[OAK-9963] Performance degradation when reading system properties on each NodeImpl

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 1.22.11
• Component/s: jcr
• Assignee: Unassigned
• Reporter: Paul Chibulcuteanu
• Created: 30/Sep/22 16:31
• Updated: 30/Sep/22 16:31

Description

Reading system properties on each NodeImpl causes performance issues especially on Java 8.
Code:

    public NodeImpl(T dlg, SessionContext sessionContext) {
        super(dlg, sessionContext);
        logWarnStringSizeThreshold = Integer.getInteger(
                OakJcrConstants.WARN_LOG_STRING_SIZE_THRESHOLD_KEY,
                OakJcrConstants.DEFAULT_WARN_LOG_STRING_SIZE_THRESHOLD_VALUE);
    }

One more important point on system property lookup in Oak. There is a huge difference between running the code on Java 8 and 11. 

The following test looks up a system property with 32 concurrent threads. Each reading the system property 10M times.

    @Test
    public void run() throws Exception {
        List<Thread> threads = new ArrayList<>();
        for (int i = 0; i < 32; i++) {
            threads.add(new Thread(this::getProperty));
        }
        Stopwatch sw = Stopwatch.createStarted();
        threads.forEach(Thread::start);
        for (Thread t : threads) {
            t.join();
        }
        System.out.println("time: " + sw);
    }

    private void getProperty() {
        for (int i = 0; i < 10_000_000; i++) {
            System.getProperty("foo");
        }
    }

On my machine the outputs are:

Java 8: ~20 seconds
Java 11: 0.5 seconds
Reason is how Properties implementation changed with Java 11. It is now using a ConcurrentHashMap.