Changes to support PrefixNode and consolidation of tree

Following are the major changes in btree library:

1. A new BtreeKey and BtreeValue type called IntervalKey and IntervalValue which allows a particular key to be part of an larger interval (say 1-3 will result in key1, key2, key3 and each key is an BtreeIntervalKey).

2. Implemented a new tree node called PrefixNode which can store interval key and value effectively stores them with prefix and suffix to store the intervals compactly.

3. Removed the Extent code which was not used and replaced it with the IntervalKey/Value as mentioned above.

4. Added filtering for read/write/remove operations, so that consumer can change the way which key/value to be custom modify. This change effectively provides compare_and_set or compare_and_remove operations atomically. This change also removed the previous on_read_cb etc, which was passed on every callback

5. Removed the requirement for BtreeKey and Value to implement some static methods like get_key_size() etc, by creating a dummyKey, dummyValue. Now the requirement of the implementor of btree key and value only needs to ensure it is default constructible (which is already the case today)

6. Created new internal btree node APIs such as multi_put, multi_get instead of each btree operations repeat similar code to read or remove operations in leaf node. As a result, a method called match_range() is introduced in the btree node, which consolidates and searches for all range operation requirements. Moved most of the common operations into variant_node class and now variant_node is derived from BtreeNode. PrefixNode overrides variant_node and implements its own version of multi_put etc. range operations.

7. Removed obsolete btree return status.

8. Consolidated the mem_btree test, mem_btree concurrent test and index_btree test into one btree_test_helper library. Modified the way range_scheduler picks the existing and working keys from boost icl to sis::Bitset as it simplifies them and also provides some essential feature which is always pick some keys to schedule instead of failing.

CMakeLists.txt

@@ -98,7 +98,7 @@ endif()
 
 if(${CMAKE_BUILD_TYPE} STREQUAL "Debug")
     message(STATUS "Debug build")
-    add_flags("-DDEBUG_RCU")
+    add_flags("-DDEBUG_RCU -D_DEBUG")
 else()
     message(STATUS "Release build")
     if((${CMAKE_CXX_COMPILER_ID} STREQUAL "GNU") OR (${CMAKE_CXX_COMPILER_ID} STREQUAL "Clang")) 

conanfile.py

@@ -5,7 +5,7 @@
 
 class HomestoreConan(ConanFile):
     name = "homestore"
-    version = "4.5.5"
+    version = "4.6.0"
 
     homepage = "https://github.com/eBay/Homestore"
     description = "HomeStore Storage Engine"

src/include/homestore/btree/btree.hpp

@@ -31,10 +31,6 @@ SISL_LOGGING_DECL(btree)
 
 namespace homestore {
 
-typedef std::function< bool(const BtreeKey&, const BtreeValue&, const BtreeRequest&) > on_kv_read_t;
-typedef std::function< bool(const BtreeKey&, const BtreeValue&, const BtreeRequest&) > on_kv_remove_t;
-typedef std::function< bool(const BtreeKey&, const BtreeKey&, const BtreeValue&, const BtreeRequest&) > on_kv_update_t;
-
 using BtreeNodePtr = boost::intrusive_ptr< BtreeNode >;
 
 struct BtreeThreadVariables {
@@ -57,11 +53,6 @@ class Btree {
     std::atomic< uint64_t > m_req_id{0};
 #endif
 
-    // Optional callback on various read or kv operations
-    on_kv_read_t m_on_read_cb{nullptr};
-    on_kv_update_t m_on_update_cb{nullptr};
-    on_kv_remove_t m_on_remove_cb{nullptr};
-
     // This workaround of BtreeThreadVariables is needed instead of directly declaring statics
     // to overcome the gcc bug, pointer here: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66944
     static BtreeThreadVariables* bt_thread_vars() {
@@ -77,8 +68,7 @@ class Btree {
 
 public:
     /////////////////////////////////////// All External APIs /////////////////////////////
-    Btree(const BtreeConfig& cfg, on_kv_read_t&& read_cb = nullptr, on_kv_update_t&& update_cb = nullptr,
-          on_kv_remove_t&& remove_cb = nullptr);
+    Btree(const BtreeConfig& cfg);
     virtual ~Btree();
     virtual btree_status_t init(void* op_context);
 
@@ -174,10 +164,6 @@ class Btree {
     void validate_sanity_child(const BtreeNodePtr& parent_node, uint32_t ind) const;
     void validate_sanity_next_child(const BtreeNodePtr& parent_node, uint32_t ind) const;
     void print_node(const bnodeid_t& bnodeid) const;
-    bool call_on_read_kv_cb(const BtreeNodePtr& node, uint32_t idx, const BtreeRequest& req) const;
-    bool call_on_remove_kv_cb(const BtreeNodePtr& node, uint32_t idx, const BtreeRequest& req) const;
-    bool call_on_update_kv_cb(const BtreeNodePtr& node, uint32_t idx, const BtreeKey& new_key,
-                              const BtreeRequest& req) const;
     void append_route_trace(BtreeRequest& req, const BtreeNodePtr& node, btree_event_t event, uint32_t start_idx = 0,
                             uint32_t end_idx = 0) const;
 
@@ -194,10 +180,10 @@ class Btree {
     btree_status_t check_split_root(ReqT& req);
 
     template < typename ReqT >
-    bool is_split_needed(const BtreeNodePtr& node, const BtreeConfig& cfg, ReqT& req) const;
+    bool is_split_needed(const BtreeNodePtr& node, ReqT& req) const;
 
     btree_status_t split_node(const BtreeNodePtr& parent_node, const BtreeNodePtr& child_node, uint32_t parent_ind,
-                              BtreeKey* out_split_key, void* context);
+                              K* out_split_key, void* context);
     btree_status_t mutate_extents_in_leaf(const BtreeNodePtr& my_node, BtreeRangePutRequest< K >& rpreq);
     btree_status_t repair_split(const BtreeNodePtr& parent_node, const BtreeNodePtr& child_node1,
                                 uint32_t parent_split_idx, void* context);

src/include/homestore/btree/btree.ipp

@@ -36,14 +36,8 @@
 
 namespace homestore {
 template < typename K, typename V >
-Btree< K, V >::Btree(const BtreeConfig& cfg, on_kv_read_t&& read_cb, on_kv_update_t&& update_cb,
-                     on_kv_remove_t&& remove_cb) :
-        m_metrics{cfg.name().c_str()},
-        m_node_size{cfg.node_size()},
-        m_on_read_cb{std::move(read_cb)},
-        m_on_update_cb{std::move(update_cb)},
-        m_on_remove_cb{std::move(remove_cb)},
-        m_bt_cfg{cfg} {
+Btree< K, V >::Btree(const BtreeConfig& cfg) :
+        m_metrics{cfg.name().c_str()}, m_node_size{cfg.node_size()}, m_bt_cfg{cfg} {
     m_bt_cfg.set_node_data_size(cfg.node_size() - sizeof(persistent_hdr_t));
 }
 
@@ -105,7 +99,7 @@ retry:
     if (ret != btree_status_t::success) { goto out; }
     is_leaf = root->is_leaf();
 
-    if (is_split_needed(root, m_bt_cfg, put_req)) {
+    if (is_split_needed(root, put_req)) {
         // Time to do the split of root.
         unlock_node(root, acq_lock);
         m_btree_lock.unlock_shared();
@@ -143,8 +137,7 @@ out:
 #ifndef NDEBUG
     check_lock_debug();
 #endif
-    if (ret != btree_status_t::success && ret != btree_status_t::fast_path_not_possible &&
-        ret != btree_status_t::cp_mismatch) {
+    if (ret != btree_status_t::success && ret != btree_status_t::cp_mismatch) {
         BT_LOG(ERROR, "btree put failed {}", ret);
         COUNTER_INCREMENT(m_metrics, write_err_cnt, 1);
     }
@@ -267,9 +260,9 @@ btree_status_t Btree< K, V >::query(BtreeQueryRequest< K >& qreq, std::vector< s
     if ((qreq.query_type() == BtreeQueryType::SWEEP_NON_INTRUSIVE_PAGINATION_QUERY ||
          qreq.query_type() == BtreeQueryType::TREE_TRAVERSAL_QUERY)) {
         if (out_values.size()) {
-            K& out_last_key = out_values.back().first;
-            qreq.set_cursor_key(out_last_key);
+            K out_last_key = out_values.back().first;
             if (out_last_key.compare(qreq.input_range().end_key()) >= 0) { ret = btree_status_t::success; }
+            qreq.shift_working_range(std::move(out_last_key), false /* non inclusive*/);
         } else {
             DEBUG_ASSERT_NE(ret, btree_status_t::has_more, "Query returned has_more, but no values added")
         }
@@ -280,8 +273,7 @@ out:
 #ifndef NDEBUG
     check_lock_debug();
 #endif
-    if (ret != btree_status_t::success && ret != btree_status_t::has_more &&
-        ret != btree_status_t::fast_path_not_possible) {
+    if ((ret != btree_status_t::success) && (ret != btree_status_t::has_more)) {
         BT_LOG(ERROR, "btree query failed {}", ret);
         COUNTER_INCREMENT(m_metrics, query_err_cnt, 1);
     }