Github user bschuchardt commented on a diff in the pull request:
https://github.com/apache/geode/pull/702#discussion_r133325744
--- Diff:
geode-core/src/main/java/org/apache/geode/internal/net/SocketCloser.java ---
@@ -96,46 +99,55 @@ public int getMaxThreads() {
return this.asyncClosePoolMaxThreads;
}
- private ThreadPoolExecutor getAsyncThreadExecutor(String address) {
- synchronized (asyncCloseExecutors) {
- ThreadPoolExecutor pool = asyncCloseExecutors.get(address);
- if (pool == null) {
- final ThreadGroup tg =
LoggingThreadGroup.createThreadGroup("Socket asyncClose", logger);
- ThreadFactory tf = new ThreadFactory() {
- public Thread newThread(final Runnable command) {
- Thread thread = new Thread(tg, command);
- thread.setDaemon(true);
- return thread;
- }
- };
- BlockingQueue<Runnable> workQueue = new
LinkedBlockingQueue<Runnable>();
- pool = new ThreadPoolExecutor(this.asyncClosePoolMaxThreads,
this.asyncClosePoolMaxThreads,
- this.asyncClosePoolKeepAliveSeconds, TimeUnit.SECONDS,
workQueue, tf);
- pool.allowCoreThreadTimeOut(true);
- asyncCloseExecutors.put(address, pool);
+ private ExecutorService getAsyncThreadExecutor(String address) {
+ ExecutorService executorService = asyncCloseExecutors.get(address);
+ if (executorService == null) {
+ // To be used for pre-1.8 jdk releases.
+ // createThreadPool();
+
+ executorService =
Executors.newWorkStealingPool(asyncClosePoolMaxThreads);
+
+ ExecutorService previousThreadPoolExecutor =
+ asyncCloseExecutors.putIfAbsent(address, executorService);
+
+ if (previousThreadPoolExecutor != null) {
+ executorService.shutdownNow();
+ return previousThreadPoolExecutor;
}
- return pool;
}
+ return executorService;
+ }
+
+ /**
+ * @deprecated this method is to be used for pre 1.8 jdk.
+ */
+ @Deprecated
+ private void createThreadPool() {
+ ExecutorService executorService;
+ final ThreadGroup threadGroup =
+ LoggingThreadGroup.createThreadGroup("Socket asyncClose", logger);
+ ThreadFactory threadFactory = new ThreadFactory() {
+ public Thread newThread(final Runnable command) {
+ Thread thread = new Thread(threadGroup, command);
+ thread.setDaemon(true);
+ return thread;
+ }
+ };
+
+ executorService = new ThreadPoolExecutor(asyncClosePoolMaxThreads,
asyncClosePoolMaxThreads,
+ asyncCloseWaitTime, asyncCloseWaitUnits, new
LinkedBlockingQueue<>(), threadFactory);
}
/**
* Call this method if you know all the resources in the closer for the
given address are no
* longer needed. Currently a thread pool is kept for each address and
if you know that an address
* no longer needs its pool then you should call this method.
*/
- public void releaseResourcesForAddress(String address) {
- synchronized (asyncCloseExecutors) {
- ThreadPoolExecutor pool = asyncCloseExecutors.get(address);
- if (pool != null) {
- pool.shutdown();
- asyncCloseExecutors.remove(address);
- }
- }
- }
- private boolean isClosed() {
- synchronized (asyncCloseExecutors) {
- return this.closed;
+ public void releaseResourcesForAddress(String address) {
+ ExecutorService executorService = asyncCloseExecutors.remove(address);
+ if (executorService != null) {
+ executorService.shutdown();
--- End diff --
When the cache closes it invokes ConnectionTable.close() which closes each
connection and _afterwards_ closes its SocketCloser. All socket closing in
this case will be performed asynchronously.
The same is done in CacheClientNotifier. All dispatching sockets are
closed and then the SocketCloser is closed.
Client proxy and connection table receiver threads are closed by shutting
down their executors.
Both of these behaviors are essential if the auto-reconnect mechanism is
going to work properly. Otherwise shutdown of the old cache may hang waiting
on the keepalive timeout.
I think extensive testing of this change is needed. Network partition
detection and auto-reconnect testing needs to be done to ensure that the cache
can be properly closed under harsh conditions.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
