LuoFucong created ZOOKEEPER-3102:
------------------------------------
Summary: Potential race condition when create ephemeral nodes
Key: ZOOKEEPER-3102
URL: https://issues.apache.org/jira/browse/ZOOKEEPER-3102
Project: ZooKeeper
Issue Type: Bug
Components: server
Affects Versions: 3.6.0
Environment: operating system: macOS High Sierra 10.13.6
java version: 8u152
Reporter: LuoFucong
The method
{code:java}
public void createNode(final String path, byte data[], List<ACL> acl, long
ephemeralOwner, int parentCVersion, long zxid, long time, Stat outputStat)
{code}
in class DataTree may conceal a potential race condition regarding the session
ephemeral nodes map "Map<Long, HashSet<String>> ephemerals".
Specifically, the codes start from line 455:
{code:java}
} else if (ephemeralOwner != 0) {
HashSet<String> list = ephemerals.get(ephemeralOwner);
if (list == null) {
list = new HashSet<String>();
ephemerals.put(ephemeralOwner, list);
}
synchronized (list) {
list.add(path);
}
}{code}
When an ephemeral owner tries to create nodes concurrently (under different
parent nodes), an empty "HashSet<String>" might be created multiple times, and
replace each other.
The following unit test reveals the race condition:
{code:java}
@Test(timeout = 60000)
public void testSessionEphemeralNodesConcurrentlyCreated()
throws InterruptedException, NodeExistsException, NoNodeException {
long session = 0x1234;
int concurrent = 10;
Thread[] threads = new Thread[concurrent];
CountDownLatch latch = new CountDownLatch(1);
for (int i = 0; i < concurrent; i++) {
String parent = "/test" + i;
dt.createNode(parent, new byte[0], null, 0, -1, 1, 1);
Thread thread = new Thread(() -> {
try {
latch.await();
} catch (InterruptedException e) {
throw new RuntimeException(e);
}
String path = parent + "/0";
try {
dt.createNode(path, new byte[0], null, session, -1, 1, 1);
} catch (Exception e) {
throw new IllegalStateException(e);
}
});
thread.start();
threads[i] = thread;
}
latch.countDown();
for (Thread thread : threads) {
thread.join();
}
int sessionEphemerals = dt.getEphemerals(session).size();
Assert.assertEquals(concurrent, sessionEphemerals);
}
{code}
The session "0x1234" has created 10 ephemeral nodes "/test\{0~9}/0"
concurrently (in 10 threads), so its ephemeral nodes size retrieved from
DataTree should be 10 while doesn't (assertion fail).
The fix should be easy:
{code:java}
private final ConcurrentMap<Long, HashSet<String>> ephemerals = new
ConcurrentHashMap<>();
...
} else if (ephemeralOwner != 0) {
HashSet<String> list = ephemerals.get(ephemeralOwner);
if (list == null) {
list = new HashSet<String>();
HashSet<String> _list;
if ((_list = ephemerals.putIfAbsent(ephemeralOwner, list)) != null) {
list = _list;
}
}
synchronized (list) {
list.add(path);
}
}
{code}
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
