Skip to content
OpenEverest Documentation

Get enterprise-grade support and services for OpenEverest from certified partners, or join community.

Get support

What’s new in OpenEverest 2.0.0 Developer Preview 2

Developer Preview — not for production

OpenEverest 2.0.0 Developer Preview 2 is an early-look release and is not feature-complete. It is intended for testing and feedback only.

There is no supported upgrade path from Developer Preview 1, and none from v1. Install Developer Preview 2 on a fresh cluster. v1 and v2 have fundamentally different architectures and data models, and running them side-by-side in the same cluster is not supported.

This release contains breaking changes to every CRD, to the HTTP API, and to the provider-runtime SDK. Manifests, providers and plugins written against Developer Preview 1 will not work and must be updated — see Breaking changes for the full mapping. No migration path is provided at this stage of the preview. This is deliberate, so that the API can be corrected before General Availability.

New to OpenEverest? Get started with our Quickstart Guide.

Read the blog post and watch the video introducing the v2 architecture.

🌟 Release highlights

everestctl becomes a real day-2 CLI for v2

Developer Preview 1 shipped v2 with a UI and an API, but no command-line story. Developer Preview 2 closes that gap: the whole v2 resource model is now drivable from everestctl, with consistent --namespace/-n, --wait, and JSON output across commands.

everestctl auth login --server https://everest.example.com
everestctl provider list
everestctl backup-class list
everestctl instance create --preset small-mongodb -n my-namespace --wait
everestctl instance status my-db -n my-namespace --watch
everestctl backup create --instance my-db --storage s3-backups -n my-namespace
everestctl restore create --instance my-db --backup my-db-20260812 -n my-namespace
everestctl backup-storage create --name s3-backups -n my-namespace
everestctl instance delete my-db -n my-namespace

New command groups: auth, instance, backup, restore, backup-storage (alias bs), backup-class (alias bc), provider (alias prov).

everestctl changes were contributed by @VijetaPriya47 through LFX Mentorship program 2026 Term 2.

Note

everestctl is not shipped as a binary with this release. everestctl install and everestctl uninstall still target v1 and cannot install v2, so publishing the binary would be misleading. To try the commands above against an existing v2 installation, build it from source:

git clone --branch v2.0.0-dev.2 https://github.com/openeverest/openeverest.git
cd openeverest
make build-cli   # produces bin/everestctl

Installation remains Helm-only for this preview.

Point-in-time recovery, end to end

PITR is now a first-class part of the v2 model rather than a modifier on a backup reference.

  • Recovery window. Instance.status.backup.storages[].pitr reports earliestRestorableTime, latestRestorableTime, state (Available/Unavailable), plus reason and message. The window is authoritative: providers truncate forward at a discontinuity and under-report rather than advertise a point that cannot be restored.
  • Point-in-time as a data source. dataSource.type now discriminates between Backup and PointInTime. A point-in-time recovery no longer has to name a Backup CR it has nothing to do with.
  • Restore into a new database. Because dataSource is the same shared type on Instance.spec.dataSource and Restore.spec.dataSource, you can seed a brand-new instance from a point in time on another instance’s stream via pointInTime.source.instanceRef.

Providers opt in by implementing the new controller.InstanceBackupStatusReporter interface; the runtime writes the aggregated per-storage window to instance.status.backup after every successful Sync.

Instance presets

InstancePreset is a new namespaced CRD that captures a ready-to-use instance shape — provider, version, topology, component sizing and storage class. Users pick a preset instead of filling in a form; power users still get the full parameter surface.

  • New API: GET /v1/clusters/{cluster}/instance-presets, .../{name}, and .../{name}/resolve (which returns the fully materialised Instance spec a preset would produce, including the resolved default storage class).
  • New CLI: everestctl instance create --preset <name>.

Secrets, ConfigMaps, and schema-driven credentials

Providers can now declare, in their provider definition, the secrets and config maps they expect (spec.secrets, spec.configMaps on Provider), each with a parameters schema. The API server exposes managed CRUD for those objects and validates payloads against the declared schema before they reach Kubernetes:

  • POST|GET /v1/clusters/{cluster}/namespaces/{namespace}/secrets and GET|DELETE .../secrets/{name}
  • POST|GET /v1/clusters/{cluster}/namespaces/{namespace}/config-maps and GET|DELETE .../config-maps/{name}

The provider-runtime gained matching validation webhooks (secret_validation.go, config_map_validation.go).

Auditable event stream with replay

GET /v1/events no longer silently drops events across a reconnect. Every event now carries a monotonic sequence and an epoch identifying the publishing process; since is "<epoch>:<seq>", and an aged-out cursor or epoch mismatch returns 410 Gone, which tells the client to resync rather than quietly lose data. Events also carry the acting user, laying the groundwork for auditing.

Breaking changes for provider and plugin developers

The API surface was normalised in four sweeps. All of them are breaking, and all of them are intentional at preview stage.

1. Cross-resource references are uniform. Every reference is now an object ({name: ...}) drawn from api/common/v1alpha1 — ObjectRef, SecretRef, TypedObjectRef — instead of a bare string or a corev1.LocalObjectReference.

Resource Before After
Instance.spec provider: <string> providerRef.name
Instance.spec.backup classRef (custom type) classRef (ObjectRef)
Instance.spec.backup.storages[] name + storageRef + main storageRef.name only (key), main inferred from pitr
Instance.status connectionSecretRef (LocalObjectReference) connectionSecretRef (*SecretRef)
Instance.status.components[] pods podRefs
Backup.spec instanceName, backupClassName, storageName instanceRef, classRef, storageRef
Backup.status / Restore.status jobName, operatorBackupRef (TypedLocalObjectReference) jobRef, operatorBackupRef (*TypedObjectRef)
Restore.spec instanceName instanceRef
BackupStorage.spec.s3 credentialsSecretName credentialsSecretRef.name
MonitoringConfig.spec.pmm credentialsSecretName credentialsSecretRef.name

2. config / customSpec / global are all parameters. One name for provider-defined free-form configuration everywhere, and one envelope (common.ParametersSchema, i.e. parametersSchema.openAPIV3Schema) for declaring its schema.

Before After
Instance.spec.global Instance.spec.parameters
Instance.spec.topology.config Instance.spec.topology.parameters
Instance.spec.components[].config Instance.spec.components[].parameters
Backup.spec.config / Restore.spec.config .spec.parameters
Provider globalConfigSchema / customSpecSchema / configSchema parametersSchema
Provider secrets[].openAPIV3Schema secrets[].parametersSchema
BackupClass config / restoreConfig / pitrConfigSchema parametersSchema / restoreParametersSchema / pitrParametersSchema
SDK Context.DecodeConfig() etc. DecodeParameters(), DecodeTopologyParameters(), DecodeComponentParameters()

ParametersSchema.Validate rejects a non-nil payload when no schema is declared, so components whose parameters are actually used must declare parametersSchema in definition/provider.yaml. Secret and config map definitions are the exception: an omitted parametersSchema there skips validation entirely rather than rejecting the payload.

3. dataSource is a discriminated union.

# before
dataSource:
  backupName: my-backup
  pitr:
    type: date            # or "latest"
    date: "2026-08-12T10:00:00Z"

# after
dataSource:
  type: Backup            # or PointInTime
  backup:
    backupRef:
      name: my-backup
# ...or...
dataSource:
  type: PointInTime
  pointInTime:
    source:
      instanceRef: {name: source-db}   # optional; defaults to the target Instance
      storageRef:  {name: s3-backups}  # required, must have pitr.enabled=true
    recoveryTarget: date               # or "latest"
    date: "2026-08-12T10:00:00Z"       # RFC 3339 with an explicit UTC offset

PITRType is renamed RecoveryTarget. The flat status.backup.storages[].latestRestorableTime is replaced by the nested pitr block described above.

4. BackupClass job specs are grouped. job and restoreJob become job.backup and job.restore, mirroring the discriminated-union shape used by the other execution modes.

Plugin API group renamed. plugin.openeverest.io/v1alpha1 is now extensions.openeverest.io/v1alpha1, and the PluginInstallation kind is now InstalledExtension. Plugin.spec.kubePermissions was removed — plugins ship their own RBAC in their Helm chart, and per-namespace plugin scoping was dropped.

Typed metadata in generated clients. CRD metadata is no longer an untyped map in the OpenAPI spec: a shared ObjectMeta schema is emitted and $ref‘d by every kind, with x-go-type mapping the Go models onto metav1.ObjectMeta. Go and TypeScript API consumers get typed metadata and lose the string-map helpers.

Provider upgrade catalog. Provider.spec.release carries the provider’s own version and minUpgradableFrom; ComponentVersion gained deprecated and removedInVersion to give users a deprecation runway.

📝 Changes

Added

  • everestctl gained a full v2 command surface — the whole v2 resource model is now drivable from the CLI, with consistent --namespace/-n, --wait and JSON output:
Group Subcommands PRs
auth login, logout #2372, #2443
instance create, list, status, delete #2480, #2563, #2519, #2762
backup create, list, delete #2599, #2597, #2863
restore create, list, delete #2656, #2648, #2847
backup-storage (bs) create, list, delete #2671, #2573, #2765
backup-class (bc) list #2565
provider (prov) list #2564

Cross-cutting flags: instance create --preset (#2518) and --wait (#2600), instance status --watch/-w (#2519), instance print columns (#2594).

  • InstancePreset CRD, resolution API and UI support, with default storage class resolution (#2407, #2454).
  • Point-in-time recovery: validation and restore gating, per-storage recovery-window status, point-in-time data sources, and restore into a new database (#2562, #2848).
  • controller.InstanceBackupStatusReporter in provider-runtime, for providers to publish per-storage backup observability data (#2848).
  • Secret management API with provider-declared schemas and runtime validation (#2534).
  • ConfigMap API with provider-declared schemas and runtime validation (#2541).
  • API auth token management with refresh tokens, and matching UI integration (#2361, #2362).
  • Audit events carrying the acting user (#2415).
  • Replay buffer on GET /v1/events so reconnecting subscribers do not silently lose events (#2591). (by @onkar717)
  • Provider upgrade catalog: spec.release.version, spec.release.minUpgradableFrom, and deprecated / removedInVersion on component versions (#2604).
  • affinity on Instance.spec.components[] (#2364) and service / loadBalancerService configuration on components.
  • Plugin hub in the v2 UI, with plugin-declared permissions (#2367, #2424).
  • Tiles view for creating instances in the UI (#2453).
  • Backup/Restore reference validation at the API layer on create (#2667). (by @Adii-45)
  • Instance reconcile-request helper in provider-runtime (#2515).
  • Finalizer on BackupStorage plus end-to-end tests (#2327, #2353).
  • RBAC fuzz tests (#2581), Dependabot coverage for v2 (#2673), CodeQL scanning of release-2.0 and workflow files (#2753), and SBOM attestation for release artifacts (#2858).
  • Self-serve issue triage and /assign process (#2764, #2796).

Changed & Improved

  • Breaking API changes. Every CRD, the HTTP API and the provider-runtime SDK changed shape. Full before/after mapping in Breaking changes:

  • Cross-resource references unified onto common.ObjectRef / SecretRef / TypedObjectRef (#2570)

  • config / customSpec / global renamed to parameters; schemas wrapped in common.ParametersSchema, including secret definitions on Provider.spec.secrets[] (#2583, #2616)
  • dataSource restructured into a Backup / PointInTime discriminated union; PITRType → RecoveryTarget; flat latestRestorableTime → nested pitr status (#2848)
  • BackupClass job specs grouped as job.backup / job.restore (#2547)
  • Plugin API group renamed to extensions.openeverest.io; PluginInstallation → InstalledExtension; spec.kubePermissions removed (#2367, #2424)

  • CRD metadata typed as a shared ObjectMeta schema in the OpenAPI spec and generated clients (#2652)

  • Breaking: POST /v1/auth/revoke is now unauthenticated (#2448).

  • Shared HTTP-poll classification extracted into pkg/cli/wait, so every --wait/--watch command behaves the same on transient errors (#2808). (by @VijetaPriya47)
  • UI upgraded from MUI 5 to MUI 7 (#2548, #2550, #2569).
  • Circular dependency removed from the UI and a strict ESLint rule added to prevent regressions (#2417). (by @StepanovPlaton)
  • golangci-lint upgraded to v2, and CI now fails on unformatted Go code (#2535, #2507).
  • GitHub Actions pinned to full-length commit SHAs and workflow token permissions scoped down (#2751, #2752).
  • Tilt development environment reworked for core development, with InstancePreset, plugin-hub and PXC provider support (#2483, #2439, #2508, #2346).

Fixed

  • provider-runtime reported /readyz as ready before the controller manager and its cache-backed client were usable (#2611). (by @Anuragkumar-687)
  • Provider status messages were silently dropped instead of surfacing on the Instance (#2366). (by @AdityaPimpalkar)
  • Nil connection reference caused a panic when an instance had no connection secret yet (#2590).
  • Init containers were counted incorrectly when computing instance resource totals (#2719). (by @Jay2006sawant)
  • MonitoringConfig finalizers and secrets were not cleaned up reliably on delete (#2408). (by @alokkumardalei-wq)
  • The server did not handle SIGTERM, so pods were killed instead of shutting down gracefully (#2820). (by @Harkirat1309)
  • OIDC well-known configuration fetch had no HTTP timeout (#2725). (by @Harsh63870)
  • RBAC informer start errors were swallowed, and the ConfigMap adapter’s Kubernetes calls had no timeout (#2830 by @alloutflo, #2822 by @Harkirat1309).
  • Nil pointer panic when the JWT private key PEM was invalid (#2851). (by @Atishyy27)
  • Token blocklist update errors were not propagated (#2769). (by @Anuragkumar-687)
  • The arm64 server image shipped amd64 binaries (#2606).
  • Helm chart dependency builds failed without a configured registry client (#2607).
  • Editing a backup storage location and the empty state of scheduled backups both errored in the UI (#2320, #2319).
  • Database credentials cache was not scoped by namespace, and the RBAC route guard was not reactive (#2173). (by @ashnaaseth2325-oss)
  • The upgrade dialog reopened after being dismissed (#2228). (by @ashnaaseth2325-oss)
  • Multiline values overflowed the Database Summary preview (#2090). (by @StepanovPlaton)
  • External links in the UI were missing noreferrer, and the login page lost spacing between the intro text and the community links (#2857, #2902).
  • everestctl flag and message fixes, including -n as a shorthand for --namespace and the namesapce typo (#2856, #2795 by @Bhupesh-081, #2444 by @sahilforkshere, #2252 by @abhisheksainimitawa, #2196 by @ayushHardeniya).
  • Documentation typos, broken links, stale extension CLI examples, the EKS z1d EBS volume limit, and the providerRef.name plugin filter reference (#2458 by @Pranav-IIITM, #2460 by @Pranav-IIITM, #2622 by @false200, #2625 by @Jay2006sawant, #2778 by @Bhupesh-081, #2818 by @Adii-45, #2624 by @Jay2006sawant, #2840 by @LarytheLord).

🚀 Installing OpenEverest 2.0.0 Developer Preview 2

OpenEverest v2 is installed via Helm. You will need a running Kubernetes cluster and helm installed. everestctl install still targets v1 and must not be used for v2.

Install the OpenEverest core:

helm repo add openeverest https://openeverest.github.io/helm-charts/
helm repo update
helm install everest-core openeverest/openeverest \
  --devel \
  --version "2.0.0-dev.2" \
  --namespace everest-system \
  --create-namespace

Retrieve the admin password:

kubectl get secret everest-accounts -n everest-system \
  -o jsonpath='{.data.users\.yaml}' | base64 --decode | yq '.admin.passwordHash'

Access the UI:

kubectl port-forward svc/everest 8080:8080 -n everest-system

Providers are released independently of the core. Install the ones you need from their own repositories, and pick a version built against 2.0.0-dev.2 — provider releases published before this one target the previous API and will not reconcile.

v1 Support and End-of-Life Timeline

OpenEverest v1 continues to be supported while v2 matures. The transition timeline is as follows:

Stage Description
Now v2 Developer Preview — testing and feedback.
v2 General Availability Expected in a few months.
GA + 3 Months v1 enters Maintenance Mode. Security patches and critical bug fixes only; no new features.
GA + 12 Months v1 reaches End of Life. No further releases.

Get involved

Try the Developer Preview, build a provider against the new API, and share your feedback. The project is open to contributions — provider plugins, generic plugins, and core improvements are all welcome.

Thanks to our contributors

This release wouldn’t have been possible without our community. Huge thanks to everyone who contributed code, fixes, and improvements to Developer Preview 2:

@abhisheksainimitawa, @Adii-45, @AdityaPimpalkar, @alloutflo, @alokkumardalei-wq, @Anuragkumar-687, @ashnaaseth2325-oss, @Atishyy27, @ayushHardeniya, @Bhupesh-081, @cansayin, @false200, @Harkirat1309, @Harsh63870, @Jay2006sawant, @LarytheLord, @onkar717, @Pranav-IIITM, @sahilforkshere, @StepanovPlaton, @VijetaPriya47

Full Changelog: https://github.com/openeverest/openeverest/compare/v2.0.0-dev.1…v2.0.0-dev.2