Company
Date Published
Author
Dimitrios Liappis
Word count
563
Language
English
Hacker News points
None

Summary

Elasticsearch 6.7 and 7.0 introduced cross-cluster replication (CCR), a feature designed to replicate indices across clusters for purposes such as disaster recovery and geographically distributed deployments. To ensure the performance, resilience, and stability of CCR, Elasticsearch expanded its benchmarking tool, Rally, to communicate with multiple clusters and collect metrics from new CCR-related API endpoints. Benchmarking involved creating a topology with a leader cluster in Europe and a follower cluster in North America, ensuring a network latency of at least 100ms. The tests included various load scenarios to ensure followers could keep up, optimized remote recovery processes, and evaluated the impact of network compression on recovery times. Stability was tested over 10 days, simulating real-world conditions with index lifecycle management and random restarts of clusters. An accessible recipe using Docker was also developed to allow users to benchmark CCR in their local environments.