Batch update

[didierofrivia]

This PR replaces the way Limitador, configured to use Redis Cached storage, updates (flushes the in memory cache) to Redis. Instead of updating counters one by one, Limitador sends the batch of counters to update to Redis.

TODO

• Batch updating to redis
• Fix rebase conficts
• Limit batch update counters to redis (?) for the near future
• Get back the values from redis...
• and update cache (?)
• Test that update counters are being updated in redis
• Test that the cached counters are updated

Notes

• In a setup where the Redis instance is (physically) far from the Limitador instance, it could happen that we encounter network partition error when the response timeout is exceeded:

  EC Frankfurt – Limitador Sao Paulo

Executing limitador: $ limitador-server /bench/limits.yaml redis_cached redis://test-limitador-redis.jeigrh.ng.0001.euc1.cache.amazonaws.com:6379
Limitador Server v1.4.0-dev (b7e4d87c) [] release
Executing limitador-driver: $ limitador-driver rpc://127.0.0.1:8081 1m
Hitting Limitador at rpc://127.0.0.1:8081 for a total duration of 1 minutes
Current: hits: 12195, mean: 0.081ms, p50: 0.062ms, p90: 0.073ms, p99: 0.092ms, p999: 0.106ms, max: 201.727ms
Current: hits: 15461, mean: 0.064ms, p50: 0.062ms, p90: 0.074ms, p99: 0.091ms, p999: 0.112ms, max: 0.335ms
Current: hits: 15367, mean: 0.064ms, p50: 0.062ms, p90: 0.074ms, p99: 0.092ms, p999: 0.112ms, max: 0.399ms
2024-04-03T14:58:00.941325Z ERROR limitador::storage::redis::redis_cached: Partition to Redis detected!
Current: hits: 6912, mean: 0.144ms, p50: 0.062ms, p90: 0.074ms, p99: 0.092ms, p999: 0.137ms, max: 352.255ms
Current: hits: 15414, mean: 0.064ms, p50: 0.062ms, p90: 0.074ms, p99: 0.091ms, p999: 0.107ms, max: 0.363ms
2024-04-03T14:58:02.941271Z ERROR limitador::storage::redis::redis_cached: Partition to Redis detected!

[eguzki]

you should rebase your branch to get some fixes in redis_cached

[didierofrivia]

this fn could be named differently, a more explicit name would be batch_update_counters... will leave it to the review

[alexsnaps]

This looks good to me… other that rename of cached_counters into cacher that I don't understand (like I don't understand the binding's name… literally).
Now, I also think the changes to keys.rs of removing the namespace: &str from the method signature are… unfortunate. As fundamentally I think we shouldn't resend the keys back, as we could "just" send the value & ttl for each ordered as we passed them in. This would avoid that change, but mostly would save bandwidth… We can iterate over these changes tho. Again, in the vein of avoiding doing work we can avoid doing.

[adam-cattermole]

I think these changes look good to me, tested it out with the driver and can see the writes in redis