distributed-counter

Sharding distributes the write load across multiple physical sub-counters, while Conflict-free Replicated Data Types (CRDTs) ensure data integrity without lost updates. This combination prevents hot-shard p99 spikes during high write contention.

HyperLogLog estimates unique counts with constant memory usage by providing approximate counting for unique-visitor metrics. It trades exact accuracy for high scalability when you can tolerate some error in distributed counting.

Use CRDTs instead of simple sharding when you need conflict-free updates across distributed nodes without coordination. CRDTs ensure data integrity automatically, whereas simple sharding alone cannot prevent lost updates during network partitions.

Implement sharded counters to distribute the write load across multiple physical sub-counters. This approach handles millions of writes per second for view counts and ad impressions without creating hot-shard p99 bottlenecks.

Yes, the distributed counter approach works for rate-limit counters and vote tallies by using sharding and CRDTs to handle high write contention. It ensures accurate counting for vote tallies and reliable throttling for rate-limit counters.

The main limitation of HyperLogLog for distributed counting is that it provides approximate counts rather than exact numbers. You should not use it when precise data integrity is required, as it trades accuracy for constant memory usage.