ci: cap recurring Blacksmith spend

.config/nextest.toml

@@ -34,5 +34,14 @@ final-status-level = "slow"
 slow-timeout = { period = "60s", terminate-after = 2 }
 retries = 1
 
+# Building a 200-table persistent catalog is intentionally CPU- and I/O-heavy.
+# Give this one overflow-chain regression exclusive access to the public runner
+# and a larger hard timeout; every other test keeps the 120-second ceiling.
+[[profile.ci.overrides]]
+filter = 'test(massive_catalog_forces_meta_overflow_chain)'
+threads-required = "num-cpus"
+slow-timeout = { period = "60s", terminate-after = 5 }
+retries = 0
+
 [profile.ci.junit]
 path = "junit.xml"

.github/workflows/blacksmith-testbox.yml

@@ -0,0 +1,102 @@
+name: Blacksmith Testbox
+
+# Permanent reddb-only pilot. Warm with `--idle-timeout 10`, run the bounded
+# comparison command, then always call `blacksmith testbox stop --id <id>`.
+# Select exactly one job with `blacksmith testbox warmup blacksmith-testbox.yml
+# --job rust-{2,4,8}vcpu --idle-timeout 10`.
+on:
+  workflow_dispatch:
+    inputs:
+      testbox_id:
+        type: string
+        description: Testbox session ID
+        required: true
+
+permissions:
+  contents: read
+
+env:
+  CARGO_TERM_COLOR: always
+  CARGO_INCREMENTAL: 0
+
+jobs:
+  rust-2vcpu:
+    name: Rust 2 vCPU
+    runs-on: blacksmith-2vcpu-ubuntu-2404
+    timeout-minutes: 90
+    steps:
+      - name: Begin Testbox
+        uses: useblacksmith/begin-testbox@233448af4bfdc6fca509a7f0974411ac6d8a8043 # v2
+        with:
+          testbox_id: ${{ inputs.testbox_id }}
+      - uses: actions/checkout@v7
+      - uses: dtolnay/rust-toolchain@1.91.0
+      - uses: ./.github/actions/install-protoc
+        with:
+          version: '28.3'
+      - uses: rui314/setup-mold@v1
+      - uses: mozilla-actions/sccache-action@v0.0.10
+      - uses: Swatinem/rust-cache@v2
+        with:
+          shared-key: testbox-rust-1.91.0-2vcpu
+          cache-on-failure: 'true'
+      - uses: taiki-e/install-action@v2
+        with:
+          tool: nextest
+      - name: Run Testbox
+        uses: useblacksmith/run-testbox@5ca05834db1d3813554d1dd109e5f2087a8d7cbc # v2
+        if: always()
+
+  rust-4vcpu:
+    name: Rust 4 vCPU
+    runs-on: blacksmith-4vcpu-ubuntu-2404
+    timeout-minutes: 90
+    steps:
+      - name: Begin Testbox
+        uses: useblacksmith/begin-testbox@233448af4bfdc6fca509a7f0974411ac6d8a8043 # v2
+        with:
+          testbox_id: ${{ inputs.testbox_id }}
+      - uses: actions/checkout@v7
+      - uses: dtolnay/rust-toolchain@1.91.0
+      - uses: ./.github/actions/install-protoc
+        with:
+          version: '28.3'
+      - uses: rui314/setup-mold@v1
+      - uses: mozilla-actions/sccache-action@v0.0.10
+      - uses: Swatinem/rust-cache@v2
+        with:
+          shared-key: testbox-rust-1.91.0-4vcpu
+          cache-on-failure: 'true'
+      - uses: taiki-e/install-action@v2
+        with:
+          tool: nextest
+      - name: Run Testbox
+        uses: useblacksmith/run-testbox@5ca05834db1d3813554d1dd109e5f2087a8d7cbc # v2
+        if: always()
+
+  rust-8vcpu:
+    name: Rust 8 vCPU
+    runs-on: blacksmith-8vcpu-ubuntu-2404
+    timeout-minutes: 90
+    steps:
+      - name: Begin Testbox
+        uses: useblacksmith/begin-testbox@233448af4bfdc6fca509a7f0974411ac6d8a8043 # v2
+        with:
+          testbox_id: ${{ inputs.testbox_id }}
+      - uses: actions/checkout@v7
+      - uses: dtolnay/rust-toolchain@1.91.0
+      - uses: ./.github/actions/install-protoc
+        with:
+          version: '28.3'
+      - uses: rui314/setup-mold@v1
+      - uses: mozilla-actions/sccache-action@v0.0.10
+      - uses: Swatinem/rust-cache@v2
+        with:
+          shared-key: testbox-rust-1.91.0-8vcpu
+          cache-on-failure: 'true'
+      - uses: taiki-e/install-action@v2
+        with:
+          tool: nextest
+      - name: Run Testbox
+        uses: useblacksmith/run-testbox@5ca05834db1d3813554d1dd109e5f2087a8d7cbc # v2
+        if: always()

.github/workflows/changesets.yml

@@ -12,7 +12,7 @@ concurrency:
 jobs:
   changesets:
     name: Version Packages
-    runs-on: blacksmith-4vcpu-ubuntu-2404
+    runs-on: ubuntu-latest
     # Skip on the version-bump commit itself to avoid an infinite loop. The
     # commit landing the version bump is authored by github-actions[bot] —
     # that push doesn't need to re-trigger this workflow.

.github/workflows/ci.yml

@@ -1,6 +1,8 @@
 name: CI
 
 on:
+  merge_group:
+    types: [checks_requested]
   push:
     branches: [ main, develop ]
     paths-ignore:
@@ -32,13 +34,13 @@ on:
 # Cost guard: new pushes cancel older runs on the same ref. Release-grade
 # validation lives in release.yml; CI push runs should stay quick.
 #
-# Cost policy (July 2026, Blacksmith bill): expensive jobs (feature matrix,
-# cross-platform, driver conformance, containers, fuzz smoke, packaging) run
-# POST-MERGE on push to main/develop — never on an open PR. PRs keep the fast
-# correctness lane: gate, lints, contract gates, quality, tests, RQL
-# conformance. `workflow_dispatch` with full_ci=true still runs everything
-# on demand. rust-cache saves are restricted to non-PR runs (save-if) so PR
-# branches never persist multi-GB caches.
+# Cost policy (August 2026, Blacksmith bill): pushes and pull requests run the
+# trunk correctness lane on free standard runners for this public repository.
+# Feature matrices, cross-platform builds, driver conformance, containers,
+# fuzz smoke, and packaging are release-adjacent and run on Blacksmith only
+# through `workflow_dispatch` with full_ci=true. Stable release builds remain
+# authoritative in release.yml. rust-cache saves are restricted to non-PR runs
+# (save-if) so PR branches never persist multi-GB caches.
 concurrency:
   group: ci-${{ github.workflow }}-${{ github.ref }}
   cancel-in-progress: true
@@ -55,7 +57,8 @@ jobs:
   # Dependabot (Dependabot PRs run the dedicated dependabot.yml workflow instead).
   gate:
     name: gate
-    runs-on: blacksmith-2vcpu-ubuntu-2404
+    runs-on: ubuntu-latest
+    timeout-minutes: 2
     if: |
       github.actor != 'dependabot[bot]' &&
       (github.event_name != 'pull_request' || github.event.pull_request.draft == false)
@@ -224,7 +227,7 @@ jobs:
   quality:
     name: Quality (fmt, check, clippy)
     needs: gate
-    runs-on: blacksmith-4vcpu-ubuntu-2404
+    runs-on: ubuntu-24.04
     timeout-minutes: 20
     env:
       SCCACHE_GHA_ENABLED: "true"
@@ -260,7 +263,7 @@ jobs:
     # every later RQL-extraction slice must keep green.
     name: RQL Conformance (sqllogictest)
     needs: gate
-    runs-on: blacksmith-4vcpu-ubuntu-2404
+    runs-on: ubuntu-24.04
     timeout-minutes: 20
     env:
       SCCACHE_GHA_ENABLED: "true"
@@ -292,7 +295,7 @@ jobs:
     # that motivated this gate.
     name: Drivers / Python (cargo check)
     needs: gate
-    runs-on: blacksmith-2vcpu-ubuntu-2404
+    runs-on: ubuntu-24.04
     timeout-minutes: 15
     env:
       SCCACHE_GHA_ENABLED: "true"
@@ -322,7 +325,7 @@ jobs:
     # operators may select for release artifacts.
     name: Feature Matrix (${{ matrix.features.label }})
     needs: gate
-    if: github.event_name == 'push' || (github.event_name == 'workflow_dispatch' && inputs.full_ci)
+    if: github.event_name == 'workflow_dispatch' && inputs.full_ci
     runs-on: blacksmith-4vcpu-ubuntu-2404
     timeout-minutes: 20
     strategy:
@@ -357,7 +360,7 @@ jobs:
   tests:
     name: Test Suite
     needs: gate
-    runs-on: blacksmith-8vcpu-ubuntu-2404
+    runs-on: ubuntu-24.04
     timeout-minutes: 20
     env:
       SCCACHE_GHA_ENABLED: "true"
@@ -427,7 +430,7 @@ jobs:
   driver-param-conformance:
     name: Driver Param Conformance
     needs: gate
-    if: github.event_name == 'push' || (github.event_name == 'workflow_dispatch' && inputs.full_ci)
+    if: github.event_name == 'workflow_dispatch' && inputs.full_ci
     runs-on: blacksmith-4vcpu-ubuntu-2404
     timeout-minutes: 10
     steps:
@@ -501,7 +504,7 @@ jobs:
     # statically links the engine, so `maturin develop` is self-contained.
     name: Driver Runnable Examples
     needs: gate
-    if: github.event_name == 'push' || (github.event_name == 'workflow_dispatch' && inputs.full_ci)
+    if: github.event_name == 'workflow_dispatch' && inputs.full_ci
     runs-on: blacksmith-4vcpu-ubuntu-2404
     timeout-minutes: 35
     env:
@@ -550,7 +553,7 @@ jobs:
     # tarball, so this also exercises the include manifest end-to-end.
     name: Publish Dry-Run (crates.io)
     needs: gate
-    if: github.event_name == 'push' || (github.event_name == 'workflow_dispatch' && inputs.full_ci)
+    if: github.event_name == 'workflow_dispatch' && inputs.full_ci
     runs-on: blacksmith-4vcpu-ubuntu-2404
     timeout-minutes: 25
     env:
@@ -577,7 +580,7 @@ jobs:
   containers:
     name: Container Stack
     needs: gate
-    if: github.event_name == 'push' || (github.event_name == 'workflow_dispatch' && inputs.full_ci)
+    if: github.event_name == 'workflow_dispatch' && inputs.full_ci
     runs-on: blacksmith-4vcpu-ubuntu-2404
     timeout-minutes: 25
     steps:
@@ -716,7 +719,7 @@ jobs:
     # (systemd install, mmap, Unix sockets) via cfg gates in the source.
     name: Windows (build + unit tests)
     needs: gate
-    if: github.event_name == 'push' || (github.event_name == 'workflow_dispatch' && inputs.full_ci)
+    if: github.event_name == 'workflow_dispatch' && inputs.full_ci
     runs-on: blacksmith-8vcpu-windows-2025
     timeout-minutes: 30
     steps:
@@ -741,7 +744,7 @@ jobs:
     # story complete. Same scope — library build + unit tests.
     name: macOS (build + unit tests)
     needs: gate
-    if: github.event_name == 'push' || (github.event_name == 'workflow_dispatch' && inputs.full_ci)
+    if: github.event_name == 'workflow_dispatch' && inputs.full_ci
     runs-on: blacksmith-6vcpu-macos-15
     timeout-minutes: 30
     steps:
@@ -760,11 +763,11 @@ jobs:
         run: cargo test --locked --lib
 
   fuzz-parsers:
-    # Post-merge bounded smoke window (cost policy above); the long
+    # On-demand bounded smoke window (cost policy above); the long
     # exploration window lives in parser-fuzz-nightly.yml.
     name: Fuzz Parsers
     needs: gate
-    if: github.event_name == 'push' || (github.event_name == 'workflow_dispatch' && inputs.full_ci)
+    if: github.event_name == 'workflow_dispatch' && inputs.full_ci
     runs-on: blacksmith-4vcpu-ubuntu-2404
     timeout-minutes: 15
     env:
@@ -819,7 +822,7 @@ jobs:
   package:
     name: cargo package dry-run
     needs: gate
-    if: github.event_name == 'push' || (github.event_name == 'workflow_dispatch' && inputs.full_ci)
+    if: github.event_name == 'workflow_dispatch' && inputs.full_ci
     runs-on: blacksmith-4vcpu-ubuntu-2404
     timeout-minutes: 15
     steps:

.github/workflows/crypto-coverage.yml

@@ -25,7 +25,7 @@ jobs:
   crypto-coverage:
     name: Crypto Coverage Gate
     if: github.actor != 'dependabot[bot]'
-    runs-on: blacksmith-4vcpu-ubuntu-2404
+    runs-on: ubuntu-latest
     permissions:
       contents: read
     timeout-minutes: 15

.github/workflows/dependabot.yml

@@ -27,7 +27,7 @@ jobs:
   check:
     name: Verify update compiles
     if: github.actor == 'dependabot[bot]'
-    runs-on: blacksmith-4vcpu-ubuntu-2404
+    runs-on: ubuntu-latest
     timeout-minutes: 20
     steps:
       - uses: actions/checkout@v7
@@ -64,7 +64,7 @@ jobs:
     name: Auto-approve patch/minor bumps
     needs: check
     if: github.actor == 'dependabot[bot]'
-    runs-on: blacksmith-2vcpu-ubuntu-2404
+    runs-on: ubuntu-latest
     permissions:
       pull-requests: write
     steps:

.github/workflows/docs.yml

@@ -20,7 +20,7 @@ concurrency:
 jobs:
   deploy:
     name: Deploy to GitHub Pages
-    runs-on: blacksmith-4vcpu-ubuntu-2404
+    runs-on: ubuntu-latest
     environment:
       name: github-pages
       url: ${{ steps.deployment.outputs.page_url }}

.github/workflows/drill-nightly.yml

@@ -1,8 +1,8 @@
-name: Nightly DR Drill
+name: Weekly DR Drill
 
 on:
   schedule:
-    - cron: '0 3 * * *'
+    - cron: '0 3 * * 2'
   workflow_dispatch:
 
 permissions:
@@ -16,7 +16,7 @@ env:
 jobs:
   drill:
     name: Backup/Restore Drills
-    runs-on: blacksmith-4vcpu-ubuntu-2404
+    runs-on: ubuntu-24.04
     timeout-minutes: 30
     env:
       SCCACHE_GHA_ENABLED: "true"

.github/workflows/dst-nightly.yml

@@ -1,8 +1,8 @@
-name: DST Nightly
+name: DST Weekly
 
 on:
   schedule:
-    - cron: "23 7 * * *"
+    - cron: "23 7 * * 4"
   workflow_dispatch:
 
 permissions:
@@ -15,7 +15,7 @@ env:
 jobs:
   dst-seed-sweep:
     name: DST seed sweep
-    runs-on: blacksmith-4vcpu-ubuntu-2404
+    runs-on: ubuntu-24.04
     timeout-minutes: 20
     steps:
       - uses: actions/checkout@v7
@@ -32,7 +32,7 @@ jobs:
 
   storage-fault-recovery:
     name: Storage fault recovery
-    runs-on: blacksmith-4vcpu-ubuntu-2404
+    runs-on: ubuntu-24.04
     timeout-minutes: 25
     permissions:
       contents: read

.github/workflows/file-coverage.yml

@@ -25,7 +25,7 @@ jobs:
   file-coverage:
     name: File Coverage Gate
     if: github.actor != 'dependabot[bot]'
-    runs-on: blacksmith-4vcpu-ubuntu-2404
+    runs-on: ubuntu-latest
     permissions:
       contents: read
     timeout-minutes: 30