[GitHub] chewbranca commented on a change in pull request #504: Opimize writing KV node append writes

[gi...@git.apache.org]

chewbranca commented on a change in pull request #504: Opimize writing KV node append writes
URL: https://github.com/apache/couchdb/pull/504#discussion_r114857890
 
 

 ##########
 File path: src/couch/src/couch_btree.erl
 ##########
 @@ -440,6 +441,71 @@ write_node(#btree{fd = Fd, compression = Comp} = Bt, NodeType, NodeList) ->
     ],
     {ok, ResultList}.
 
+% Don't make our append-only write optimization for
+% kp nodes.
+write_node(Bt, _OldNode, kp_node, _OldList, NewList) ->
+    write_node(Bt, kp_node, NewList);
+
+% If we're creating a new kv node then there's no
+% possibility for the optimization
+write_node(Bt, _OldNode, NodeType, [], NewList) ->
+    write_node(Bt, NodeType, NewList);
+
+% Disable the optimization for nodes that only
+% have a single element so we don't end up increasing
+% the number of reads when folding a btree
+write_node(Bt, _OldNode, NodeType, [_], NewList) ->
+    write_node(Bt, NodeType, NewList);
+
+% If a KV node has had a new key appended to the
+% end of its list we can instead take the appended
+% KVs and create a new node while reusing the old
+% node already on disk. This saves us both the effort
+% of writing data that's already on disk as well as
+% saves us the disk space that would have been
+% orphaned by not reusing the old node.
+write_node(Bt, OldNode, NodeType, OldList, NewList) ->
+    case is_append_only(OldList, NewList) of
+        false ->
+            write_node(Bt, NodeType, NewList);
+        {true, Suffix} ->
+            case old_node_full(OldList) of
+                true ->
+                    {ok, Results} = write_node(Bt, NodeType, Suffix),
+                    {OldLastKey, _} = lists:last(OldList),
+                    {ok, [{OldLastKey,OldNode} | Results]};
+                false ->
+                    write_node(Bt, NodeType, NewList)
+            end
+    end.
+
+% This function will blow up if OldList == NewList
+% on purpose as an assertion that modify_node
+% doesn't provide this input.
 
 Review comment:
   Dangling comment?
 
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services