Changes in 0.11.x

This edition describes the 0.11 release candidate, v0.11.0-rc.7. Use the upgrade guide for an existing deployment and the header selector for an installed 0.10 build.

Commands

Administrative commands use an entity followed by an action. Existing command spellings remain aliases with the same permissions, output and exit codes. Scripts can move to the new names independently of the deployment upgrade.

Operation	0.10 spelling	0.11 spelling
List nodes	authority list-nodes	authority nodes list
List resources	authority list-resources	authority resources list
Declare a manual resource	authority set-resource	authority resources set
Create a node token	authority create-token	authority tokens create
Bootstrap an admin identity	authority bootstrap-admin	authority identities bootstrap --role admin
Verify audit integrity	authority verify-audit	authority audit verify
List client resources	ls	resources list
Generate SSH configuration	ssh-config	ssh config show

The CLI reference lists primary paths. Legacy alias tables retain the old spellings. connect, join, login, up, down, status, doctor and the short SSH invocation remain available.

Resident reload and diagnostics

contexts use PATH switches a running resident without replacing its process. gdsgate reload re-reads its current file. Managed addresses are bound to the cluster; previous sessions keep their original connection. The loaded file and reload result appear in status, including JSON. Bound OS settings have explicit restart requirements in the reload reference.

doctor --role selects the component on hosts running several roles. JSON reports include the role and source file. Unrelated local state does not change the role.

Server configuration reload

gdsgate reload --role authority|proxy|connector|agent-host|all applies supported changes to a running server. Files are validated before application; a refusal retains the live configuration. Connector declarations update on the existing tunnel and open sessions retain their backend. Linux service reload uses the same acknowledged command. The reload reference lists supported settings, restart requirements, admission checks and platform commands.

Windows server installation sets machine ownership and service read access on the configuration. setup change preserves these permissions for later reloads.

Configuration selection and setup

connect selects the client configuration it writes. Use an existing file for another cluster:

gdsgate contexts use ~/.gdsgate/clusters/production/config.toml
gdsgate contexts show
gdsgate login
gdsgate resources list

Access and administrative RPCs use the same selected cluster. Installed server roles use their installed configuration records. An explicit --config overrides the selection for that invocation. A missing or unreadable selection stops the command. See configuration selection.

A running resident reloads the selected context. Run login when the target cluster has no saved identity. Generated commands saved manually retain their original configuration path; regenerate those copies after switching.

setup prepare supplies platform paths and ports, preserves explicit overrides, and resumes a partially completed installation. CLI, TUI and saved bundles use the same installation receipt. --start installs and starts native services; containers run the same binary under their supervisor. A retry preserves a running service. See setup and services.

HA and node enrollment

A new PostgreSQL Authority starts with HA and keyring synchronization enabled. It can begin as one member. Add Authority, Proxy or paired roles with verified join; the protected bootstrap delivers the shared settings and keyrings. Each role retains its own identity and local state. Existing standalone stores require an explicit migration and are not reinitialized by repeating setup.

Registration addresses, trust anchors and names come from the verified role bundle. Unknown trust requires terminal confirmation or an approved fingerprint. Non-interactive installation requires explicit trust. Token credentials are sent after target and cluster verification.

Private membership updates the Proxy's Authority pools. Adding another Authority works through a surviving Proxy and ready Authority. Root rotation still requires all retained acknowledgements. The external load balancer's backend list remains an operator responsibility.

Choose direct Authority failover or an Authority load balancer. Both use the same component procedures and platform tabs.

Catalog and names

New installations publish permitted Connector backends automatically. Authority assigns the node identity, namespace and publication limits. User access still requires policy. Existing clusters keep their shared policy; existing Connector identities keep their assigned publication mode.

Automatic resources have qualified IDs and names. A short name is offered when the complete catalog makes it unambiguous. Manual logical resources can have several explicitly assigned Connector nodes. Administrative removal suppresses automatic republishing until an authorized restore.

See publish and migrate resources for identity, collisions, SSH onward defaults, bindings and recovery.

Kubernetes

One setup command configures API access, discovery, or both. It writes the existing backend/source declarations and preserves unrelated resources and explicit settings. Discovery-only configuration publishes discovered targets without exposing the cluster API as a resource.

Inside a Pod, setup uses the ServiceAccount mount. Outside the cluster, select the kubeconfig context explicitly. Keep the API and discovery credentials separate when their permissions differ. See Kubernetes setup.

Upgrade an existing deployment

Take a paired database/key backup and rehearse restoration before changing the fleet. Keep existing configurations, node identities and resource IDs. Use the complete rc.10/rc.13-to-0.11 procedure for preflight, the six store migrations, binary replacement, service unit regeneration, Authority/Proxy/Connector/Client startup and paired recovery. service install reloads the service manager but does not restart a running process, even with --start; restart separately to load the new executable. This documentation does not promise mixed-version HA or downgrade after new state has been written.

The CLI aliases support existing scripts. Automatic publication is a separate, explicit decision for an existing cluster. Verify policy denial, serving state, session audit and finalized recordings after the upgrade. For native SSH, test both the generated name and the resident address.