No eviction with asMap.keySet.toArray

[zzrough]

hey,

thanks for the great library.

I'm using it for a long time, but recently I came up with something strange regarding expireAfterWrite eviction.
I don't get exactly get the semantics of the Map view (asMap), since it can or can not be up to date like the API docs tells.

But still, I've got this issue in production that puzzles me:

    public static void main(String[] args) {
        Object PRESENT = new Object();

        Cache<String,Object> cache = Caffeine.newBuilder()
                                             .expireAfterWrite(Duration.ofSeconds(3))
                                             .build();

        Consumer<Long> sleepQuietly = millis -> {
            try {
                Thread.sleep(millis);
            } catch (InterruptedException e) {
            }
        };

        new Thread(() -> {
            String name = "u1";

            for (int i = 0; i < 10; i++) {
                cache.put(name, PRESENT);
                // System.out.println(name + ": crowd = " + List.copyOf(cache.asMap().keySet()));            // ERR
                System.out.println(name + ": crowd = " + cache.asMap().keySet().stream().collect(toList())); // OK
                sleepQuietly.accept(500L);
            }
        }).start();

        new Thread(() -> {
            String name = "u2";

            sleepQuietly.accept(1000L);
            cache.put(name, PRESENT);
            // System.out.println(name + ": crowd = " + List.copyOf(cache.asMap().keySet()));            // ERR
            System.out.println(name + ": crowd = " + cache.asMap().keySet().stream().collect(toList())); // OK
        }).start();
    }

Simply put:

• eviction is after write 3s
• u1 thread put every 500ms
• u2 thread put after 1s

If I use List.copyOf that relies on toArray() on the key set view, u1 never sees u2 being evicted:

u1: crowd = [u1]
u1: crowd = [u1]
u2: crowd = [u1, u2]
u1: crowd = [u1, u2]
u1: crowd = [u1, u2]
u1: crowd = [u1, u2]
u1: crowd = [u1, u2]
u1: crowd = [u1, u2]
u1: crowd = [u1, u2]
u1: crowd = [u1, u2]
u1: crowd = [u1, u2]

If the set is traversed using a stream, or the iterator in anyway, I've got the proper eviction:

u1: crowd = [u1]
u1: crowd = [u1]
u2: crowd = [u1, u2]
u1: crowd = [u1, u2]
u1: crowd = [u1, u2]
u1: crowd = [u1, u2]
u1: crowd = [u1, u2]
u1: crowd = [u1, u2]
u1: crowd = [u1, u2]
u1: crowd = [u1]
u1: crowd = [u1]

Is this normal that keySet iteration works but not the toArray method? If I use List.copyOf or new ArrayList this does not work.
I certainly was not expected this :)

thanks a lot,

[zzrough]

(calling cleanUp manually each time did not do a difference)

[ben-manes]

Thanks for the test case. It looks like this was optimized to use the backing CHM's faster toArray, but missed that it created an inconsistency by not checking for expiration and only iterating if weakKeys is enabled. The values and entrySet iterate, but since keySet might be optimizable it has this overload which contains that oversight. Guava always iterates, so we should be consistent.

caffeine/caffeine/src/main/java/com/github/benmanes/caffeine/cache/BoundedLocalCache.java

Lines 2792 to 2802 in 8c91f65

	@Override
	public Object[] toArray() {
	if (cache.collectKeys()) {
	List<Object> keys = new ArrayList<>(size());
	for (Object key : this) {
	keys.add(key);
	}
	return keys.toArray();
	}
	return cache.data.keySet().toArray();
	}

[zzrough]

Thanks for the very quick response and detailed explanation. Le 11 novembre 2020 20:17:38 GMT+01:00, Ben Manes <notifications@github.com> a écrit :

…

Thanks for the test case. It looks like this was optimized to use the backing CHM's faster `toArray`, but missed that it created an inconsistency by not checking for expiration and only iterating if `weakKeys` is enabled. The `values` and `entrySet` iterate, but since `keySet` might be optimizable it has this overload which contains that oversight. Guava always iterates, so we should be consistent. https://github.com/ben-manes/caffeine/blob/8c91f6544f8f7cbfa5275aa05cce0713ea7f2863/caffeine/src/main/java/com/github/benmanes/caffeine/cache/BoundedLocalCache.java#L2792-L2802 -- You are receiving this because you authored the thread. Reply to this email directly or view it on GitHub: #472 (comment)

--S

[zzrough]

wow both a quick response and a quick fix, amazing thanks a lot!

[ben-manes]

haha, yes. If this is frustrating then I can cut a release over the weekend. Just let me know.

[zzrough]

don't bother with an extra release if it's much work. I've worked around it simply using a collector like in the testcase.

Just release when you feel it's needed. I'll subscribe to the project release notifications.

Thanks again, have a great day.

[ben-manes]

Released 2.8.7