Merge branch 'merge-tokudb-5.6' into 10.0

9 files changed, 66 insertions, 50 deletions

diff --git a/storage/tokudb/PerconaFT/ft/ft-ops.cc b/storage/tokudb/PerconaFT/ft/ft-ops.cc
index 30a8710d7aa..ad9ecb1d074 100644
--- a/storage/tokudb/PerconaFT/ft/ft-ops.cc
+++ b/storage/tokudb/PerconaFT/ft/ft-ops.cc
@@ -651,10 +651,8 @@ void toku_ftnode_clone_callback(void *value_data,
     // set new pair attr if necessary
     if (node->height == 0) {
         *new_attr = make_ftnode_pair_attr(node);
-        for (int i = 0; i < node->n_children; i++) {
-            BLB(node, i)->logical_rows_delta = 0;
-            BLB(cloned_node, i)->logical_rows_delta = 0;
-        }
+        node->logical_rows_delta = 0;
+        cloned_node->logical_rows_delta = 0;
     } else {
         new_attr->is_valid = false;
     }
@@ -702,6 +700,10 @@ void toku_ftnode_flush_callback(CACHEFILE UU(cachefile),
             if (ftnode->height == 0) {
                 FT_STATUS_INC(FT_FULL_EVICTIONS_LEAF, 1);
                 FT_STATUS_INC(FT_FULL_EVICTIONS_LEAF_BYTES, node_size);
+                if (!ftnode->dirty) {
+                    toku_ft_adjust_logical_row_count(
+                        ft, -ftnode->logical_rows_delta);
+                }
             } else {
                 FT_STATUS_INC(FT_FULL_EVICTIONS_NONLEAF, 1);
                 FT_STATUS_INC(FT_FULL_EVICTIONS_NONLEAF_BYTES, node_size);
@@ -714,11 +716,12 @@ void toku_ftnode_flush_callback(CACHEFILE UU(cachefile),
                         BASEMENTNODE bn = BLB(ftnode, i);
                         toku_ft_decrease_stats(&ft->in_memory_stats,
                                                bn->stat64_delta);
-                        if (!ftnode->dirty)
-                            toku_ft_adjust_logical_row_count(
-                                ft, -bn->logical_rows_delta);
                     }
                 }
+                if (!ftnode->dirty) {
+                    toku_ft_adjust_logical_row_count(
+                        ft, -ftnode->logical_rows_delta);
+                }
             }
         }
         toku_ftnode_free(&ftnode);
@@ -944,8 +947,6 @@ int toku_ftnode_pe_callback(void *ftnode_pv,
                     basements_to_destroy[num_basements_to_destroy++] = bn;
                     toku_ft_decrease_stats(&ft->in_memory_stats,
                                            bn->stat64_delta);
-                    toku_ft_adjust_logical_row_count(ft,
-                                                     -bn->logical_rows_delta);
                     set_BNULL(node, i);
                     BP_STATE(node, i) = PT_ON_DISK;
                     num_partial_evictions++;
@@ -2652,7 +2653,7 @@ static std::unique_ptr<char[], decltype(&toku_free)> toku_file_get_parent_dir(
     return result;
 }
 
-static bool toku_create_subdirs_if_needed(const char *path) {
+bool toku_create_subdirs_if_needed(const char *path) {
     static const mode_t dir_mode = S_IRUSR | S_IWUSR | S_IXUSR | S_IRGRP |
                                    S_IWGRP | S_IXGRP | S_IROTH | S_IXOTH;
 
@@ -4563,6 +4564,8 @@ int toku_ft_rename_iname(DB_TXN *txn,
                          bs_new_name);
     }
 
+    if (!toku_create_subdirs_if_needed(new_iname_full.get()))
+        return get_error_errno();
     r = toku_os_rename(old_iname_full.get(), new_iname_full.get());
     if (r != 0)
         return r;
diff --git a/storage/tokudb/PerconaFT/ft/ft-ops.h b/storage/tokudb/PerconaFT/ft/ft-ops.h
index 70cf045d43c..df8ffe287df 100644
--- a/storage/tokudb/PerconaFT/ft/ft-ops.h
+++ b/storage/tokudb/PerconaFT/ft/ft-ops.h
@@ -288,3 +288,8 @@ void toku_ft_set_direct_io(bool direct_io_on);
 void toku_ft_set_compress_buffers_before_eviction(bool compress_buffers);
 
 void toku_note_deserialized_basement_node(bool fixed_key_size);
+
+// Creates all directories for the path if necessary,
+// returns true if all dirs are created successfully or
+// all dirs exist, false otherwise.
+bool toku_create_subdirs_if_needed(const char* path);
diff --git a/storage/tokudb/PerconaFT/ft/logger/recover.cc b/storage/tokudb/PerconaFT/ft/logger/recover.cc
index a9c30c0e37a..9eaa56bdc53 100644
--- a/storage/tokudb/PerconaFT/ft/logger/recover.cc
+++ b/storage/tokudb/PerconaFT/ft/logger/recover.cc
@@ -987,7 +987,8 @@ static int toku_recover_frename(struct logtype_frename *l, RECOVER_ENV renv) {
         return 1;
 
     if (old_exist && !new_exist &&
-        (toku_os_rename(old_iname_full.get(), new_iname_full.get()) == -1 ||
+        (!toku_create_subdirs_if_needed(new_iname_full.get()) ||
+         toku_os_rename(old_iname_full.get(), new_iname_full.get()) == -1 ||
          toku_fsync_directory(old_iname_full.get()) == -1 ||
          toku_fsync_directory(new_iname_full.get()) == -1))
         return 1;
diff --git a/storage/tokudb/PerconaFT/ft/node.cc b/storage/tokudb/PerconaFT/ft/node.cc
index 12e5fda226e..07309ff7f94 100644
--- a/storage/tokudb/PerconaFT/ft/node.cc
+++ b/storage/tokudb/PerconaFT/ft/node.cc
@@ -386,7 +386,8 @@ static void bnc_apply_messages_to_basement_node(
     const pivot_bounds &
         bounds,  // contains pivot key bounds of this basement node
     txn_gc_info *gc_info,
-    bool *msgs_applied) {
+    bool *msgs_applied,
+    int64_t* logical_rows_delta) {
     int r;
     NONLEAF_CHILDINFO bnc = BNC(ancestor, childnum);
 
@@ -394,7 +395,6 @@ static void bnc_apply_messages_to_basement_node(
     // apply messages from this buffer
     STAT64INFO_S stats_delta = {0, 0};
     uint64_t workdone_this_ancestor = 0;
-    int64_t logical_rows_delta = 0;
 
     uint32_t stale_lbi, stale_ube;
     if (!bn->stale_ancestor_messages_applied) {
@@ -470,7 +470,7 @@ static void bnc_apply_messages_to_basement_node(
                             gc_info,
                             &workdone_this_ancestor,
                             &stats_delta,
-                            &logical_rows_delta);
+                            logical_rows_delta);
         }
     } else if (stale_lbi == stale_ube) {
         // No stale messages to apply, we just apply fresh messages, and mark
@@ -482,7 +482,7 @@ static void bnc_apply_messages_to_basement_node(
             .gc_info = gc_info,
             .workdone = &workdone_this_ancestor,
             .stats_to_update = &stats_delta,
-            .logical_rows_delta = &logical_rows_delta};
+            .logical_rows_delta = logical_rows_delta};
         if (fresh_ube - fresh_lbi > 0)
             *msgs_applied = true;
         r = bnc->fresh_message_tree
@@ -503,7 +503,7 @@ static void bnc_apply_messages_to_basement_node(
             .gc_info = gc_info,
             .workdone = &workdone_this_ancestor,
             .stats_to_update = &stats_delta,
-            .logical_rows_delta = &logical_rows_delta};
+            .logical_rows_delta = logical_rows_delta};
 
         r = bnc->stale_message_tree
                 .iterate_on_range<struct iterate_do_bn_apply_msg_extra,
@@ -521,8 +521,6 @@ static void bnc_apply_messages_to_basement_node(
     if (stats_delta.numbytes || stats_delta.numrows) {
         toku_ft_update_stats(&t->ft->in_memory_stats, stats_delta);
     }
-    toku_ft_adjust_logical_row_count(t->ft, logical_rows_delta);
-    bn->logical_rows_delta += logical_rows_delta;
 }
 
 static void
@@ -536,6 +534,7 @@ apply_ancestors_messages_to_bn(
     bool* msgs_applied
     )
 {
+    int64_t logical_rows_delta = 0;
     BASEMENTNODE curr_bn = BLB(node, childnum);
     const pivot_bounds curr_bounds = bounds.next_bounds(node, childnum);
     for (ANCESTORS curr_ancestors = ancestors; curr_ancestors; curr_ancestors = curr_ancestors->next) {
@@ -548,13 +547,16 @@ apply_ancestors_messages_to_bn(
                 curr_ancestors->childnum,
                 curr_bounds,
                 gc_info,
-                msgs_applied
+                msgs_applied,
+                &logical_rows_delta
                 );
             // We don't want to check this ancestor node again if the
             // next time we query it, the msn hasn't changed.
             curr_bn->max_msn_applied = curr_ancestors->node->max_msn_applied_to_node_on_disk;
         }
     }
+    toku_ft_adjust_logical_row_count(t->ft, logical_rows_delta);
+    node->logical_rows_delta += logical_rows_delta;
     // At this point, we know all the stale messages above this
     // basement node have been applied, and any new messages will be
     // fresh, so we don't need to look at stale messages for this
diff --git a/storage/tokudb/PerconaFT/ft/node.h b/storage/tokudb/PerconaFT/ft/node.h
index 52eefec0936..db189e36d59 100644
--- a/storage/tokudb/PerconaFT/ft/node.h
+++ b/storage/tokudb/PerconaFT/ft/node.h
@@ -157,36 +157,49 @@ private:
 
 // TODO: class me up
 struct ftnode {
-    MSN      max_msn_applied_to_node_on_disk; // max_msn_applied that will be written to disk
+    // max_msn_applied that will be written to disk
+    MSN max_msn_applied_to_node_on_disk;
     unsigned int flags;
-    BLOCKNUM blocknum;   // Which block number is this node?
-    int    layout_version; // What version of the data structure?
-    int    layout_version_original;	// different (<) from layout_version if upgraded from a previous version (useful for debugging)
-    int    layout_version_read_from_disk;  // transient, not serialized to disk, (useful for debugging)
-    uint32_t build_id;       // build_id (svn rev number) of software that wrote this node to disk
-    int    height; /* height is always >= 0.  0 for leaf, >0 for nonleaf. */
-    int    dirty;
+    // Which block number is this node?
+    BLOCKNUM blocknum;
+    // What version of the data structure?
+    int layout_version;
+    // different (<) from layout_version if upgraded from a previous version
+    // (useful for debugging)
+    int layout_version_original;
+    // transient, not serialized to disk, (useful for debugging)
+    int layout_version_read_from_disk;
+    // build_id (svn rev number) of software that wrote this node to disk
+    uint32_t build_id;
+    // height is always >= 0.  0 for leaf, >0 for nonleaf.
+    int height;
+    int dirty;
     uint32_t fullhash;
+    // current count of rows add or removed as a result of message application
+    // to this node as a basement, irrelevant for internal nodes, gets reset
+    // when node is undirtied. Used to back out tree scoped LRC id node is
+    // evicted but not persisted
+    int64_t logical_rows_delta;
 
-    // for internal nodes, if n_children==fanout+1 then the tree needs to be rebalanced.
-    // for leaf nodes, represents number of basement nodes
+    // for internal nodes, if n_children==fanout+1 then the tree needs to be
+    // rebalanced. for leaf nodes, represents number of basement nodes
     int n_children;
     ftnode_pivot_keys pivotkeys;
 
-    // What's the oldest referenced xid that this node knows about? The real oldest
-    // referenced xid might be younger, but this is our best estimate. We use it
-    // as a heuristic to transition provisional mvcc entries from provisional to
-    // committed (from implicity committed to really committed).
+    // What's the oldest referenced xid that this node knows about? The real
+    // oldest referenced xid might be younger, but this is our best estimate.
+    // We use it as a heuristic to transition provisional mvcc entries from
+    // provisional to committed (from implicity committed to really committed).
     //
-    // A better heuristic would be the oldest live txnid, but we use this since it
-    // still works well most of the time, and its readily available on the inject
-    // code path.
+    // A better heuristic would be the oldest live txnid, but we use this since
+    // it still works well most of the time, and its readily available on the
+    // inject code path.
     TXNID oldest_referenced_xid_known;
 
     // array of size n_children, consisting of ftnode partitions
-    // each one is associated with a child
-    // for internal nodes, the ith partition corresponds to the ith message buffer
-    // for leaf nodes, the ith partition corresponds to the ith basement node
+    // each one is associated with a child  for internal nodes, the ith
+    // partition corresponds to the ith message buffer for leaf nodes, the ith
+    // partition corresponds to the ith basement node
     struct ftnode_partition *bp;
     struct ctpair *ct_pair;
 };
@@ -199,7 +212,6 @@ struct ftnode_leaf_basement_node {
     MSN max_msn_applied;            // max message sequence number applied
     bool stale_ancestor_messages_applied;
     STAT64INFO_S stat64_delta;      // change in stat64 counters since basement was last written to disk
-    int64_t logical_rows_delta;
 };
 typedef struct ftnode_leaf_basement_node *BASEMENTNODE;
 
diff --git a/storage/tokudb/PerconaFT/ft/serialize/ft_node-serialize.cc b/storage/tokudb/PerconaFT/ft/serialize/ft_node-serialize.cc
index 5914f8a1050..56876b474d4 100644
--- a/storage/tokudb/PerconaFT/ft/serialize/ft_node-serialize.cc
+++ b/storage/tokudb/PerconaFT/ft/serialize/ft_node-serialize.cc
@@ -996,7 +996,6 @@ BASEMENTNODE toku_clone_bn(BASEMENTNODE orig_bn) {
     bn->seqinsert = orig_bn->seqinsert;
     bn->stale_ancestor_messages_applied = orig_bn->stale_ancestor_messages_applied;
     bn->stat64_delta = orig_bn->stat64_delta;
-    bn->logical_rows_delta = orig_bn->logical_rows_delta;
     bn->data_buffer.clone(&orig_bn->data_buffer);
     return bn;
 }
@@ -1007,7 +1006,6 @@ BASEMENTNODE toku_create_empty_bn_no_buffer(void) {
     bn->seqinsert = 0;
     bn->stale_ancestor_messages_applied = false;
     bn->stat64_delta = ZEROSTATS;
-    bn->logical_rows_delta = 0;
     bn->data_buffer.init_zero();
     return bn;
 }
@@ -1432,6 +1430,7 @@ static FTNODE alloc_ftnode_for_deserialize(uint32_t fullhash, BLOCKNUM blocknum)
     node->fullhash = fullhash;
     node->blocknum = blocknum;
     node->dirty = 0;
+    node->logical_rows_delta = 0;
     node->bp = nullptr;
     node->oldest_referenced_xid_known = TXNID_NONE;
     return node; 
diff --git a/storage/tokudb/PerconaFT/ft/txn/roll.cc b/storage/tokudb/PerconaFT/ft/txn/roll.cc
index 9f3977743a0..4f374d62173 100644
--- a/storage/tokudb/PerconaFT/ft/txn/roll.cc
+++ b/storage/tokudb/PerconaFT/ft/txn/roll.cc
@@ -227,7 +227,8 @@ int toku_rollback_frename(BYTESTRING old_iname,
         return 1;
 
     if (!old_exist && new_exist &&
-        (toku_os_rename(new_iname_full.get(), old_iname_full.get()) == -1 ||
+        (!toku_create_subdirs_if_needed(old_iname_full.get()) ||
+         toku_os_rename(new_iname_full.get(), old_iname_full.get()) == -1 ||
          toku_fsync_directory(new_iname_full.get()) == -1 ||
          toku_fsync_directory(old_iname_full.get()) == -1))
         return 1;
diff --git a/storage/tokudb/PerconaFT/util/dmt.h b/storage/tokudb/PerconaFT/util/dmt.h
index 71cde8814ab..99be296d0e9 100644
--- a/storage/tokudb/PerconaFT/util/dmt.h
+++ b/storage/tokudb/PerconaFT/util/dmt.h
@@ -589,7 +589,6 @@ private:
 
     void convert_from_tree_to_array(void);
 
-    __attribute__((nonnull(2,5)))
     void delete_internal(subtree *const subtreep, const uint32_t idx, subtree *const subtree_replace, subtree **const rebalance_subtree);
 
     template<typename iterate_extra_t,
@@ -627,16 +626,12 @@ private:
     __attribute__((nonnull))
     void rebalance(subtree *const subtree);
 
-    __attribute__((nonnull(3)))
     static void copyout(uint32_t *const outlen, dmtdata_t *const out, const dmt_node *const n);
 
-    __attribute__((nonnull(3)))
     static void copyout(uint32_t *const outlen, dmtdata_t **const out, dmt_node *const n);
 
-    __attribute__((nonnull(4)))
     static void copyout(uint32_t *const outlen, dmtdata_t *const out, const uint32_t len, const dmtdata_t *const stored_value_ptr);
 
-    __attribute__((nonnull(4)))
     static void copyout(uint32_t *const outlen, dmtdata_t **const out, const uint32_t len, dmtdata_t *const stored_value_ptr);
 
     template<typename dmtcmp_t,
diff --git a/storage/tokudb/PerconaFT/util/omt.h b/storage/tokudb/PerconaFT/util/omt.h
index 799ed0eae7c..c7ed2ca546f 100644
--- a/storage/tokudb/PerconaFT/util/omt.h
+++ b/storage/tokudb/PerconaFT/util/omt.h
@@ -284,7 +284,6 @@ public:
      *               By taking ownership of the array, we save a malloc and memcpy,
      *               and possibly a free (if the caller is done with the array).
      */
-    __attribute__((nonnull))
     void create_steal_sorted_array(omtdata_t **const values, const uint32_t numvalues, const uint32_t new_capacity);
 
     /**
@@ -667,7 +666,6 @@ private:
 
     void set_at_internal(const subtree &subtree, const omtdata_t &value, const uint32_t idx);
 
-    __attribute__((nonnull(2,5)))
     void delete_internal(subtree *const subtreep, const uint32_t idx, omt_node *const copyn, subtree **const rebalance_subtree);
 
     template<typename iterate_extra_t,