Credit: Hisense
Directory to save the logging info
。关于这个话题，pg电子官网提供了深入分析
Кирилл Дмитриев. Фото: Reuters，详情可参考手游
We often have new self-managed users asking us to provide recommendations around orchestration and how to scale to dozens, if not hundreds, of nodes. While technologies such as Kubernetes have made the deployment of multiple instances of stateless applications relatively simple, this pattern should, in nearly all cases, not be required for ClickHouse. Unlike other databases, which may be restricted to a machine size due to inherent limits, e.g., JVM heap size, ClickHouse was designed from the ground up to utilize the full resources of a machine. We commonly find successful deployments with ClickHouse deployed on servers with hundreds of cores, terabytes of RAM, and petabytes of disk space. Most analytical queries have a sort, filter, and aggregation stage. Each of these can be parallelized independently and will, by default, use as many threads as cores, thus utilizing the full machine resources for a query.