Replicove v0.3.0-alpha.2

Replicove v0.3.0-alpha.2 fixes cleanup of copied PVCs in replica-owned guest namespaces. A guest-created Pod can keep a PVC protected after its deletion is requested; cleanup now also starts deletion of the owned namespace so those consumers can drain. Existing borrowed namespaces and their unrelated resources remain protected. Custom-resource and controller cleanup keeps its dependency order, and all deletions still require the recorded object identity and ownership markers.

helm upgrade --install replicove oci://ghcr.io/nimeshbuilds/charts/replicove \
  --version 0.3.0-alpha.2 \
  --namespace replicove-system --create-namespace --wait --timeout 3m

Prebuilt CLI archives for Linux and macOS on amd64 and arm64 are attached below; each includes the executable, license and a quickstart for this release. Download the matching platform archive and verify it with SHA256SUMS. Go is only needed when building from the tagged source. The Helm chart and native CRD/operator manifests are separate assets; the operator image is published to GHCR.

Use matching CLI, chart, image and CRDs from this release. The expanded lifecycle test covers selected dependencies, Secret Snapshot/Follow behavior, EmptyVolumes with independent guest writes, drift repair, pruning and complete storage cleanup. The release links the exact source commit and published-artifact verification run below. Use a source build for an unreleased revision.

• Reusable test recipes: create a fresh managed replica or mirror, wait for readiness, obtain a bounded guest session, execute an explicit local command and verify cleanup. Metadata/provenance JSON and JUnit reports distinguish command failure from cleanup failure. Recipes are local documents; there is no dedicated GitHub Action product integration.
• Preflight and plan evidence: inspect caller/API checks, captured source identities, dependencies, transformations and observed omissions without exposing resource payloads. A report is not an atomic source snapshot or an importable historical replay.
• Destination pools and admission: render separate operator/state/destination installations with quotas, choose among explicitly listed destinations, and reserve capacity durably in the operator. Queued requests retain their original TTL.
• PostgreSQL 17 copies: a separate administrator database grant authorizes read-only source access, approved column masks and explicit table equality filters. A consistent logical dump is restored in isolation, sanitized and validated, then copied into fresh final storage before applications or guest access start. Undeclared data can remain sensitive; no general anonymization or automatic tenant extraction is promised.
• Six bounded chaos fault types: PodDelete, ScaleZero, NetworkIsolation, CPUStress, MemoryStress and CustomJob operate only on qualified owned test targets, with grant limits, durable rollback and verified cleanup. Network/Job faults require enforced host policies and reject conflicting allow rules. These are shared-worker workload faults, not privileged host/node faults.
• Optional agent and UI interfaces: namespace-scoped stdio MCP uses the caller's Kubernetes identity, read-only by default with explicit mutation opt-in. The dashboard serves read-only metadata over loopback with a private session URL. There is no remote MCP certificate/CA service or shared dashboard deployment.

Mirrors, databases and chaos use the same operator image and are disabled by default. They can be enabled later with complete CRD, image, values and RBAC updates. Existing selected configuration/operator/Helm replication, access, persistent control planes and CSI mirror resets remain available.

Upgrade all six CRDs before Helm: ClusterReplica, ReplicaGrant, ReplicaAccess, ReplicaMirror, ReplicaMirrorRun and ReplicaExperiment. Helm does not upgrade its crds/ directory automatically. Preserve the immutable encryption key, protected state, namespaces and reviewed values; replace a completed bootstrap Job only when its immutable template changes. An old saved image digest overrides a new tag. Keep affected modules, operator and storage controllers available until all finalizers finish.

PostgreSQL copies are limited to new managed targets and cannot be combined with CSI mirrors, existing targets or in-place refresh. The pinned vCluster permits one managed runtime per host namespace; use separate pool destinations for concurrency. Managed mirror resets interrupt availability. CSI captures remain per-volume crash-consistent, with no atomic application or multi-volume guarantee.

Release artifacts include Linux amd64/arm64 operator images, macOS/Linux amd64/arm64 CLI archives, digest-pinned OCI chart/native manifests and checksums. OCI build metadata is not a separate signed release attestation. The release workflow creates the version tag only after the exact main revision and published-artifact verification pass. The release links its source commit and artifact-verification run.

Read the feature map, installation and optional modules, security model and validation record. Cloud identity, broader data adapters, vCluster Platform, distribution certification and production support remain outside this experimental release.

Image digest: sha256:92072ef308880bc93aa429c5e10e0ddc3dce79b4f99cd3b0c2cee2e9c1e62775

Source commit: afa15cb1723ccb9d2a204b8aeafe88231e782d24

Published-artifact test run.

Downloads and tested walkthroughs

Use the CLI download and source-build guide for platform selection, checksum verification and installation. The ten executable scenario guides cover the current feature groups and explain prerequisites, expected results and cleanup.

All 16 released-artifact scenario variants passed in run 36054672172 at fixture revision b3e6ab392482c452bf7c26afc08e6f4301e4fc57. That revision also passed all 16 source CI jobs and strict documentation checks; the merged documentation has the same tested file contents. These are small Linux amd64/disposable kind-cluster checks, not cloud or production qualification.

The supplemental scenario evidence records individual outcomes, exact release/fixture identities, functional assertions, mirror retention/transition checks, JSON/JUnit results and cleanup. Its SHA-256 is c79b5e186745b8f4e25333c2c145e187ef26499c02fb79fce6b8d4e2e2bc3fac. The original binary/chart/manifest assets and their SHA256SUMS are unchanged.

Replicove v0.3.0-alpha.1

Replicove v0.3.0-alpha.1 adds reusable integration-test runs, scoped PostgreSQL 17 copies, bounded workload faults, better plan evidence and optional local interfaces. All 16 source CI jobs passed at 93bcfbc, including the three mirror upgrade paths and new PostgreSQL, chaos and test-runner suites. The validation record records that source evidence separately from exact-main and release-artifact verification.

helm upgrade --install replicove oci://ghcr.io/nimeshbuilds/charts/replicove \
  --version 0.3.0-alpha.1 \
  --namespace replicove-system --create-namespace --wait --timeout 3m

Use matching CLI, chart, image and CRDs from this release. Use a source build for an unreleased revision.

• Reusable test recipes: create a fresh managed replica or mirror, wait for readiness, obtain a bounded guest session, execute an explicit local command and verify cleanup. Metadata/provenance JSON and JUnit reports distinguish command failure from cleanup failure. Recipes are local documents; there is no dedicated GitHub Action product integration.
• Preflight and plan evidence: inspect caller/API checks, captured source identities, dependencies, transformations and observed omissions without exposing resource payloads. A report is not an atomic source snapshot or an importable historical replay.
• Destination pools and admission: render separate operator/state/destination installations with quotas, choose among explicitly listed destinations, and reserve capacity durably in the operator. Queued requests retain their original TTL.
• PostgreSQL 17 copies: a separate administrator database grant authorizes read-only source access, approved column masks and explicit table equality filters. A consistent logical dump is restored in isolation, sanitized and validated, then copied into fresh final storage before applications or guest access start. Undeclared data can remain sensitive; no general anonymization or automatic tenant extraction is promised.
• Six bounded chaos fault types: PodDelete, ScaleZero, NetworkIsolation, CPUStress, MemoryStress and CustomJob operate only on qualified owned test targets, with grant limits, durable rollback and verified cleanup. Network/Job faults require enforced host policies and reject conflicting allow rules. These are shared-worker workload faults, not privileged host/node faults.
• Optional agent and UI interfaces: namespace-scoped stdio MCP uses the caller's Kubernetes identity, read-only by default with explicit mutation opt-in. The dashboard serves read-only metadata over loopback with a private session URL. There is no remote MCP certificate/CA service or shared dashboard deployment.

Mirrors, databases and chaos use the same operator image and are disabled by default. They can be enabled later with complete CRD, image, values and RBAC updates. Existing selected configuration/operator/Helm replication, access, persistent control planes and CSI mirror resets remain available.

Upgrade all six CRDs before Helm: ClusterReplica, ReplicaGrant, ReplicaAccess, ReplicaMirror, ReplicaMirrorRun and ReplicaExperiment. Helm does not upgrade its crds/ directory automatically. Preserve the immutable encryption key, protected state, namespaces and reviewed values; replace a completed bootstrap Job only when its immutable template changes. An old saved image digest overrides a new tag. Keep affected modules, operator and storage controllers available until all finalizers finish.

PostgreSQL copies are limited to new managed targets and cannot be combined with CSI mirrors, existing targets or in-place refresh. The pinned vCluster permits one managed runtime per host namespace; use separate pool destinations for concurrency. Managed mirror resets interrupt availability. CSI captures remain per-volume crash-consistent, with no atomic application or multi-volume guarantee.

Release artifacts include Linux amd64/arm64 operator images, macOS/Linux amd64/arm64 CLI archives, digest-pinned OCI chart/native manifests and checksums. OCI build metadata is not a separate signed release attestation. The release workflow creates the version tag only after the exact main revision and published-artifact verification pass. The release links its source commit and artifact-verification run.

Read the feature map, installation and optional modules, security model and validation record. Cloud identity, broader data adapters, vCluster Platform, distribution certification and production support remain outside this experimental release.

Image digest: sha256:fa57593efc4fbfbfc1bf7539930fb3fa23799517b989aa69c27cfe0349edb817

Source commit: 003b59a9ecdc2729034fa21e5d2a3478686ef704

Published-artifact test run.

Replicove v0.2.0-alpha.2

Replicove v0.2.0-alpha.2 fixes mirror infrastructure being selected as application configuration during broad capture/refresh, and documents adding mirroring after installation. New upgrade tests preserve a running ordinary replica when enabling the module from either 0.1.0-alpha.1 or 0.2.0-alpha.1.

The optional mirror module creates writable copies of selected host workloads and CSI volume data, with manual or scheduled resets. The host remains the source of truth; resets discard guest changes within the owned scope.

helm upgrade --install replicove oci://ghcr.io/nimeshbuilds/charts/replicove \
  --version 0.2.0-alpha.2 \
  --namespace replicove-system --create-namespace --wait --timeout 3m

For data copies, follow the mirror quickstart to enable mirrors.enabled, qualify host NetworkPolicy/CSI support, and delegate exact source PVCs. The module uses the same operator image/chart, installs the pinned snapshot controller/APIs when absent, and reuses existing snapshot infrastructure.

• ReplicaMirror and ReplicaMirrorRun YAML APIs plus replicove mirror commands for creation, access, sync/reset, scheduling, test leases, retention and cleanup.
• Independent restored storage, encrypted snapshot identities, protection for active revisions, candidate readiness/isolation checks, and recovery from interrupted creation or partial restoration.
• Dedicated new vClusters or administrator-qualified existing runtimes with separate generation namespaces and namespaced guest access.
• Existing selected configuration/operator/Helm replication, secrets, plan approval, refresh, persistent control planes and ordinary replica lifecycle remain available.
• Stable local CLI listening socket across streaming tunnel reconnection, without replaying application requests.
• Generated API/CLI references, API-tested examples, a full mirror guide and strict documentation/link checks.
• Linux amd64/arm64 operator image, macOS/Linux amd64/arm64 CLI downloads, digest-pinned OCI chart/native manifests and SHA-256 checksums. OCI SBOM/provenance metadata is not a separate signed release attestation.

Upgrades must apply the release CRDs before the Helm upgrade; Helm does not upgrade its crds/ directory automatically. Preserve the immutable state key. Keep the operator and storage/snapshot controllers running until mirror finalizers finish cleanup.

Follow enable mirroring later for an existing installation, including saved values, explicit image selection, source RBAC, native/GitOps ownership and cleanup before disabling. The feature map separates all shipped capabilities from planned integrations. Both upgrade paths run against published artifacts before this release is created.

vCluster 0.37.1 allows one runtime per host namespace. Managed resets prepare recovery points, wait for test leases, then remove the old runtime before starting its replacement; expect an interruption and no automatic rollback. Existing-target mirrors can prepare separate generation namespaces within their registered runtime. An unrelated runtime blocks new managed installation with an explicit reason.

The exact main revision must pass the full CI suite. Published-artifact tests additionally install the anonymous OCI chart/image and exercise mirrors, real data resets, isolation, existing targets, access, expiry and cleanup before this GitHub prerelease is created. Check the linked test run and source commit below for evidence.

Experimental: CSI recovery points are per-volume crash-consistent, not atomic application/database or multi-volume snapshots. External services, IRSA/cloud identity, provider CSI certification, vCluster Platform, the proposed MCP identity service and dashboard remain outside this release. Read the compatibility limits and security model.

Image digest: sha256:c54f80e06d2c5c8b1d350d92e5c43f4892776ad6acc8b9bb55760ca48a291992

Source commit: ceddb562d10a376b3373623d9a798caa041d2b11

Published-artifact test run.

Replicove v0.2.0-alpha.1

Replicove v0.2.0-alpha.1 adds optional workload mirrors: writable copies of selected host workloads and CSI volume data, with manual or scheduled resets. The host remains the source of truth; resets discard guest changes within the owned scope.

helm upgrade --install replicove oci://ghcr.io/nimeshbuilds/charts/replicove \
  --version 0.2.0-alpha.1 \
  --namespace replicove-system --create-namespace --wait --timeout 3m

For data copies, follow the mirror quickstart to enable mirrors.enabled, qualify host NetworkPolicy/CSI support, and delegate exact source PVCs. The module uses the same operator image/chart, installs the pinned snapshot controller/APIs when absent, and reuses existing snapshot infrastructure.

• ReplicaMirror and ReplicaMirrorRun YAML APIs plus replicove mirror commands for creation, access, sync/reset, scheduling, test leases, retention and cleanup.
• Independent restored storage, encrypted snapshot identities, protection for active revisions, candidate readiness/isolation checks, and recovery from interrupted creation or partial restoration.
• Dedicated new vClusters or administrator-qualified existing runtimes with separate generation namespaces and namespaced guest access.
• Existing selected configuration/operator/Helm replication, secrets, plan approval, refresh, persistent control planes and ordinary replica lifecycle remain available.
• Stable local CLI listening socket across streaming tunnel reconnection, without replaying application requests.
• Generated API/CLI references, API-tested examples, a full mirror guide and strict documentation/link checks.
• Linux amd64/arm64 operator image, macOS/Linux amd64/arm64 CLI downloads, digest-pinned OCI chart/native manifests and SHA-256 checksums. OCI SBOM/provenance metadata is not a separate signed release attestation.

Upgrades must apply the release CRDs before the Helm upgrade; Helm does not upgrade its crds/ directory automatically. Preserve the immutable state key. Keep the operator and storage/snapshot controllers running until mirror finalizers finish cleanup.

vCluster 0.37.1 allows one runtime per host namespace. Managed resets prepare recovery points, wait for test leases, then remove the old runtime before starting its replacement; expect an interruption and no automatic rollback. Existing-target mirrors can prepare separate generation namespaces within their registered runtime. An unrelated runtime blocks new managed installation with an explicit reason.

The exact main revision must pass the full CI suite. Published-artifact tests additionally install the anonymous OCI chart/image and exercise mirrors, real data resets, isolation, existing targets, access, expiry and cleanup before this GitHub prerelease is created. Check the linked test run and source commit below for evidence.

Experimental: CSI recovery points are per-volume crash-consistent, not atomic application/database or multi-volume snapshots. External services, IRSA/cloud identity, provider CSI certification, vCluster Platform, the proposed MCP identity service and dashboard remain outside this release. Read the compatibility limits and security model.

Image digest: sha256:306ca6cd22b77d890ef3db0fe0039a7a7aea3c3553dfb21c4ce3ee715f3be0a1

Source commit: e00c9ed4d9bab09b2fd1726edf0582ac27475977

Published-artifact test run.

Replicove v0.1.0-alpha.1

Replicove's first packaged alpha makes selected Kubernetes replica environments installable through a public Helm chart, native YAML, or the CLI.

helm upgrade --install replicove oci://ghcr.io/nimeshbuilds/charts/replicove \
  --version 0.1.0-alpha.1 \
  --namespace replicove-system --create-namespace --wait --timeout 3m

• Operator: ghcr.io/nimeshbuilds/replicove:0.1.0-alpha.1 for Linux amd64 and arm64. The chart and release YAML pin its digest.
• Installs CRDs, operator, permissions, encryption key, and a destination namespace. Creates new vClusters on demand; explicitly registered existing vClusters are preserved.
• Selected resources and Helm components, grants, overrides, secrets, plans/approval, refresh, scoped access, persistent control planes, and TTL cleanup.
• CLI downloads for macOS/Linux amd64/arm64, standalone CRDs/install YAML, Helm package, and SHA-256 checksums are attached below.
• Image build provenance and SBOM are attached as OCI attestations. These are build metadata, not a separate signed release attestation.

The complete main CI suite must pass before publication. Published-artifact checks verify anonymous pulls and source labels, exercise Helm provisioning and existing-vCluster reuse, and start the operator/CLI on native arm64. Live Kubernetes coverage remains the documented disposable amd64 kind hosts; this is not production or cloud certification.

Experimental: no blanket cluster cloning, source volume data restoration, qualified IRSA/cloud identity, or vCluster Platform provisioning. Retain keys and ownership records until cleanup finishes. Read the Helm quickstart, feature docs, and compatibility limits.

Image digest: sha256:1efc1bf2167c356bdbf532b3df30d47b7404a8583a7e1925ef3bbcffd92c18a4

Source commit: 4c2df9f7d01a3e929d2794932e3144a133d2abed

Published-artifact test run.