Java concurrency, stream behavior

Question

Methods getFirst() and getSecond() of this class is invoked concurrently. It is a part of a web app.

Internal maps populated as well with no concurrency.

public class MyClass {
    private Map<String, List<List<String>>> first;

    private Map<String, List<List<String>>> second;

    public MyClass() {
        first = new ConcurrentHashMap<>();
        second = new ConcurrentHashMap<>();
    }

    public Set<String> getFirst(String key, String token, int a, int b) {
        return get(first, key, token, a, b);
    }

    public Set<String> getSecond(String key, String token, int a, int b) {
        return get(second, key, token, a, b);
    }

    private Set<String> get(final Map<String, List<List<String>>> map, final String key, final String token,
        final int a, final int b) {
        Set<String> result = new TreeSet<>();
        map.get(key).stream().filter(i -> i.size() <= b && i.size() >= a).forEach(
            s -> result
                .addAll(s.stream().filter(p -> StringUtils.containsIgnoreCase(p, token)).collect(Collectors.toList())));
        return result;
    }
}

I tested it with something like ab -n 10000 -c 100 (Apache's utility). And I log it. I get the same set all times. But if I change map.get(key).stream() to map.get(key).parallelStream() and do the same steps I get sometimes different result size (always smaller).

What is it?

Answer 1

You are using TreeSet.addAll() inside the forEach of parallel stream. The forEach body can be executed simultaneously several times in different threads for different elements and TreeSet is not thread-safe. To fix the problem quickly you can either synchronize modification of result or use forEachOrdered . However it would be cleaner and more performant to flatMap your stream and collect it at once without forEach . Try this version:

return map.get(key).stream()
        .filter(i -> i.size() <= b && i.size() >= a)
        .flatMap(List::stream).filter(p -> StringUtils.containsIgnoreCase(p, token))
        .collect(Collectors.toCollection(TreeSet::new));