2.5.0 - suggestions and two-phase upload

[2.5.0] - 2021-03-05

Issues and suggestions

• Show an issue when ConfiguringRoles state gets stuck for more than 5s.

• Show a suggestion to disable broken instances.

• Show a suggestion to force reapply clusterwide configuration.

Configuration options

• New cartridge.cfg option webui_enabled (default: true). Otherwise, HTTP server remains operable (and GraphQL too), but serves user-defined roles API only.

• New cartridge.cfg option http_host (default: 0.0.0.0) which allows to specify the bind address of the HTTP server.

Miscellaneous enhancements

• Allow observing cluster from an unconfigured instance WebUI.

• Introduce a new graphql parser (libgraphqlparser instead of lulpeg). It conforms to the newer GraphQL specification and provides better error messages. The "null" literal is now supported. But some other GraphQL expressions are considered invalid (e.g. empty subselection).

• Properly handle etcd index updates while polling stateful failover updates. The problem affected long-running clusters and resulted in flooding logs with the "Etcd cluster id mismatch" warnings.

• Refactor two-phase commit (patch_clusterwide) logics: don't use hardcoded timeout for the prepare stage, move upload to a separate stage.

• Eliminate GraphQL error "No value provided for non-null ReplicaStatus" when a replica is removed from the box.space._cluster.

• Allow specifying server zone in join_server API.

• Don't make formatting ugly during config upload.

• Hide the bootstrap button when it's not necessary (e.g. before the cluster is bootstrapped, and in vshardless cluster too).

• Properly display an error if changing server zone fails.

• Fix disabled instances style.