Changes are listed newest first by documentation revision. The API remains v1. Check service capabilities for the operations available to your application.
2026-11-06.2#
- Database grants now treat underscores in schema names literally, keeping access scoped to the selected database.
- Database-user renames require a running SQL service so removal of the old login can be queued safely.
- SSH account operation polling now includes shared gateway execution status.
2026-11-06.1#
- Public slice-token CRUD for databases, database users, SFTP accounts and container SSH access accounts.
- Explicit
databases:manage,databases:delete,access:manageandaccess:deletegrants; existing tokens and default grants are unchanged. - Scoped generated-password retrieval and idempotent rotation. Passwords and secret references are excluded from inventory and operation receipts.
- Managed database service detail, label editing and decommissioning. SQL schema names are immutable and required application schemas remain protected.
- Account/database operation polling distinguishes configuration receipt status from runtime execution, including unavailable or failed work.
- See database and access management for the request and polling walkthrough.
2026-10-21.8#
- Portal can edit permissions on existing active slice and service tokens, including
deployment:configure, without rotating credentials. - Added explicit confirmation for additional grants, stale-edit protection and tenant-scoped audit records under the existing workspace credential security policy.
2026-10-21.7#
- Added service-scoped deployment settings read/write, covering image, API upload and GitHub sources, build commands, repository credentials and release webhooks.
- Added opt-in
deployment:configurepermissions to service and slice tokens; enabling automatic deployment also requiresdeploy. - Added explicit configured-source deployment with revision checks, idempotent retries and sanitized progress polling.
- Added complete GitHub setup, credential rotation, webhook, deployment and source-switch examples to the deployment guide.
2026-10-21.6#
- Portal custom applications can switch between container image, API upload and GitHub deployment sources without recreating the service or deploying on save.
- GitHub builds support repository Dockerfiles as well as Node.js commands. Existing runtime configuration and permissions remain in place.
- Switching sources disables automatic GitHub deployment and rejects uploads from an inactive API source. Existing service deployment tokens can also target custom Node.js runtimes selected for API upload.
2026-10-21.5#
- External data imports now use the backup restore service lifecycle and normal database TCP connection. No socket-only database setup is required.
- Validation requires explicit stopping consent and reports
preparingprogress. Application resume is opt-in after verified import, withresumingprogress. - Upload validation, scoped SQL permissions, explicit replacement approval and failure protection remain enforced. Destination agents require the updated restore lifecycle capability.
2026-10-21.4#
- Added slice-scoped managed service discovery, creation and observed status. MariaDB is the first available type; the service catalogue can support more managed types later.
- MariaDB creation registers a private managed service, persistent storage, root credential and logical database. Deployment remains a separate action.
- Portal offers Add managed service on Services slices, including review before deployment and the ability to deploy a configured service later.
2026-10-21.3#
- Data imports support opt-in
mode:"replace"validation and target-bound clearing consent. Empty destinations remain the default requirement. - Portal offers Clear and import, a destination preview and typed target ID.
- Requires updated destination agents advertising
private_data_import_replace_v1. Interrupted replacements remain protected for manual recovery and never clear again automatically. Keep a separate backup before approving replacement.
2026-10-21.2#
- Renamed the upload/import workflow to Data imports.
Use
/v1/deploy/slices/{sliceId}/data-importsfor new integrations. All/backup-importsendpoints remain compatible aliases with the same operation IDs, uploaded data, reviews and retry keys at runtime. - Portal now offers Services → slice → Import data, using databases and attached writable storage without requiring a backup schedule or recovery point.
- Token permission strings, request/response formats and destination safeguards are unchanged. Deploy the API before updating clients to the new route name.
2026-10-21.1#
- Private backup uploads and imports:
resumable
.sql.gzand.zipuploads, separate validation and approval, optional automatic root detection, explicit folder selection when ambiguous, and durable progress/recovery guidance for empty isolated destinations. - Expanded OpenAPI capability, preview and verification schemas, permission requirements, response statuses and retry guidance.
- Individual backup restore: clarified preview versus start, overwrite confirmation and the distinction between restoring a retained artifact and importing a local file.
- Existing token grants do not change. Imports require explicit
backups:uploadandbackups:import; retained restores requirebackups:restore. Existing restore request formats remain compatible. No import feature flag is required; destination support and isolation checks still apply.
2026-10-17.4#
- Retained logs: fixed cleanup verification failures for long deletion periods. Purge and retention requests keep their existing format; no client changes are required.
2026-10-17.3#
- Clarified the guides and API reference to focus on customer actions, permissions and response handling. No API request changes are required.
2026-10-17.2#
- Job output: known secrets are redacted before response truncation.
Incomplete or unreadable job output is withheld, with
output_coverage:unavailable; clients should check coverage before using output. - Registry credential rotation: failed
validation preserves the working credential. Concurrent configuration changes
return
CONFIGURATION_REVIEW_STALEand require a fresh review before retrying. - Slice backups: documented the existing inventory, group, cleanup and restore recovery endpoints in OpenAPI, including permissions, revision checks and retry semantics. No permissions are granted automatically; existing tokens still need the documented grants.
2026-10-17.1#
- Service-scoped registry login metadata, set/rotate and removal endpoints, with permission, secret-handling and separate-deployment guidance.
- Clearer deployment failure diagnostics distinguish image-download failures from application startup and readiness failures.
- Safe public
REGISTRY_ACCESS_DENIEDandIMAGE_PULL_FAILEDdiagnostics.
2026-10-16.2#
- Replay-safe service-token issuance, scoped metadata list/revoke and recovery guidance.
- Stable same-image deployment handles while the exact image remains retained.
- Coordinated slice measurements report shared resource limits and indicate when samples are incomplete.
- Structured retry guidance for transient runtime admission failures.
- Ingress policy/application evidence, edge rejection headers and observed counters.
- Recognized password arguments and known secret values are redacted from newly retained logs.
2026-10-16.1 - Runtime operations and deployment fixes#
Added#
- Native container readiness: manifests accept
readiness: {"type":"docker"}for images with a HEALTHCHECK. Deployment requires fresh successful health evidence from the intended container image. MariaDB and Redis no longer need an HTTP health wrapper. - Private applications: slice service creation accepts
public_route:falseto provision an application without a public HTTP route or published host port. Services communicate through their slice network. - Maintenance jobs:
POST /v1/deploy/services/{serviceId}/jobsruns bounded commands in a temporary container using the deployed image, mounted data and network. Jobs support status polling, cancellation, time/resource limits and bounded output with secret redaction. - Lifecycle operations: service
/start,/stopand/restartendpoints operate on deployed containers while preserving storage. Start and restart require fresh readiness; first deployment still uses the deployment flow. - Resource samples: service
/metrics/samplescollects container CPU, memory, throttling and identity information. Slice/metricsaggregates recent member samples and reports incomplete coverage. Clients retain samples for history. - Capability discovery: service
/capabilitiesreports runtime support and token permissions./operationsand/operations/{id}expose asynchronous job, lifecycle and metrics results.
See Runtime operations and private database services for requests, limits, polling and cancellation behavior, and the OpenAPI reference for schemas.
Fixed#
- HTTP readiness now requires the exact expected status, defaulting to 200.
A 403 or 404 cannot complete readiness successfully. Use
expected_statusfor a different intended status in the supported 200-399 range. - Fixed deployment eligibility checks for newly provisioned services.
- New persistent volumes are initialized on first use. Missing existing storage requires recovery and is never silently replaced with an empty volume.
- Deployment failures expose safe storage reasons
STORAGE_VOLUME_UNAVAILABLEandSTORAGE_RECOVERY_REQUIRED, and the affected storage ID when available. - Build logs expose cursors, retention/truncation coverage and build progress.
Source validation identifies missing root files, including
.dockerignore; builder setup failures are distinguished from source archive problems. - Unstarted deployments can be cancelled in
awaiting_uploadorreadystate. Once start has been accepted, cancellation returnsDEPLOYMENT_ALREADY_STARTED. Polling examples now handle the terminalcancelledstate.
See Deploy a service for build logs, diagnostics and upload cancellation.
Client compatibility#
- Existing tokens gain no permissions automatically. Jobs require
jobs:run; lifecycle actions requirelifecycle:manage. Keepreadfor status polling. A slice token can delegate only permissions it already holds. - Check
/capabilitiesbefore using an operation. Contact support forRUNTIME_CAPABILITY_UNAVAILABLE;TOKEN_PERMISSION_DENIEDrequires a token with the appropriate permission. - Clients relying on redirects or other non-200 health responses must configure an explicit expected status or provide a health endpoint returning 200.
- Contact support if existing storage remains unavailable. Do not create an empty replacement for a volume that should contain data.
mount_pathselects the location of a volume inside your container.- For a retry of the same request, retain its idempotency key. A new deployment attempt after a terminal failure uses a new key.