Merge pull request ceph#54930 from rzarzynski/wip-osd-ec-rework

osd: rework EC for the sake of integration with crimson

Author: ljflores

src/common/scrub_types.cc

@@ -55,10 +55,12 @@ static void encode(const osd_shard_t& shard, bufferlist& bl) {
 
 void shard_info_wrapper::set_object(const ScrubMap::object& object)
 {
-  for (auto attr : object.attrs) {
-    bufferlist bl;
-    bl.push_back(attr.second);
-    attrs.insert(std::make_pair(attr.first, std::move(bl)));
+  // logically no-op, changes the comparator from std::less<void>
+  // while avoiding `reinterpret_cast<const std::map<std::string,
+  // ceph::bufferlist>&>(object.attrs)`
+  attrs.clear();
+  for (const auto& kv : object.attrs) {
+    attrs.insert(kv);
   }
   size = object.size;
   if (object.omap_digest_present) {

src/crimson/osd/osd_operations/scrub_events.cc

@@ -221,7 +221,7 @@ ScrubScan::ifut<> ScrubScan::scan_object(
       if (i.second.length() == 0) {
 	entry.attrs[i.first];
       } else {
-	entry.attrs.emplace(i.first, i.second.front());
+	entry.attrs.emplace(i.first, i.second);
       }
     }
   }).handle_error_interruptible(

src/crimson/osd/pg_backend.h

@@ -451,4 +451,22 @@ class PGBackend
 		      std::vector<pg_log_entry_t>&& log_entries) = 0;
   friend class ReplicatedRecoveryBackend;
   friend class ::crimson::osd::PG;
+
+protected:
+  boost::container::flat_set<hobject_t> temp_contents;
+
+  template <class... Args>
+  void add_temp_obj(Args&&... args) {
+    temp_contents.insert(std::forward<Args>(args)...);
+  }
+  void clear_temp_obj(const hobject_t &oid) {
+    temp_contents.erase(oid);
+  }
+  template <class T>
+  void clear_temp_objs(const T &cont) {
+    for (const auto& oid : cont) {
+      clear_temp_obj(oid);
+    }
+  }
+  friend class RecoveryBackend;
 };

src/crimson/osd/recovery_backend.cc

@@ -32,6 +32,16 @@ hobject_t RecoveryBackend::get_temp_recovery_object(
   return hoid;
 }
 
+void RecoveryBackend::add_temp_obj(const hobject_t &oid)
+{
+  backend->add_temp_obj(oid);
+}
+
+void RecoveryBackend::clear_temp_obj(const hobject_t &oid)
+{
+  backend->clear_temp_obj(oid);
+}
+
 void RecoveryBackend::clean_up(ceph::os::Transaction& t,
 			       std::string_view why)
 {

src/crimson/osd/recovery_backend.h

@@ -201,12 +201,9 @@ class RecoveryBackend {
 
   boost::container::flat_set<hobject_t> temp_contents;
 
-  void add_temp_obj(const hobject_t &oid) {
-    temp_contents.insert(oid);
-  }
-  void clear_temp_obj(const hobject_t &oid) {
-    temp_contents.erase(oid);
-  }
+  void add_temp_obj(const hobject_t &oid);
+  void clear_temp_obj(const hobject_t &oid);
+
   void clean_up(ceph::os::Transaction& t, std::string_view why);
   virtual seastar::future<> on_stop() = 0;
 private:

src/crimson/osd/scrub/scrub_validator.cc

@@ -96,11 +96,9 @@ shard_evaluation_t evaluate_object_shard(
     if (xiter == obj.attrs.end()) {
       ret.shard_info.set_info_missing();
     } else {
-      bufferlist bl;
-      bl.push_back(xiter->second);
       ret.object_info = object_info_t{};
       try {
-	auto bliter = bl.cbegin();
+	auto bliter = xiter->second.cbegin();
 	::decode(*(ret.object_info), bliter);
       } catch (...) {
 	ret.shard_info.set_info_corrupted();
@@ -120,11 +118,9 @@ shard_evaluation_t evaluate_object_shard(
     if (xiter == obj.attrs.end()) {
       ret.shard_info.set_snapset_missing();
     } else {
-      bufferlist bl;
-      bl.push_back(xiter->second);
       ret.snapset = SnapSet{};
       try {
-	auto bliter = bl.cbegin();
+	auto bliter = xiter->second.cbegin();
 	::decode(*(ret.snapset), bliter);
       } catch (...) {
 	ret.shard_info.set_snapset_corrupted();
@@ -138,11 +134,9 @@ shard_evaluation_t evaluate_object_shard(
     if (xiter == obj.attrs.end()) {
       ret.shard_info.set_hinfo_missing();
     } else {
-      bufferlist bl;
-      bl.push_back(xiter->second);
       ret.hinfo = ECUtil::HashInfo{};
       try {
-	auto bliter = bl.cbegin();
+	auto bliter = xiter->second.cbegin();
 	decode(*(ret.hinfo), bliter);
       } catch (...) {
 	ret.shard_info.set_hinfo_corrupted();

src/mon/OSDMonitor.cc

@@ -8032,10 +8032,7 @@ int OSDMonitor::prepare_new_pool(string& name,
     /* crimson-osd requires that the pool be replicated and that pg_num/pgp_num
      * be static.  User must also have specified set-allow-crimson */
     const auto *suffix = " (--crimson specified or osd_pool_default_crimson set)";
-    if (pool_type != pg_pool_t::TYPE_REPLICATED) {
-      *ss << "crimson-osd only supports replicated pools" << suffix;
-      return -EINVAL;
-    } else if (pg_autoscale_mode != "off") {
+    if (pg_autoscale_mode != "off") {
       *ss << "crimson-osd does not support changing pg_num or pgp_num, "
 	  << "pg_autoscale_mode must be set to 'off'" << suffix;
       return -EINVAL;

src/osd/CMakeLists.txt

@@ -17,6 +17,7 @@ set(osd_srcs
   PGLog.cc
   PrimaryLogPG.cc
   ReplicatedBackend.cc
+  ECCommon.cc
   ECBackend.cc
   ECTransaction.cc
   PGBackend.cc