github paimonsoror/telltaledns v0.6.0
TelltaleDNS 0.6.0

latest releases: edge, v0.7.0
one month ago

A release for Kubernetes: your pods show up as named workloads, failover drills test that devices really move to another node when one stops answering, and the web UI can serve HTTPS.

Highlights

  • Kubernetes workloads as devices. With kubernetes.enabled in the Helm chart, the controller watches pods and namespaces (read-only) and names each pod's address by its workload, such as media/deploy/jellyfin. Workloads appear on the Clients page's Workloads tab and across the dashboard, query log, and analytics like any other device. A workload keeps its history through pod restarts and new pod addresses, and a CronJob's runs count as one device. Groups come from pod and namespace labels, [[kubernetes.map]] rules, or a default group, so policy can follow a namespace or label. The chart forwards cluster.local to CoreDNS, so you can point the kubelet's cluster DNS at TelltaleDNS with CoreDNS second. The guide's "Your cluster's DNS through TelltaleDNS" section walks through it.

  • Failover drills. Cluster → Drill… makes one node stop answering chosen devices (some devices, a group, or everyone) for 15 to 300 seconds while the other nodes serve. The report says who followed to another node and how fast, who was stuck, who stayed quiet, and who came back. Drills refuse to start unless another node is ready and answering, and a restart cancels one. The default drops queries; REFUSED mode answers at once with EDE 14 instead. [drill] schedule = "monthly" runs one each month at a quiet hour, and a drill_report alert sends every finished report to your alert destinations.

  • Internal names get their own top list. Reverse lookups of private addresses, cluster.local (with Kubernetes on), and names you add to [telemetry] internal_names no longer fill the top domains and NXDOMAIN lists. The dashboard's Top domains card switches between Internet and Internal. The query log, live view, a device's domains, blocked lists, exports, and metrics still include them.

  • HTTPS for the web UI and API. [api] tls = { self_signed = true } makes a certificate on first start and keeps it; { cert, key } uses your own PEM files and reloads them within 10 seconds. One port serves both: plain HTTP requests are redirected to HTTPS, except the health checks (/livez, /readyz, /healthz). Settings → System shows the certificate's fingerprint to compare with the browser's before trusting it. Your own certificate's expiry is checked and alerted on like a DNS listener's.

  • Steadier encrypted upstreams. A DoT or DoH handshake now finishes even when the query that started it gives up, so the next query finds a ready connection. A connection that goes silent is closed after 3 seconds and its queries retry at once on a new one. New metrics count connections opened, handshakes, retries, and why connections closed.

  • Also new: Flow and Heatmap fill the window and follow links and Back. In a cluster, the sidebar footer shows whether the node is the primary or a replica. On the project site, a Kubernetes overview page and a "Starts on a Pi. Grows into Kubernetes." section on the overview.

Fixes

  • Upstream answers with extra bytes: an upstream answer with bytes after its last record kept them in the cache, and the OPT record added for an EDNS client then landed after them, so the client got an answer it couldn't parse. The extra bytes are now dropped. Found by the nightly fuzzing.
  • Kubernetes and a Pi in one cluster: the chart's route to CoreDNS (udp://10.43.0.10) was shared with every node, including ones outside Kubernetes, which can't reach it; their cluster.local lookups ended in SERVFAIL and their health went severe. node_only = true now works on [[upstream]], [[upstream_group]], and [[route]] (until now only on [[record]]), and the chart sets it.
  • Helm readiness: a resolver pod that stopped answering stayed in the Service for about 14 seconds. The readiness probe now runs every 2 seconds and gives up after 2 failures (about 4 seconds).
  • Helm, own pods' DNS: with the kubelet's cluster DNS pointed at TelltaleDNS, TelltaleDNS's own pods asked their own Service first. kubernetes.ownPodsUseClusterDns (on by default) sends them to clusterDns directly.

Upgrading from 0.5

Most installs need no change. Check these if they apply to you:

  • Clusters: upgrade replicas first, then the primary. A 0.5 node refuses a configuration that uses a setting it doesn't know ([kubernetes], [drill], node_only on upstreams, groups, and routes), keeps the one it had, and logs upgrade this node.
  • Helm chart: the readiness probe is faster (above). With kubernetes.enabled, TelltaleDNS's own pods now get dnsPolicy: None with clusterDns as their server; set kubernetes.ownPodsUseClusterDns: false to keep the old behavior. If your nodes resolve through TelltaleDNS, leave kubernetes.podCidr and serviceCidr empty: CoreDNS's fallthrough would send unknown reverse lookups back in a loop.
  • Top lists: from the upgrade on, private reverse lookups and cluster.local count in the new Internal list, so the top domains and NXDOMAIN lists may look shorter. Scripts reading GET /api/v1/stats/top?kind=domains can ask for kind=internal too.
  • Query log and metrics: a query dropped by a drill has the new status drilled.
  • API listener: changing [api] enabled, listen, or tls now needs a restart, and the API reports it as such. telltale ctl and telltale mcp --stdio follow the config to https://, trust the node's own self-signed certificate, and take --ca-file for others.

Known issues

  • DNSSEC validate mode: right after a start, an upstream that closes fresh DoT connections can make the first validations fail, so a signed name may get SERVFAIL until its keys are cached. Keep permissive or off for now; see ADR-098.
  • _dns.resolver.arpa (encrypted-DNS discovery) is forwarded upstream instead of being answered locally.

For developers

  • New ADRs: 148–152 (all Proposed); ADR-122 amended (the Kubernetes pages in the site's Features menu).
  • New MCP tools: list_workloads, get_workload, plan_failover_drill, drill_report; top_items takes kind: internal.
  • New API paths: /kubernetes/workloads, /kubernetes/status, /kubernetes/map/{key}, /nodes/{id}/drill, /drills, /drills/{id}, /clients/affinity. New token scope ops:drill.
  • New peer calls: workloads.table (push and pull) and the primary's relay for drill windows and reports; an older primary refuses relay, and a drill then reports only what the drilled node can reach.

Full Changelog: v0.5.0...v0.6.0

Don't miss a new telltaledns release

NewReleases is sending notifications on new releases.