[
https://issues.apache.org/jira/browse/POOL-419?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Phil Steitz resolved POOL-419.
------------------------------
Fix Version/s: 2.12.2
Resolution: Fixed
In 078f737490805b38b8993287c6b3bcfbd27332da, I committed a fix for this in the
2_X branch. Thanks for the report, Raju, and for the test, which I committed
unchanged other than to adapt for 2_X and some Checkstyle nits. My fix was
to fully synchronize returnObject on the pooled object's monitor. The code
already acquires the monitor (on some paths multiple times) and the only
potential contention is with the evictor, which should not be visiting checked
out objects, so I think within pool itself, the added sync should not cause
performance problems.
I thought about recommending WONT_FIX for this issue because having multiple
client threads invoke pool lifecycle methods on the same pooled object
concurrently could be viewed as violating the API contract. But we don't
explicitly disallow this anywhere, so we need to address these kinds of issues
when they are reported. In general, the sync protections on pooled objects
within pool are to handle contention between client threads and the evictor,
not multiple client threads concurrently invoking lifecycle methods on the same
pooled object.
> GenericObjectPoolConfig getNumActive return negative value
> ----------------------------------------------------------
>
> Key: POOL-419
> URL: https://issues.apache.org/jira/browse/POOL-419
> Project: Commons Pool
> Issue Type: Bug
> Affects Versions: 2.12.0
> Reporter: john ms
> Priority: Major
> Fix For: 2.12.2
>
>
> We've notice unexpected counting that is return from getNumActive .
> and it means that allObjects and idleObjects are out of sync.
> Calling returnObject and invalidateObject on the same pooled object from two
> different threads is causing getNumActive to return negative value.
> Why two threads are working on the same object at the same time and the real
> running use case is much complex to described and not relevant, I think.
> Bellow is a simple program to demo the behavior
> {*}Expected result{*}:
> getNumActive=0
>
> {*}Actual result{*}:
> getNumActive=\{negative number}
> If it help debugging the RC, I've notice that returnObject is not
> synchronized over pooled object and verify it's status before using it, like
> it's done in invalidateObject.
> {*}Workaround{*}:
> Synchronize the pooled object before calling returnObject/invalidateObject
>
> Java 17
> {code:java}
> public class PoolObject extends BasePooledObjectFactory<PoolObject> {
> @Override
> public PoolObject create() {
> return new PoolObject();
> }
> @Override
> public PooledObject<PoolObject> wrap(PoolObject poolObject) {
> return new DefaultPooledObject<>(poolObject);
> }
> @Override
> public void passivateObject(PooledObject<PoolObject> pooledObject) {
> }
> }{code}
>
> {code:java}
> public static void main (String [] args) throws Exception {
> ExecutorService executor = Executors.newCachedThreadPool();
> GenericObjectPoolConfig<PoolObject> poolConfig = new
> GenericObjectPoolConfig<>();
> poolConfig.setMaxTotal(200000);
> poolConfig.setMaxIdle(200000);
> poolConfig.setMinIdle(200000);
> ObjectPool<PoolObject> objectPool = new GenericObjectPool<>(new
> PoolObject(), poolConfig);
> for (int i=0; i<1000000; i++){
> PoolObject poolObject= objectPool.borrowObject();
> FutureTask invalidateObject = new FutureTask<>(() -> {
> Thread.sleep(RandomUtils.nextInt(0,10));
> //synchronized (poolObject) { // Workaround bug
> objectPool.invalidateObject(poolObject);
> //}
> return true;
> });
> executor.execute(invalidateObject);
> FutureTask returnObject = new FutureTask<>(() -> {
> Thread.sleep(RandomUtils.nextInt(0,10));
> //synchronized (poolObject) { // Workaround bug
> objectPool.returnObject(poolObject);
> //}
> return true;
> });
> executor.execute(returnObject);
> }
> Thread.sleep(2000);
> executor.shutdown();
> System.out.println("getNumActive=" + objectPool.getNumActive());
> }{code}
>
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)