refs #5770 Only check one basement node on pin, remove the assumption that adjacent

available nodes are query-able.

ft/ft-cachetable-wrappers.cc

@@ -193,6 +193,11 @@ toku_create_new_ftnode (
         NULL);
 }
 
+//
+// On success, this function assumes that the caller is trying to pin the node
+// with a PL_READ lock. If message application is needed,
+// then a PL_WRITE_CHEAP lock is grabbed
+//
 int
 toku_pin_ftnode_batched(
     FT_HANDLE brt,
@@ -202,15 +207,22 @@ toku_pin_ftnode_batched(
     ANCESTORS ancestors,
     const PIVOT_BOUNDS bounds,
     FTNODE_FETCH_EXTRA bfe,
-    pair_lock_type lock_type,
     bool apply_ancestor_messages, // this bool is probably temporary, for #3972, once we know how range query estimates work, will revisit this
     FTNODE *node_p,
     bool* msgs_applied)
 {
     void *node_v;
     *msgs_applied = false;
-    pair_lock_type needed_lock_type = lock_type;
-try_again_for_write_lock:
+    FTNODE node = nullptr;
+    MSN max_msn_in_path = ZERO_MSN;
+    bool needs_ancestors_messages = false;
+    // this function assumes that if you want ancestor messages applied,
+    // you are doing a read for a query. This is so we can make some optimizations
+    // below.
+    if (apply_ancestor_messages) {
+        paranoid_invariant(bfe->type == ftnode_fetch_subset);
+    }
+    
     int r = toku_cachetable_get_and_pin_nonblocking_batched(
             brt->ft->cf,
             blocknum,
@@ -221,25 +233,52 @@ try_again_for_write_lock:
             toku_ftnode_fetch_callback,
             toku_ftnode_pf_req_callback,
             toku_ftnode_pf_callback,
-            needed_lock_type,
+            PL_READ,
             bfe, //read_extraargs
             unlockers);
-    if (r==0) {
-        FTNODE node = static_cast<FTNODE>(node_v);
-        MSN max_msn_in_path;
-        bool needs_ancestors_messages = false;
-        if (apply_ancestor_messages && node->height == 0) {
-            needs_ancestors_messages = toku_ft_leaf_needs_ancestors_messages(brt->ft, node, ancestors, bounds, &max_msn_in_path);
-            if (needs_ancestors_messages && needed_lock_type == PL_READ) {
-                toku_unpin_ftnode_read_only(brt->ft, node);
-                needed_lock_type = PL_WRITE_CHEAP;
-                goto try_again_for_write_lock;
-            }
+    if (r != 0) {
+        assert(r == TOKUDB_TRY_AGAIN); // Any other error and we should bomb out ASAP.
+        goto exit;
     }
+    node = static_cast<FTNODE>(node_v);
     if (apply_ancestor_messages && node->height == 0) {
+        needs_ancestors_messages = toku_ft_leaf_needs_ancestors_messages(
+            brt->ft, 
+            node, 
+            ancestors, 
+            bounds, 
+            &max_msn_in_path, 
+            bfe->child_to_read
+            );
         if (needs_ancestors_messages) {
-                invariant(needed_lock_type != PL_READ);
-                toku_apply_ancestors_messages_to_node(brt, node, ancestors, bounds, msgs_applied);
+            toku_unpin_ftnode_read_only(brt->ft, node);
+            int rr = toku_cachetable_get_and_pin_nonblocking_batched(
+                    brt->ft->cf,
+                    blocknum,
+                    fullhash,
+                    &node_v,
+                    NULL,
+                    get_write_callbacks_for_node(brt->ft),
+                    toku_ftnode_fetch_callback,
+                    toku_ftnode_pf_req_callback,
+                    toku_ftnode_pf_callback,
+                    PL_WRITE_CHEAP,
+                    bfe, //read_extraargs
+                    unlockers);
+            if (rr != 0) {
+                assert(rr == TOKUDB_TRY_AGAIN); // Any other error and we should bomb out ASAP.
+                r = TOKUDB_TRY_AGAIN;
+                goto exit;
+            }
+            node = static_cast<FTNODE>(node_v);
+            toku_apply_ancestors_messages_to_node(
+                brt, 
+                node, 
+                ancestors, 
+                bounds, 
+                msgs_applied,
+                bfe->child_to_read
+                );
         } else {
             // At this point, we aren't going to run
             // toku_apply_ancestors_messages_to_node but that doesn't
@@ -264,20 +303,12 @@ try_again_for_write_lock:
             // requires a write lock, and a write lock requires you to
             // resolve checkpointing.
             if (!node->dirty) {
-                    toku_ft_bn_update_max_msn(node, max_msn_in_path);
+                toku_ft_bn_update_max_msn(node, max_msn_in_path, bfe->child_to_read);
             }
         }
-            invariant(needed_lock_type != PL_READ || !*msgs_applied);
-        }
-        if ((lock_type != PL_READ) && node->height > 0) {
-            toku_move_ftnode_messages_to_stale(brt->ft, node);
     }
     *node_p = node;
-        // printf("%*sPin %ld\n", 8-node->height, "", blocknum.b);
-    } else {
-        assert(r==TOKUDB_TRY_AGAIN); // Any other error and we should bomb out ASAP.
-        // printf("%*sPin %ld try again\n", 8, "", blocknum.b);
-    }
+exit:
     return r;
 }
 

ft/ft-cachetable-wrappers.h

@@ -150,7 +150,6 @@ toku_pin_ftnode_batched(
     ANCESTORS ancestors,
     const PIVOT_BOUNDS pbounds,
     FTNODE_FETCH_EXTRA bfe,
-    pair_lock_type lock_type,
     bool apply_ancestor_messages, // this bool is probably temporary, for #3972, once we know how range query estimates work, will revisit this
     FTNODE *node_p,
     bool* msgs_applied

ft/ft-internal.h

@@ -727,13 +727,6 @@ STAT64INFO_S toku_get_and_clear_basement_stats(FTNODE leafnode);
 #define VERIFY_NODE(t,n) ((void)0)
 #endif
 
-//#define FT_TRACE
-#ifdef FT_TRACE
-#define WHEN_FTTRACE(x) x
-#else
-#define WHEN_FTTRACE(x) ((void)0)
-#endif
-
 void toku_ft_status_update_pivot_fetch_reason(struct ftnode_fetch_extra *bfe);
 void toku_ft_status_update_flush_reason(FTNODE node, uint64_t uncompressed_bytes_flushed, uint64_t bytes_written, tokutime_t write_time, bool for_checkpoint);
 void toku_ft_status_update_serialize_times(FTNODE node, tokutime_t serialize_time, tokutime_t compress_time);
@@ -982,11 +975,11 @@ struct pivot_bounds {
 
 __attribute__((nonnull))
 void toku_move_ftnode_messages_to_stale(FT ft, FTNODE node);
-void toku_apply_ancestors_messages_to_node (FT_HANDLE t, FTNODE node, ANCESTORS ancestors, struct pivot_bounds const * const bounds, bool* msgs_applied);
+void toku_apply_ancestors_messages_to_node (FT_HANDLE t, FTNODE node, ANCESTORS ancestors, struct pivot_bounds const * const bounds, bool* msgs_applied, int child_to_read);
 __attribute__((nonnull))
-bool toku_ft_leaf_needs_ancestors_messages(FT ft, FTNODE node, ANCESTORS ancestors, struct pivot_bounds const * const bounds, MSN *const max_msn_in_path);
+bool toku_ft_leaf_needs_ancestors_messages(FT ft, FTNODE node, ANCESTORS ancestors, struct pivot_bounds const * const bounds, MSN *const max_msn_in_path, int child_to_read);
 __attribute__((nonnull))
-void toku_ft_bn_update_max_msn(FTNODE node, MSN max_msn_applied);
+void toku_ft_bn_update_max_msn(FTNODE node, MSN max_msn_applied, int child_to_read);
 
 __attribute__((const,nonnull))
 size_t toku_ft_msg_memsize_in_fifo(FT_MSG cmd);

ft/tests/orthopush-flush.cc

@@ -696,7 +696,7 @@ flush_to_leaf(FT_HANDLE t, bool make_leaf_up_to_date, bool use_flush) {
         struct ancestors ancestors = { .node = parentnode, .childnum = 0, .next = NULL };
         const struct pivot_bounds infinite_bounds = { .lower_bound_exclusive = NULL, .upper_bound_inclusive = NULL };
         bool msgs_applied;
-        toku_apply_ancestors_messages_to_node(t, child, &ancestors, &infinite_bounds, &msgs_applied);
+        toku_apply_ancestors_messages_to_node(t, child, &ancestors, &infinite_bounds, &msgs_applied, -1);
 
         FIFO_ITERATE(parent_bnc->buffer, key, keylen, val, vallen, type, msn, xids, is_fresh,
                      {
@@ -921,7 +921,7 @@ flush_to_leaf_with_keyrange(FT_HANDLE t, bool make_leaf_up_to_date) {
         .upper_bound_inclusive = toku_clone_dbt(&ubi, childkeys[7])
     };
     bool msgs_applied;
-    toku_apply_ancestors_messages_to_node(t, child, &ancestors, &bounds, &msgs_applied);
+    toku_apply_ancestors_messages_to_node(t, child, &ancestors, &bounds, &msgs_applied, -1);
 
     FIFO_ITERATE(parent_bnc->buffer, key, keylen, val, vallen, type, msn, xids, is_fresh,
                  {
@@ -1104,7 +1104,7 @@ compare_apply_and_flush(FT_HANDLE t, bool make_leaf_up_to_date) {
     struct ancestors ancestors = { .node = parentnode, .childnum = 0, .next = NULL };
     const struct pivot_bounds infinite_bounds = { .lower_bound_exclusive = NULL, .upper_bound_inclusive = NULL };
     bool msgs_applied;
-    toku_apply_ancestors_messages_to_node(t, child2, &ancestors, &infinite_bounds, &msgs_applied);
+    toku_apply_ancestors_messages_to_node(t, child2, &ancestors, &infinite_bounds, &msgs_applied, -1);
 
     FIFO_ITERATE(parent_bnc->buffer, key, keylen, val, vallen, type, msn, xids, is_fresh,
                  {