Fix counting bug in buffer
Quoting @davisp:
There's a bug in rexi_buffer that can lead to the counter in its state
running negative. I noticed this on malort looking for memory usage. A
quick reading of the code suggests its due to us getting a timeout
message with an empty queue. Theoretically this could happen if we've
exceeded the MAX_MEMORY threshold when sending a message or even with
just grabbing the buffered count when its idle.
BugzID: 28049
diff --git a/src/rexi_buffer.erl b/src/rexi_buffer.erl
index f75399c..26f3c97 100644
--- a/src/rexi_buffer.erl
+++ b/src/rexi_buffer.erl
@@ -58,37 +58,25 @@
{noreply, State#state{buffer = Q2, count = C+1}, 0}
end.
-handle_info(timeout, #state{sender = nil} = State) ->
- #state{buffer = Q, count = C} = State,
- Sender = case queue:out_r(Q) of
- {{value, {Dest, Msg}}, Q2} ->
- case erlang:send(Dest, Msg, [noconnect, nosuspend]) of
- ok ->
- nil;
- _Else ->
- spawn_monitor(erlang, send, [Dest, Msg])
- end;
- {empty, Q2} ->
- nil
- end,
- if Sender =:= nil, C > 1 ->
- {noreply, State#state{buffer = Q2, count = C-1}, 0};
- true ->
- NewState = State#state{buffer = Q2, sender = Sender, count = C-1},
- % When Sender is nil and C-1 == 0 we're reverting to an
- % idle state with no outstanding or queued messages. We'll
- % use this oppurtunity to hibernate this process and
- % run a garbage collection.
- case {Sender, C-1} of
- {nil, 0} ->
- {noreply, NewState, hibernate};
- _ ->
- {noreply, NewState, infinity}
- end
- end;
-handle_info(timeout, State) ->
- % Waiting on a sender to return
+handle_info(timeout, #state{sender = nil, buffer = {[],[]}, count = 0}=State) ->
{noreply, State};
+handle_info(timeout, #state{sender = nil, count = C} = State) when C > 0 ->
+ #state{buffer = Q, count = C} = State,
+ {{value, {Dest, Msg}}, Q2} = queue:out_r(Q),
+ NewState = State#state{buffer = Q2, count = C-1},
+ case erlang:send(Dest, Msg, [noconnect, nosuspend]) of
+ ok when C =:= 1 ->
+ % We just sent the last queued messsage, we'll use this opportunity
+ % to hibernate the process and run a garbage collection
+ {noreply, NewState, hibernate};
+ ok when C > 1 ->
+ % Use a zero timeout to recurse into this handler ASAP
+ {noreply, NewState, 0};
+ _Else ->
+ % We're experiencing delays, keep buffering internally
+ Sender = spawn_monitor(erlang, send, [Dest, Msg]),
+ {noreply, NewState#state{sender = Sender}}
+ end;
handle_info({'DOWN', Ref, _, Pid, _}, #state{sender = {Pid, Ref}} = State) ->
{noreply, State#state{sender = nil}, 0}.
