Allow users to specify the redis key via a strategy of some kind

[spencergibb]

Previously, there was a bug that leaked rate limiter keys across routes #2288

#2517 added the routeId to that key, but that may not be ideal in all situations.

See #2517 (comment)

[iis-MarkKuang]

I read #2517 and felt the same way, there are 2 scenarios in my case where some flexibility could be helpful:

1. Share the same bucket quota across more than one route;
2. Allow for different bucket quota and replenish rate for different keys.

In my case, I had to write my own version of the RedisRateLimiter(to be exact, re-write the isAllowed function) to implement these, where the passed in ApiKey is retrieved from the Exchange object, as the previous ApiKey check filter stores the apikey in the exchange once the apikey is validated.

public Mono<Response> isAllowed(ApiKey apiKey) {
    // Check if the rate limiter is initialized.
    if (!this.initialized.get()) {
      throw new IllegalStateException("RedisRateLimiter is not initialized");
    }

    // Use apikey from context(Put in by previous ApiKeyAuthGatewayFactory).
    if (apiKey == null) {
      log.warn("ApiKey doesn't exist, " + apiKey.getId());
      return Mono.just(new Response(false, getHeaders(0L)));
    }

    int requestedTokens = 1;

    // Invokes LUA script for rate-limiting based on tokens bucket algorithm.
    try {
      List<String> keys = getKeys(apiKey.getId());

      int replenishRate = apiKey.getReplenishRate();
      int burstCapacity = apiKey.getBurstCapacity();

      // The arguments to the LUA script. time() returns unix timestamp in seconds.
      List<String> scriptArgs = Arrays.asList(String.valueOf(replenishRate), String.valueOf(burstCapacity), "", String.valueOf(requestedTokens));
      // Get whether allowed(boolean) and tokens left(int) by executing lua script.
      Flux<List<Long>> flux = this.redisTemplate.execute(this.script, keys, scriptArgs);
      return flux.onErrorResume(throwable -> {
        log.error("Error calling rate limiter lua", throwable);
        return Flux.just(Arrays.asList(1L, -1L));
      }).reduce(new ArrayList<Long>(), (longs, l) -> {
        longs.addAll(l);
        return longs;
      }).map(results -> {
        boolean allowed = results.get(0) == 1L;
        Long tokensLeft = results.get(1);

        Response response = new Response(allowed, getHeaders(tokensLeft));

        return response;
      });
    }
    catch (Exception e) {

      log.error(RATE_LIMITER_GENERAL_ERROR, e);
    }
    return Mono.just(new Response(true, getHeaders(-1L)));
  }

Maybe some design is needed to form the redis/bucket4j rate limit keys to better support these requirements.

[LorenzoLuconi]

I think this change causes more problems than solutions

I just added a comment to #2517.

[imazzeii]

if RedisRateLimiter.getKeys is turned into an instance method it could be easily overriden