Improve Scalability Of Metrics Services

This paper is devoted to the problem of limited service scalability in a distributed faulttolerant system, which was caused by cross-partition queries to the database storing the state of service shards. System scalability and performance became significantly limited under the increased load on the database caused by competition for resources between shards. To circumvent this limitation, a new protocol based on the Event-carried state transfer approach was developed and implemented. The protocol allows service shards to interact in an asynchronous environment through message passing, avoiding direct requests to the state of other shards. This solution ensures exclusive ownership by each shard of its partitions, which reduces competition for resources and improves overall system performance. Special attention is paid to the protocol design and complexity of communication between shards under asynchronous data processing model. Index terms: distributed systems, scalability, consistency, Event-carried state transfer, Cross-Device, attributions, asynchrony.