cc @slorello89 for heads up on these, may affect others!

Hey Nick - the error your seeing here is caused by slightly different behavior between the Release Candidate and the actual release of Redis 7:

While iterating the PEL, if XAUTOCLAIM stumbles upon a message which doesn't exist in the stream anymore (either trimmed or deleted by [XDEL](https://redis.io/commands/xdel)) it does not claim it, and deletes it from the PEL in which it was found. This feature was introduced in Redis 7.0. These message IDs are returned to the caller as a part of XAUTOCLAIMs reply.

Might seem innocuous but when you take it in context with:

The optional <count> argument, which defaults to 100, is the upper limit of the number of entries that the command attempts to claim.

Tells you what's going on, the test is sending the XAUTOCLAIM with a COUNT of 1 - it attempts to claim the deleted message. THAT COUNTS as its one message it will attempt to claim, and it breaks off. It returns the next message as the place to start the next autoclaim, returns a null entry as the actual messages claimed, and returns the deleted message in the third spot.

The fix is simple enough if you want me to push it (just change the counts to 2 :) )

@slorello89 If the test is wrong and server is right, absolutely - would be much appreciated. I'd love for tests to be wrong here :)

@NickCraver

Insofar as XAUTOCLAIM the server is behaving correctly, so the update is correct

WRT the other issue with the perf test - this is a BC coming to Redis 7.2:

redis/redis#10182
redis/redis#10300

See the Breaking section of the release notes - basically what's happening is that Redis 7.2 is freezing the clock during script execution. Consequently, that test is no longer correct and will always fail. Obviously there's a couple ways to remediate, but I'm not entirely sure what that test is really testing?

Anyway - hopefully this is helpful

@slorello89 very helpful - thank you!!

@mgravell any thoughts about the perf test/intent? For quick context, this is the entire content of the test:

    [Fact]
    public void HackyGetPerf()
    {
        using var conn = GetScriptConn();

        var key = Me();
        var db = conn.GetDatabase();
        db.StringSet(key + "foo", "bar", flags: CommandFlags.FireAndForget);
        var result = (long)db.ScriptEvaluate(@"
redis.call('psetex', KEYS[1], 60000, 'timing')
for i = 1,5000 do
    redis.call('set', 'ignore','abc')
end
local timeTaken = 60000 - redis.call('pttl', KEYS[1])
redis.call('del', KEYS[1])
return timeTaken
", new RedisKey[] { key }, null);
        Log(result.ToString());
        Assert.True(result > 0);
    }

The test originally comes from a Booksleeve migration 9 years ago...I'm leaning towards "delete it": c002a00