Redis optimization: avoid querying Redis when threshold is violated

[oleg-smith]

Hi, I would like to verify Redis integration behavior for this scenario:

• a bucket backed by Redis
• the bucket bandwidth is 50 requests/min, classic
• say, all 50 tokens are consumed in the very first second of a minute.

1. Would the library query Redis for the rest of the minute, knowing the threshold had been already violated?

2. If it will query Redis by default, is there a way to configure it not to?

Thanks!

[vladimir-bukhtoyarov]

Hello @oleg-smith

Refill is not a single source of increasing tokens in the bucket. There are API that can to increasing tokens in the bucket:

• addTokens
• forceAddTokens
• reset
• replaceConfiguration

So, proposed optimization can not be applied by default because library does not know the scope of API that user intents to call. Anyway, this kind of optimization can be configured expilitly. There are big package of library that is not documented yet https://github.com/bucket4j/bucket4j/tree/8.4/bucket4j-core/src/main/java/io/github/bucket4j/distributed/proxy/optimization this package is addressed to implement varios optimization strategeis, feel free to read javadocs for classes in this packet, it can be interesting.

I have just added optimization for your use case in this commit d1e4d75 The class HazelcastSkipSyncOnZeroExample shows how to use this optimization:

The 8.4.0 is going to be released at Jule, so just copy and past my code to your project if you have no time to wait.

[oleg-smith]

Thanks very much!

[oleg-smith]

Hi @vladimir-bukhtoyarov, I tried it this way:

remoteProxyManager = LettuceBasedProxyManager
           ...
            .build();

Then, accessing the same bucket from multiple threads this way:

var optimizationListener = new DefaultOptimizationListener();
        var optimization = new SkipSyncOnZeroOptimization(optimizationListener, TimeMeter.SYSTEM_NANOTIME)
            .withListener(optimizationListener);

        return remoteProxyManager.builder()
            .withOptimization(optimization)
            .build(
                key.getBytes(),
                () -> BucketConfiguration.builder()
                    .addLimit(Bandwidth.classic(capacity, Refill.intervally(capacity, period)))
                    .build()
            );

in debug, I'm seeing isNeedToExecuteRemoteImmediately always returns true:

Also, in the debug, the optimization object is always a new object the way I use it above - I assume that is the issue.

• Should I cache the optimization object additionally by the bucket key using CHM, for instance?

• Is optimization object thread-safe by contract?

• Could you please suggest a better way for configuring an optimization and preserving its state, when accessed from multiple threads?

[vladimir-bukhtoyarov]

The whole code

[vladimir-bukhtoyarov]

I have fixed the example. TimeMeter.SYSTEM_MILLISECONDS should be used

[oleg-smith]

thanks, let me try!

Btw, isn't it enough to use new SkipSyncOnZeroOptimization(optimizationListener, TimeMeter.SYSTEM_MILLISECONDS)?

without .withListener() part

[vladimir-bukhtoyarov]

Yes it is enough. Listener is needed only to collect statistics, nothing more.

[oleg-smith]

Yeah, I meant that the listener is passed 2 times - 1st in the constructor, and 2nd in the withListener, basically making a copy

that just seemed excessive

[vladimir-bukhtoyarov]

released with version 8.4.0