in_forward: Fixed a problem where connections are not closed on fluentd's shutdown. #861
Conversation
| end | ||
|
|
||
| def listen | ||
| log.info "listening fluent socket on #{@bind}:#{@port}" | ||
| s = Coolio::TCPServer.new(@bind, @port, Handler, @linger_timeout, log, method(:on_message)) | ||
| s = Coolio::TCPServer.new(@bind, @port, Handler, @linger_timeout, log, method(:on_message)) do |connection| | ||
| @connections.reject! { |conn| conn.closed? } |
There was a problem hiding this comment.
Should this block guard using Mutex?
There was a problem hiding this comment.
I think it doesn't need to be guarded.
(I would like to confirm just in case; does this block mean do |connection| ... end, not { |conn| conn.closed? }, right?)
@connections is used for closing accepted connections at shutdown, so already closed connections don't need to be included in. I think it is safe that several processes or threads come in this block at the same time.
There was a problem hiding this comment.
It looks needed for me too to guard @connections.reject! using Mutex.
I mean that:
@mutex.synchronize do
@connections.reject!{ |conn| conn.closed? }
@connections << connection
endConnection callback may run twice immediately if many connection arrived at a time.
There was a problem hiding this comment.
Thank you for your comments.
Surely, I think I should guard it by mutex, but please let me ask you a question.
Can this callback be run simultaneously by threads?
It seems that only one thread(@thread) handles waiting connections, communicating with accepted connections and replying heartbeat request. So only one thread can run this callback at a time, cannot it?
|
This issue about closing sockets appropriately will be re-checked after porting |
|
I created #1401 to solve this problem at v0.14 server plugin helper based code. |
At shutdown,
in_forwardplugin doesn't close connections accepted by Coolio::TCPServer, so clients continue to sending requests which main process cannot recognize and are to be lost until the process exits. (I think #801 arises from the same cause as this)I have fixed this problem by keeping accepted connections and closing them at shutdown.