Skip to content

Releases

v0.8.0

Choose a tag to compare

@ssahani ssahani released this 22 Nov 13:53
2a02977

Added

  • Resource pools (Experimental, ZORVIA_EXPERIMENTAL=1): named VM sets via pool.zorvia.io/<name> labels with optional max VMs / vCPUs / memory limits stored in the state store, a read-only usage report, a datacenter > cluster > folder > pool view, /v1/inventory/pools API (cluster.admin, audited) and zorvia inventory pool .... Limits are advisory: VM create returns pool_warnings and never blocks. Fake-data unit tests only, no web page. See docs/RESOURCE_POOLS.md.

  • Volume encryption: real AWS KMS and Google Cloud KMS key wrappers replace CloudKmsStub (Experimental). REST over reqwest, no SDK and no new crate; AWS SigV4 implemented in-tree and checked against the published AWS test vectors; wrap bound to the namespace/vm/volume context (AWS EncryptionContext, GCP AAD); credentials from env, a mounted file or GCP workload identity only; https only, timeouts, 64 KiB response cap, throttling retries with backoff; error text never carries response bodies. Selected by ZORVIA_AWS_KMS_KEY_ID / ZORVIA_GCP_KMS_KEY. Tested against mock servers only, never a real cloud KMS; no Azure Key Vault. See docs/VOLUME_ENCRYPTION.md.

  • Small polish: zorvia dr pair delete NAME --yes (server refuses with 409 while a failover or failback is active); --wait/--timeout for zorvia rescue reset-password|install-packages|inspect; ZORVIA_AUTOSCALER_PROMETHEUS_TOKEN_FILE (bearer token for the Prometheus source, file only, https only, re-read per request); GET /api/images/upload-config and the web upload panel shows the real size cap; a test that bash/zsh/fish completions include dr pair, volume encryption and rescue. Mock-server tested only.

  • Autoscaler decisions view (Experimental, read-only): the loop keeps a bounded in-memory ring of recent decisions per target (timestamp, metric, load, source, decision, applied/recommended/held/skipped, reason; 100 per target, identical holds folded, memory only). New GET /api/v1/autoscaler/status and /decisions?limit=&target= (cluster.admin, EXPERIMENTAL_DISABLED when off), zorvia autoscaler status|decisions, and an /app/autoscaler console page with the experimental, disabled state. Unit, mock-server and component tested; not exercised on a live cluster. See docs/AUTOSCALER_LOOP.md.

  • Certificate lifecycle (Experimental, ZORVIA_EXPERIMENTAL=1): inventory of kubernetes.io/tls Secrets, the serving certificate and configured CA files (subject, issuer, SANs, validity, key type/size, self-signed, weak-algorithm flags; metadata only, tls.key never read), leader-gated expiry alerts once per threshold (ZORVIA_CERT_ALERT_DAYS, default 30/14/7/1), cert-manager Ready state and an admin-only, audited, confirmed renew (Issuing condition, like cmctl renew). /api/certificates/*, zorvia certs ..., /app/certificates. Parsing uses the existing OpenSSL dependency (no new crate). Unit-tested with generated certificates and a fake store; not run against a live cluster or cert-manager. See docs/CERTIFICATE_LIFECYCLE.md.

  • Guest LUKS mode for volume encryption (Experimental, partial): mode: guest_luks plan/apply (blank DataVolume plus wrapped-key Secret), one-time unlock tokens (POST /volume-encryption/guest-luks/token, single use, 30-900 s TTL, bound to namespace/VM/volume, constant-time compare, rate limited, audited), an unauthenticated redeem route the guest calls with the token, and a generated cloud-init (script, unit, crypttab/fstab) that never embeds key material. CLI: --guest-luks, zorvia volume encryption guest-token. Missing: automatic token delivery at every VM start; never run in a real guest. See docs/VOLUME_ENCRYPTION.md.

  • Rescue agent: install-packages now bind-mounts /proc, /dev and /sys into the guest chroot and always unmounts them (reverse order, even on error); the outbound-network requirement is documented. The rescue-agent binary now builds with --no-default-features --features rescue-agent (the Dockerfile's flags). A test keeps the secrets RBAC rule identical across the Helm chart, deploy/k8s.yaml and deploy/k3s-zorvia-web.yaml. See docs/RESCUE.md.

  • CLI for the API-backed experimental features: zorvia dr pair list|create|status|failover|failback-plan|resume (dry-run default, exact confirmation phrase), zorvia volume encryption status|plan|apply|rotate, zorvia rescue reset-password|install-packages|inspect|status (password from a no-echo prompt or stdin only). Clear refusal on 403 EXPERIMENTAL_DISABLED. Mock-server tested only. No golden-gates or autoscaler status command (no server API). See docs/CLI_REMOTE.md.

  • Disk image upload through the CDI upload proxy (Experimental, ZORVIA_EXPERIMENTAL=1): zorvia guest image-upload, POST /api/images/uploads/{name}, POST /api/images/validate; magic-byte format detection (qcow2/raw/vmdk/vhdx/vhd/vdi/iso), streamed with a size cap and SHA-256 check, upload token never exposed. Tested against a local mock proxy only. See docs/IMAGE_UPLOAD.md.

  • Opt-in autoscaler reconcile loop (Experimental, ZORVIA_AUTOSCALER_LOOP=1): hysteresis, cooldown, min/max bounds, recommendation-only default, leader-gated, audited. Not live-verified; the existing monitoring load source is simulated so nothing is applied from it. See docs/AUTOSCALER_LOOP.md.

  • Autoscaler real load source (Experimental): the loop now reads VM CPU/memory from the Kubernetes metrics API (PodMetrics for virt-launcher pods, mapped to VM size), with an optional Prometheus fallback (ZORVIA_AUTOSCALER_PROMETHEUS_URL, https required except loopback, 1 MiB cap, timeouts, no redirects). Missing, stale or malformed data yields no decision. New target field load_metric (cpu/memory). Mock-server tested only. See docs/AUTOSCALER_LOOP.md.

  • SAML 2.0 SSO groundwork (Experimental, partial, opt-in ZORVIA_SAML_ENABLED=1): SP metadata, SP-initiated AuthnRequest (HTTP-Redirect), strict assertion validation with replay cache and group-to-role mapping. XML signature verifica

(Truncated; see CHANGELOG.md for the full list.)