ivmaykov commented on a change in pull request #753: ZOOKEEPER-3204: Reconfig
tests are constantly failing on 3.5 after applying Java 11 fix
URL: https://github.com/apache/zookeeper/pull/753#discussion_r254438888
##########
File path:
zookeeper-server/src/main/java/org/apache/zookeeper/server/NettyServerCnxnFactory.java
##########
@@ -324,30 +260,42 @@ public void operationComplete(ChannelFuture future)
CnxnChannelHandler channelHandler = new CnxnChannelHandler();
- NettyServerCnxnFactory() {
- bootstrap = new ServerBootstrap(
- new NioServerSocketChannelFactory(
- Executors.newCachedThreadPool(),
- Executors.newCachedThreadPool()));
- // parent channel
- bootstrap.setOption("reuseAddress", true);
- // child channels
- bootstrap.setOption("child.tcpNoDelay", true);
- /* set socket linger to off, so that socket close does not block */
- bootstrap.setOption("child.soLinger", -1);
- bootstrap.setPipelineFactory(new ChannelPipelineFactory() {
- @Override
- public ChannelPipeline getPipeline() throws Exception {
- ChannelPipeline p = Channels.pipeline();
- if (secure) {
- initSSL(p);
- }
- p.addLast("servercnxnfactory", channelHandler);
+ private ServerBootstrap configureBootstrapAllocator(ServerBootstrap
bootstrap) {
+ ByteBufAllocator testAllocator = TEST_ALLOCATOR.get();
+ if (testAllocator != null) {
+ return bootstrap
+ .option(ChannelOption.ALLOCATOR, testAllocator)
+ .childOption(ChannelOption.ALLOCATOR, testAllocator);
+ } else {
+ return bootstrap;
+ }
+ }
- return p;
- }
- });
+ NettyServerCnxnFactory() {
x509Util = new ClientX509Util();
+
+ EventLoopGroup bossGroup = NettyUtils.newNioOrEpollEventLoopGroup();
+ EventLoopGroup workerGroup = NettyUtils.newNioOrEpollEventLoopGroup();
Review comment:
I think we probably want a separate accept group since ZK could have a lot
of clients (tens of thousands) reconnecting within a very short time after a
machine fails and a new leader is elected.
Maybe I could set the accept group to use 4 threads for now? I think it's
unlikely that a ZK server will bind on more than 4 network interfaces. It's a
bit more conservative than using 1 thread, and if we only bind on 1 address
then we end up "wasting" 3 threads instead of dozens. What do you think?
I guess we could also make the number of bind threads configurable. Or maybe
there's a dynamic way to detect the number of network interfaces present?
----------------------------------------------------------------
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:
[email protected]
With regards,
Apache Git Services
