Release notes — 0.18.0

Released 2026-07-04 with DNS Conduit 0.18.0.

Changes merged to main that are not yet tagged. Staging area for the next operator release.

Breaking changes — Rule Rhai host API

Rule Rhai scripts must be updated before upgrading to this release. Legacy host symbols are removed; there are no compatibility aliases.

Run conduitctl validate --file <config> after updating every .rhai script referenced in your config.

Migration table

Legacy	New	Notes
`table_lookup(table, key)`	`lookup(table, key)`	Global function; same lookup semantics
`txn.metric_inc(name, delta)`	`metrics.inc(name, delta)`	`metrics` object is in hook scope
`txn.metric_inc_labels(name, delta, labels)`	`metrics.inc_labels(name, delta, labels)`	Same
`log_info(msg)`	`log.info(msg)`	`log` object in hook scope
`log_warn(msg)`	`log.warn(msg)`	Same
`question_qname(txn)`	`txn.question().qname`	Use `txn.question()` when you also need qtype, class, or opcode

New read-only surface: `runtime.routing`

On pools with health enabled, scripts can read Route-consistent routing/health state at hook entry:

runtime.routing().pool(name) — configured_count, eligible_count, fail_open_active, and pool aggregates (minimum latency EWMA via min_latency_ewma_ms, max_outstanding)
runtime.routing().backend(pool, id) — applied, observed, eligible, frozen, latency EWMA (latency_ewma_ms), weight_factor, outstanding, last_transition_unix_ms (id = backend name or ip:port)
runtime.routing().backend_for_attempt(pool, backend_id) — same backend view for a named attempt; pass txn.selected_pool() and txn.selected_backend_name() for the current attempt

Use method call syntax: runtime.routing().pool("default"), not runtime.routing.pool(...).

Values reflect host state at hook entry on this worker (the same sources Route uses). There is no stronger transactional guarantee against concurrent probe or passive health updates.

Unknown pool or backend at runtime returns an empty view (documented sentinel field values).

Unchanged on `txn`

Policy effects remain on txn: set_pool, tags, retry/drop, egress overrides, question(), response(), sampling helpers, selected_*, and related methods.

Operator action

Update every rule .rhai file using the migration table above.
Re-run conduitctl validate on each config before reload.
See Host API overview, Runtime API, Data sources and lookups, and User metrics for the updated reference.

Backend health

Shipped in the same release as the Rhai host API changes above:

Active health probing, passive fast-trip, and health-aware Route selection (including latency EWMA-weighted backend shares) — Backend health
Operator controls via conduitctl health (show, set down/up, freeze, resume) — gRPC and conduitctl — health
Health-related Prometheus metrics when metrics.profile: full — Built-in metrics — Backend health
Config: pools[].health

Scripts read live health through runtime.routing on the hot path — not control-plane RPC per query. See Runtime API — routing.

Health and forward metrics

conduit_probe_results_total (full only) — active health-probe outcomes per backend (success, failure, timeout, send_error).
conduit_forward_errors_total now includes a backend label (pool, backend, reason) so timeouts and other forward errors are attributable per upstream. Existing PromQL that only groups by pool / reason still works; series identity gains the new label.
Passive fast-trip logs each counting failure and the trip event at WARN with query and client context — Backend health.

All changes in this release (automated pull request list).