I’m configuring a multi-node clustered setup of Phabricator with hosts in both North America and Western Europe. The database and repository clustering have been successfully established, although any cluster operations (e.g., repository clones via SSH and viewing repositories in diffusion via HTTP) which require coordination from the EU node consistently exhibit 5-10 seconds of latency. Due to the operational implications of this performance I have some questions:
Is it possible to de-prioritize requests routed to specific hosts in a cluster (i.e., to assign a given priority)? Currently these operations do not complete until there is acknowledgement from all cluster members which is very slow.
Cluster hosts were set by running
bin/config set cluster.addresses [‘list’, ‘of’, ‘addresses’] from the phabricator CLI on all hosts in the cluster. Are there known possible common configuration settings which could contribute such observed latent connections?