First Steps Towards Replacing sqlite_orm with sqlite_modern_cpp

.gitmodules

@@ -75,3 +75,6 @@
 [submodule "src/rgw/driver/sfs/sqlite/sqlite_orm"]
 	path = src/rgw/driver/sfs/sqlite/sqlite_orm
 	url = https://github.com/aquarist-labs/sqlite_orm.git
+[submodule "src/rgw/driver/sfs/sqlite_modern_cpp"]
+	path = src/rgw/driver/sfs/sqlite/sqlite_modern_cpp
+	url = https://github.com/SqliteModernCpp/sqlite_modern_cpp.git

src/rgw/CMakeLists.txt

@@ -320,6 +320,7 @@ endif()
 
 if(WITH_RADOSGW_SFS)
   target_include_directories(rgw_common PUBLIC "${CMAKE_SOURCE_DIR}/src/rgw/store/sfs")
+  target_include_directories(rgw_common PUBLIC "${CMAKE_SOURCE_DIR}/src/rgw/store/sfs/sqlite/sqlite_modern_cpp/hdr")
   target_link_libraries(rgw_common PRIVATE global sfs)
 endif()
 

src/rgw/driver/sfs/CMakeLists.txt

@@ -25,6 +25,7 @@ set(sfs_srcs
   sqlite/dbconn.cc
   sqlite/errors.cc
   sqlite/sqlite_list.cc
+  sqlite/conversion_utils.cc
   bucket.cc
   multipart.cc
   object.cc

src/rgw/driver/sfs/object_state.h

@@ -15,9 +15,12 @@
 #define RGW_SFS_OBJECT_STATE_H
 
 #include <iostream>
+
+#include "include/ceph_assert.h"
 #if FMT_VERSION >= 90000
 #include <fmt/ostream.h>
 #endif
+#include "sqlite/dbapi.h"
 
 namespace rgw::sal::sfs {
 
@@ -47,6 +50,36 @@ inline std::string str_object_state(ObjectState state) {
   return result;
 }
 
+template <>
+struct dbapi::sqlite::has_sqlite_type<ObjectState, SQLITE_INTEGER, void>
+    : ::std::true_type {};
+
+inline int bind_col_in_db(
+    sqlite3_stmt* stmt, int inx, const rgw::sal::sfs::ObjectState& val
+) {
+  return sqlite3_bind_int(stmt, inx, static_cast<int>(val));
+}
+inline void store_result_in_db(
+    sqlite3_context* db, const rgw::sal::sfs::ObjectState& val
+) {
+  sqlite3_result_int(db, static_cast<int>(val));
+}
+inline rgw::sal::sfs::ObjectState
+get_col_from_db(sqlite3_stmt* stmt, int inx, dbapi::sqlite::result_type<rgw::sal::sfs::ObjectState>) {
+  if (sqlite3_column_type(stmt, inx) == SQLITE_NULL) {
+    ceph_abort_msg("cannot make enum value from NULL");
+  }
+  return static_cast<rgw::sal::sfs::ObjectState>(sqlite3_column_int(stmt, inx));
+}
+
+inline rgw::sal::sfs::ObjectState
+get_val_from_db(sqlite3_value* value, dbapi::sqlite::result_type<rgw::sal::sfs::ObjectState>) {
+  if (sqlite3_value_type(value) == SQLITE_NULL) {
+    ceph_abort_msg("cannot make enum value from NULL");
+  }
+  return static_cast<rgw::sal::sfs::ObjectState>(sqlite3_value_int(value));
+}
+
 }  // namespace rgw::sal::sfs
 
 inline std::ostream& operator<<(

src/rgw/driver/sfs/sqlite/bindings/real_time.h

@@ -14,6 +14,8 @@
 #pragma once
 
 #include "common/ceph_time.h"
+#include "include/ceph_assert.h"
+#include "rgw/driver/sfs/sqlite/dbapi.h"
 #include "rgw/driver/sfs/sqlite/sqlite_orm.h"
 
 /// ceph::real_time is represented as a uint64 (unsigned).
@@ -90,3 +92,41 @@ struct row_extractor<ceph::real_time> {
 };
 
 }  // namespace sqlite_orm
+
+namespace rgw::sal::sfs::dbapi::sqlite {
+
+template <>
+struct has_sqlite_type<ceph::real_time, SQLITE_INTEGER, void>
+    : ::std::true_type {};
+
+inline int bind_col_in_db(
+    sqlite3_stmt* stmt, int inx, const ceph::real_time& val
+) {
+  return sqlite3_bind_int64(
+      stmt, inx, rgw::sal::sfs::sqlite::time_point_to_int64(val)
+  );
+}
+inline void store_result_in_db(
+    sqlite3_context* db, const ceph::real_time& val
+) {
+  sqlite3_result_int64(db, rgw::sal::sfs::sqlite::time_point_to_int64(val));
+}
+inline ceph::real_time
+get_col_from_db(sqlite3_stmt* stmt, int inx, result_type<ceph::real_time>) {
+  if (sqlite3_column_type(stmt, inx) == SQLITE_NULL) {
+    ceph_abort_msg("cannot make enum value from NULL");
+  }
+  return rgw::sal::sfs::sqlite::time_point_from_int64(
+      sqlite3_column_int64(stmt, inx)
+  );
+}
+
+inline ceph::real_time
+get_val_from_db(sqlite3_value* value, result_type<ceph::real_time>) {
+  if (sqlite3_value_type(value) == SQLITE_NULL) {
+    ceph_abort_msg("cannot make enum value from NULL");
+  }
+  return rgw::sal::sfs::sqlite::time_point_from_int64(sqlite3_value_int64(value)
+  );
+}
+}  // namespace rgw::sal::sfs::dbapi::sqlite

src/rgw/driver/sfs/sqlite/conversion_utils.cc

@@ -0,0 +1,37 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t
+// vim: ts=8 sw=2 smarttab ft=cpp
+/*
+ * Ceph - scalable distributed file system
+ * SFS SAL implementation
+ *
+ * Copyright (C) 2023 SUSE LLC
+ *
+ * This is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License version 2.1, as published by the Free Software
+ * Foundation. See file COPYING.
+ */
+
+#include <string>
+
+namespace rgw::sal::sfs::sqlite {
+
+std::string prefix_to_escaped_like(const std::string& prefix, char escape) {
+  std::string like_expr;
+  like_expr.reserve(prefix.length() + 10);
+  for (const char c : prefix) {
+    switch (c) {
+      case '%':
+        [[fallthrough]];
+      case '_':
+        like_expr.push_back(escape);
+        [[fallthrough]];
+      default:
+        like_expr.push_back(c);
+    }
+  }
+  like_expr.push_back('%');
+  return like_expr;
+}
+
+}  // namespace rgw::sal::sfs::sqlite

src/rgw/driver/sfs/sqlite/conversion_utils.h

@@ -128,22 +128,12 @@ void assign_db_value(const SOURCE& source, std::vector<char>& dest) {
   dest = blob_vector;
 }
 
+std::string prefix_to_escaped_like(const std::string& prefix, char escape);
+
 template <typename COL>
 sqlite_orm::internal::like_t<COL, std::basic_string<char>, const char*>
 prefix_to_like(COL col, const std::string& prefix) {
-  std::string like_expr;
-  like_expr.reserve(prefix.length() + 10);
-  for (const char c : prefix) {
-    switch (c) {
-      case '%':
-      case '_':
-        like_expr.push_back('\a');
-      default:
-        like_expr.push_back(c);
-    }
-  }
-  like_expr.push_back('%');
-  return sqlite_orm::like(col, like_expr, "\a");
+  return sqlite_orm::like(col, prefix_to_escaped_like(prefix, '\a'), "\a");
 }
 
 }  // namespace rgw::sal::sfs::sqlite

src/rgw/driver/sfs/sqlite/dbapi.h

@@ -0,0 +1,9 @@
+#pragma once
+
+#include <sqlite3.h>
+
+namespace rgw::sal::sfs::dbapi {
+
+#include "sqlite_modern_cpp/hdr/sqlite_modern_cpp.h"
+
+}  // namespace rgw::sal::sfs::dbapi

src/rgw/driver/sfs/sqlite/dbconn.h

@@ -26,6 +26,7 @@
 #include "buckets/multipart_definitions.h"
 #include "common/ceph_mutex.h"
 #include "common/dout.h"
+#include "dbapi.h"
 #include "lifecycle/lifecycle_definitions.h"
 #include "objects/object_definitions.h"
 #include "rgw/rgw_perf_counters.h"
@@ -255,6 +256,8 @@ inline auto _make_storage(const std::string& path) {
 using Storage = decltype(_make_storage(""));
 using StorageRef = Storage*;
 
+// TODO(https://github.com/aquarist-labs/s3gw/issues/788): Make
+// dbapi::sqlite::database the primary interface for sqlite3.
 class DBConn {
  private:
   std::unordered_map<std::thread::id, Storage> storage_pool;
@@ -282,6 +285,10 @@ class DBConn {
     return sqlite_conns;
   }
 
+  dbapi::sqlite::database get() {
+    return dbapi::sqlite::database(get_storage()->filename());
+  }
+
   static std::string getDBPath(CephContext* cct) {
     auto rgw_sfs_path = cct->_conf.get_val<std::string>("rgw_sfs_data_path");
     auto db_path =

src/rgw/driver/sfs/sqlite/sqlite_list.cc

@@ -13,6 +13,7 @@
 
 #include <limits>
 
+#include "dbapi.h"
 #include "rgw/driver/sfs/sqlite/conversion_utils.h"
 #include "rgw/driver/sfs/sqlite/objects/object_definitions.h"
 #include "rgw/driver/sfs/sqlite/versioned_object/versioned_object_definitions.h"
@@ -101,64 +102,45 @@ bool SQLiteList::versions(
   // more available logic: request one more than max. if we get that
   // much set out_more_available, but return only up to max
   ceph_assert(max < std::numeric_limits<size_t>::max());
-  const size_t query_limit = max + 1;
-
-  auto storage = conn->get_storage();
-  auto rows = storage->select(
-      columns(
-          &DBObject::name, &DBVersionedObject::version_id,
-          &DBVersionedObject::mtime, &DBVersionedObject::etag,
-          &DBVersionedObject::size, &DBVersionedObject::version_type,
-          is_equal(
-              // IsLatest logic
-              // - Use the id as secondary condition if multiple version
-              // with same max(commit_time) exists
-              sqlite_orm::select(
-                  &DBVersionedObject::id, from<DBVersionedObject>(),
-                  where(
-                      is_equal(
-                          &DBObject::uuid, &DBVersionedObject::object_id
-                      ) and
-                      is_equal(
-                          &DBVersionedObject::object_state,
-                          ObjectState::COMMITTED
-                      )
-                  ),
-                  multi_order_by(
-                      order_by(&DBVersionedObject::commit_time).desc(),
-                      order_by(&DBVersionedObject::id).desc()
-                  ),
-                  limit(1)
-              ),
-              &DBVersionedObject::id
-          )
-      ),
-      inner_join<DBVersionedObject>(
-          on(is_equal(&DBObject::uuid, &DBVersionedObject::object_id))
-      ),
-      where(
-          is_equal(&DBVersionedObject::object_state, ObjectState::COMMITTED) and
-          is_equal(&DBObject::bucket_id, bucket_id) and
-          greater_than(&DBObject::name, start_after_object_name) and
-          prefix_to_like(&DBObject::name, prefix)
-      ),
-      // Sort:
-      // names a-Z
-      // first delete markers, then versions - (See: LC CurrentExpiration)
-      // newest to oldest version
-      multi_order_by(
-          order_by(&DBObject::name).asc(),
-          order_by(&DBVersionedObject::commit_time).desc(),
-          order_by(&DBVersionedObject::id).desc()
-      ),
-      limit(query_limit)
-  );
-
-  ceph_assert(rows.size() <= static_cast<size_t>(query_limit));
-  const size_t return_limit = std::min(max, rows.size());
-  out.reserve(return_limit);
-  for (size_t i = 0; i < return_limit; i++) {
-    const auto& row = rows[i];
+  const uint32_t query_limit = max + 1;
+  dbapi::sqlite::database db = conn->get();
+  auto rows = db << R"sql(
+      SELECT
+         o.name, vo.version_id, vo.mtime, vo.etag, vo.size, vo.version_type,
+         (vo.id = ( SELECT id FROM versioned_objects
+           WHERE object_id = o.uuid
+           AND object_state = ?
+           ORDER BY commit_time desc, id desc
+           LIMIT 1
+         )) AS is_latest
+      FROM objects as o
+      INNER JOIN versioned_objects as vo
+      ON (o.uuid = vo.object_id)
+      WHERE vo.object_state = ?
+      AND o.bucket_id = ?
+      AND o.name > ?
+      AND o.name LIKE ? ESCAPE CHAR(7)
+      ORDER BY o.name ASC,
+        vo.commit_time DESC,
+        vo.id DESC
+      LIMIT ?;)sql"
+                 << ObjectState::COMMITTED << ObjectState::COMMITTED
+                 << bucket_id << start_after_object_name
+                 << prefix_to_escaped_like(prefix, '\a') << query_limit;
+  out.reserve(max);
+  if (out_more_available) {
+    *out_more_available = false;
+  }
+  for (std::tuple<
+           std::string, std::string, ceph::real_time, std::string, int64_t,
+           VersionType, bool>
+           row : rows) {
+    if (out.size() >= max) {
+      if (out_more_available) {
+        *out_more_available = true;
+      }
+      break;
+    }
     rgw_bucket_dir_entry e;
     e.key.name = std::get<0>(row);
     e.key.instance = std::get<1>(row);
@@ -169,9 +151,7 @@ bool SQLiteList::versions(
     e.flags = to_dentry_flag(std::get<5>(row), std::get<6>(row));
     out.emplace_back(e);
   }
-  if (out_more_available) {
-    *out_more_available = rows.size() == query_limit;
-  }
+
   return true;
 }
 

src/rgw/driver/sfs/version_type.h

@@ -14,6 +14,8 @@
 #ifndef RGW_SFS_VERSION_TYPE_H
 #define RGW_SFS_VERSION_TYPE_H
 
+#include "sqlite/dbapi.h"
+
 namespace rgw::sal::sfs {
 
 enum class VersionType {
@@ -22,6 +24,36 @@ enum class VersionType {
   LAST_VALUE = DELETE_MARKER
 };
 
+template <>
+struct dbapi::sqlite::has_sqlite_type<VersionType, SQLITE_INTEGER, void>
+    : ::std::true_type {};
+
+inline int bind_col_in_db(
+    sqlite3_stmt* stmt, int inx, const rgw::sal::sfs::VersionType& val
+) {
+  return sqlite3_bind_int(stmt, inx, static_cast<int>(val));
+}
+inline void store_result_in_db(
+    sqlite3_context* db, const rgw::sal::sfs::VersionType& val
+) {
+  sqlite3_result_int(db, static_cast<int>(val));
+}
+inline rgw::sal::sfs::VersionType
+get_col_from_db(sqlite3_stmt* stmt, int inx, dbapi::sqlite::result_type<rgw::sal::sfs::VersionType>) {
+  if (sqlite3_column_type(stmt, inx) == SQLITE_NULL) {
+    ceph_abort_msg("cannot make enum value from NULL");
+  }
+  return static_cast<rgw::sal::sfs::VersionType>(sqlite3_column_int(stmt, inx));
+}
+
+inline rgw::sal::sfs::VersionType
+get_val_from_db(sqlite3_value* value, dbapi::sqlite::result_type<rgw::sal::sfs::VersionType>) {
+  if (sqlite3_value_type(value) == SQLITE_NULL) {
+    ceph_abort_msg("cannot make enum value from NULL");
+  }
+  return static_cast<rgw::sal::sfs::VersionType>(sqlite3_value_int(value));
+}
+
 }  // namespace rgw::sal::sfs
 
 #endif  // RGW_SFS_VERSION_TYPE_H