "false" || (interr.empty() && n == 0)) {
+      if ((p.flags & pg_pool_t::FLAG_EC_OPTIMIZATIONS) != 0) {
+	ss << "allow_ec_optimizations cannot be disabled once enabled";
+	return -EINVAL;
+      }
     } else {
       ss << "expecting value 'true', 'false', '0', or '1'";
       return -EINVAL;
diff --git a/src/osd/ECSwitch.h b/src/osd/ECSwitch.h
@@ -29,7 +29,7 @@ class ECSwitch : public PGBackend
   friend class ECReadPred;
 
   ECLegacy::ECBackendL legacy;
-  ECLegacy::ECBackendL optimized;
+  ECBackend optimized;
   bool is_optimized_actual;
 
 public:
@@ -44,12 +44,13 @@ class ECSwitch : public PGBackend
     PGBackend(cct, pg, store, coll, ch),
     legacy(pg, cct, ec_impl, stripe_width, this),
     optimized(pg, cct, ec_impl, stripe_width, this),
-    is_optimized_actual(false) {}
+    is_optimized_actual(get_parent()->get_pool().allows_ecoptimizations()) {}
 
   bool is_optimized() const
   {
-    // FIXME: Interface not yet implemented.
-    //ceph_assert(is_optimized_actual == get_parent()->get_pool().allows_ecoptimizations());
+    // FIXME: Once we trust this, we can remove this assert, as it adds
+    //        function call overhead.
+    ceph_assert(is_optimized_actual == get_parent()->get_pool().allows_ecoptimizations());
     return is_optimized_actual;
   }
 
@@ -84,7 +85,7 @@ class ECSwitch : public PGBackend
   private:
     const ECSwitch *switcher;
     std::unique_ptr<ECLegacy::ECBackendL::ECRecPred> legacy;
-    std::unique_ptr<ECLegacy::ECBackendL::ECRecPred> optimized;
+    std::unique_ptr<ECBackend::ECRecPred> optimized;
   };
 
   class ECReadPred : public IsPGReadablePredicate
@@ -111,7 +112,7 @@ class ECSwitch : public PGBackend
   private:
     const ECSwitch *switcher;
     std::unique_ptr<ECLegacy::ECBackendL::ECReadPred> legacy;
-    std::unique_ptr<ECLegacy::ECBackendL::ECReadPred> optimized;
+    std::unique_ptr<ECBackend::ECReadPred> optimized;
   };
 
   RecoveryHandle *open_recovery_op() override
@@ -181,7 +182,11 @@ class ECSwitch : public PGBackend
     else {
       legacy.on_change();
     }
-    //FIXME: Switch to new EC here.
+
+    if (!is_optimized_actual)
+      is_optimized_actual = get_parent()->get_pool().allows_ecoptimizations();
+    else
+      ceph_assert(get_parent()->get_pool().allows_ecoptimizations());
   }
 
   void clear_recovery_state() override
@@ -345,7 +350,7 @@ class ECSwitch : public PGBackend
   object_size_to_shard_size(const uint64_t size, int shard) const override
   {
     if (is_optimized()) {
-      return optimized.object_size_to_shard_size(size);
+      return optimized.object_size_to_shard_size(size, shard);
     }
     return legacy.object_size_to_shard_size(size);
     // All shards are the same size.
diff --git a/src/osd/osd_types.cc b/src/osd/osd_types.cc
@@ -4199,6 +4199,8 @@ bool PastIntervals::is_new_interval(
   uint32_t new_crush_barrier,
   int32_t old_crush_member,
   int32_t new_crush_member,
+  bool old_allow_ec_optimizations,
+  bool new_allow_ec_optimizations,
   pg_t pgid) {
   return old_acting_primary != new_acting_primary ||
     new_acting != old_acting ||
@@ -4222,7 +4224,8 @@ bool PastIntervals::is_new_interval(
     old_crush_count != new_crush_count ||
     old_crush_target != new_crush_target ||
     old_crush_barrier != new_crush_barrier ||
-    old_crush_member != new_crush_member;
+    old_crush_member != new_crush_member ||
+    old_allow_ec_optimizations != new_allow_ec_optimizations;
 }
 
 bool PastIntervals::is_new_interval(
@@ -4271,6 +4274,7 @@ bool PastIntervals::is_new_interval(
 		    plast->peering_crush_bucket_target, pi->peering_crush_bucket_target,
 		    plast->peering_crush_bucket_barrier, pi->peering_crush_bucket_barrier,
 		    plast->peering_crush_mandatory_member, pi->peering_crush_mandatory_member,
+		    plast->allows_ecoptimizations(), pi->allows_ecoptimizations(),
 		    pgid);
 }
 
diff --git a/src/osd/osd_types.h b/src/osd/osd_types.h
@@ -1297,6 +1297,7 @@ struct pg_pool_t {
     // Pool features are restricted to those supported by crimson-osd.
     // Note, does not prohibit being created on classic osd.
     FLAG_CRIMSON = 1<<18,
+    FLAG_EC_OPTIMIZATIONS = 1<<19, // enable optimizations, once enabled, cannot be disabled
   };
 
   static const char *get_flag_name(uint64_t f) {
@@ -1320,6 +1321,7 @@ struct pg_pool_t {
     case FLAG_EIO: return "eio";
     case FLAG_BULK: return "bulk";
     case FLAG_CRIMSON: return "crimson";
+    case FLAG_EC_OPTIMIZATIONS: return "ec_optimizations";
     default: return "???";
     }
   }
@@ -1376,6 +1378,8 @@ struct pg_pool_t {
       return FLAG_BULK;
     if (name == "crimson")
       return FLAG_CRIMSON;
+    if (name == "ec_optimizations")
+      return FLAG_EC_OPTIMIZATIONS;
     return 0;
   }
 
@@ -1790,6 +1794,10 @@ struct pg_pool_t {
     return has_flag(FLAG_EC_OVERWRITES);
   }
 
+  bool allows_ecoptimizations() const {
+    return has_flag(FLAG_EC_OPTIMIZATIONS);
+  }
+
   bool is_crimson() const {
     return has_flag(FLAG_CRIMSON);
   }
@@ -3453,6 +3461,8 @@ class PastIntervals {
     uint32_t new_crush_barrier,
     int32_t old_crush_member,
     int32_t new_crush_member,
+    bool old_allow_ec_optimizations,
+    bool new_allow_ec_optimizations,
     pg_t pgid
     );
 
diff --git a/src/osdc/Objecter.cc b/src/osdc/Objecter.cc
@@ -2934,6 +2934,8 @@ int Objecter::_calc_target(op_target_t *t, Connection *con, bool any_change)
 	pi->peering_crush_bucket_barrier,
 	t->peering_crush_mandatory_member,
 	pi->peering_crush_mandatory_member,
+	t->allows_ecoptimizations,
+	pi->allows_ecoptimizations(),
 	prev_pgid)) {
     force_resend = true;
   }
@@ -2989,6 +2991,7 @@ int Objecter::_calc_target(op_target_t *t, Connection *con, bool any_change)
     t->peering_crush_bucket_target = pi->peering_crush_bucket_target;
     t->peering_crush_bucket_barrier = pi->peering_crush_bucket_barrier;
     t->peering_crush_mandatory_member = pi->peering_crush_mandatory_member;
+    t->allows_ecoptimizations = pi->allows_ecoptimizations();
     ldout(cct, 10) << __func__ << " "
 		   << " raw pgid " << pgid << " -> actual " << t->actual_pgid
 		   << " acting " << t->acting
diff --git a/src/osdc/Objecter.h b/src/osdc/Objecter.h
@@ -1856,6 +1856,7 @@ class Objecter : public md_config_obs_t, public Dispatcher {
     int min_size = -1; ///< the min size of the pool when were were last mapped
     bool sort_bitwise = false; ///< whether the hobject_t sort order is bitwise
     bool recovery_deletes = false; ///< whether the deletes are performed during recovery instead of peering
+    bool allows_ecoptimizations = false; ///< whether EC plugin optimizations are enabled.
     uint32_t peering_crush_bucket_count = 0;
     uint32_t peering_crush_bucket_target = 0;
     uint32_t peering_crush_bucket_barrier = 0;
diff --git a/src/test/osd/ceph_test_rados_io_sequence/ceph_test_rados_io_sequence.cc b/src/test/osd/ceph_test_rados_io_sequence/ceph_test_rados_io_sequence.cc
diff --git a/src/test/osd/ceph_test_rados_io_sequence/ceph_test_rados_io_sequence.h b/src/test/osd/ceph_test_rados_io_sequence/ceph_test_rados_io_sequence.h
