MB-4411 A worker should wait for a signal from its parent process.

A batch worker might finish its execution before its parent process gets the worker's monitor Ref value. Consequently, this causes a crash in mc_daemon process. To resolve this issue, each batch worker should wait for a signal from its parent process before starting its execution. Change-Id: Iab63c1608a6e3fe6367654d996b7484d9f56b623 Reviewed-on: http://review.couchbase.org/10752 Tested-by: Chiyoung Seo <chiyoung.seo@gmail.com> Reviewed-by: Aliaksey Artamonau <aliaksiej.artamonau@gmail.com>
couchbase · Nov 10, 2011 · 0f5c8ce · 0f5c8ce
1 parent 2ea6aba
commit 0f5c8ce
Show file tree

Hide file tree

Showing 2 changed files with 21 additions and 10 deletions.
diff --git a/src/mc_batch_sup.erl b/src/mc_batch_sup.erl
@@ -27,13 +27,19 @@ init([]) ->
 start_worker(Sup, CurrentVBucket, CurrentList, BucketName, Socket) ->
    {ok, Pid} = supervisor:start_child(Sup, [CurrentVBucket, CurrentList, BucketName, Socket]),
    Ref = monitor(process, Pid),
+   Pid ! can_start,
    {ok, Ref}.
 
 start_link_worker(CurrentVBucket, CurrentList, BucketName, Socket) ->
     {ok, proc_lib:spawn_link(?MODULE, sync_update_docs,
                              [CurrentVBucket, CurrentList, BucketName, Socket])}.
 
 sync_update_docs(CurrentVBucket, CurrentList, BucketName, Socket) ->
+    receive
+        can_start ->
+            ok
+    end,
+
     UpdateOptions =
         [clobber, return_errors_only] ++
         case couch_config:get("mc_couch", "optimistic_writes", "true") of

diff --git a/src/mc_daemon.erl b/src/mc_daemon.erl
@@ -28,7 +28,7 @@
           errors = [],
           next_vb_batch,
           current_vbucket,
-          current_vbucket_list,
+          current_vbucket_list = [],
           max_workers = 4,
           worker_sup,
           caller = nil,
@@ -275,9 +275,9 @@ batching({?NOOP, Opaque}, From, State) ->
           socket = Socket,
           current_vbucket = CurrentVBucket, current_vbucket_list = CurrentList
     } = State,
-    mc_batch_sup:sync_update_docs(CurrentVBucket, CurrentList, State#state.db, Socket),
-    State2 = State#state{current_vbucket = 0, current_vbucket_list = []},
-    case num_workers(State) of
+    State2 = maybe_start_worker(State#state{next_vb_batch = {CurrentVBucket, CurrentList},
+                                            current_vbucket = 0, current_vbucket_list = []}),
+    case num_workers(State2) of
         0 ->
             mc_connection:respond(Socket, ?NOOP, Opaque, #mc_response{}),
             gen_fsm:reply(From, ok),
@@ -336,12 +336,17 @@ maybe_start_worker(State) ->
             worker_refs = WorkerRefs,
             next_vb_batch = {VBucket, ItemList}
           } = State,
-    {ok, WorkerRef} = mc_batch_sup:start_worker(WorkerSup, VBucket, ItemList,
-                                                State#state.db, Socket),
-    State#state{
-      next_vb_batch = {},
-      worker_refs = [WorkerRef|WorkerRefs]
-    }.
+    case ItemList /= [] of
+        true ->
+            {ok, WorkerRef} = mc_batch_sup:start_worker(WorkerSup, VBucket, ItemList,
+                                                        State#state.db, Socket),
+            State#state{
+                next_vb_batch = {VBucket, []},
+                worker_refs = [WorkerRef|WorkerRefs]
+            };
+        false ->
+            State
+    end.
 
 handle_info({'DOWN', Ref, process, _Pid, normal}, batching, State) ->
     #state{caller = From, worker_refs = WorkerRefs} = State,