fix(cargo-anvil): preserve mutation runner resource diagnostics

.anvil.lock

@@ -1,7 +1,7 @@
 version = 1
 tool = "anvil"
 tool_version = "0.13.0"
-catalog_checksum = "sha256:dfaf58166631905ff5990c0dbf4741a15c9db94d33804212faad51d7f9c3790b"
+catalog_checksum = "sha256:09e913b54324919fbe67ed027b40c3e7b3e53cd9e0698f72f24cae686332d451"
 
 [[file]]
 path = ".anvil/container/Dockerfile.dockerignore"
@@ -17,7 +17,7 @@ checksum = "sha256:9940d1947482150ac08fcb9b4150da99f5ae60642f4caeea137577ce0e709
 
 [[file]]
 path = ".github/actions/anvil-run-group/action.yml"
-checksum = "sha256:d564a0ce424cda58c8f5b2476cf0c8b36e759afbdf8227608bd1f3fc74913fdf"
+checksum = "sha256:6f61042746607c4afe3fcef1b827461f848cb0ebfc2e69e1eda9f17e889f04cf"
 
 [[file]]
 path = ".github/actions/anvil-setup/action.yml"
@@ -161,7 +161,7 @@ checksum = "sha256:adc3b25f4f0b563695335f85870d9f319606c5ca80fa3d118cf221388f4bb
 
 [[file]]
 path = "justfiles/anvil/checks/mutants-diff.just"
-checksum = "sha256:b8e32d8c03f3c68a2b4b9eba7998ccabb36d29abff39f21d46005fdc281bb7c3"
+checksum = "sha256:fa74227f1c6121d68bb37b054e890e8d7f329e2b15caa624b35da564fec13d7f"
 
 [[file]]
 path = "justfiles/anvil/checks/mutants-full.just"

.github/actions/anvil-run-group/action.yml

@@ -61,6 +61,59 @@ runs:
         # Capture the group result before propagating failure so the
         # best-effort reporter can consume both outputs.
         log="$RUNNER_TEMP/anvil-$ANVIL_GROUP.log"
+        resource_pid=""
+        if [[ "$ANVIL_GROUP" == "pr-mutants" && "$RUNNER_OS" == "Linux" ]]; then
+          # Mutants intentionally crash. Core files are not used by this check.
+          ulimit -c 0
+          (
+            sleep_pid=""
+            trap 'if [[ -n "$sleep_pid" ]]; then kill "$sleep_pid"; fi; exit 143' TERM
+            while true; do
+              {
+                sample_time="$(date -u +%FT%TZ)" || echo "::warning::Mutation clock probe failed"
+                echo "[anvil-resources] $sample_time"
+                free -m || echo "::warning::Mutation memory probe failed"
+                df -m . || echo "::warning::Mutation disk probe failed"
+                ps -eo pid,ppid,rss,pcpu,comm --sort=-rss | sed -n '1,12p' || echo "::warning::Mutation process probe failed"
+                if [[ -f mutants.out/debug.log ]]; then
+                  tail -n 1 mutants.out/debug.log || echo "::warning::Mutation debug-log probe failed"
+                fi
+              } 2>&1 | tee -a "$RUNNER_TEMP/anvil-$ANVIL_GROUP-resources.log" || echo "::warning::Mutation resource log write failed"
+              sleep 30 &
+              sleep_pid=$!
+              wait "$sleep_pid"
+              sleep_pid=""
+            done
+          ) &
+          resource_pid=$!
+          stop_resources() {
+            if ! kill "$resource_pid" 2>/dev/null; then
+              echo "::warning::Mutation resource sampler exited before cleanup"
+            fi
+            wait "$resource_pid" || {
+              sampler_status=$?
+              if [[ "$sampler_status" != 143 ]]; then
+                echo "::warning::Mutation resource sampler exited with status $sampler_status"
+              fi
+            }
+          }
+          core_pattern=""
+          restore_core_pattern=false
+          cleanup_mutation() {
+            stop_resources
+            if [[ "$restore_core_pattern" == true ]]; then
+              sudo -n sysctl -w "kernel.core_pattern=$core_pattern" || echo "::warning::Could not restore mutation core-dump routing"
+            fi
+          }
+          trap cleanup_mutation EXIT
+          if [[ "${RUNNER_ENVIRONMENT:-}" == "github-hosted" ]]; then
+            # Linux ignores ulimit for piped handlers such as Ubuntu's apport.
+            # Only ephemeral hosted runners may change this host-wide setting.
+            core_pattern="$(sysctl -n kernel.core_pattern)"
+            restore_core_pattern=true
+            sudo -n sysctl -w kernel.core_pattern=core
+          fi
+        fi
         set +e
         just "anvil-$ANVIL_GROUP" 2>&1 | tee "$log"
         # Preserve Just's status rather than tee's pipeline status.
@@ -74,6 +127,18 @@ runs:
         echo "exit_code=$status" >> "$GITHUB_OUTPUT"
         exit "$status"
 
+    - name: Upload mutation diagnostics
+      if: always() && inputs.group == 'pr-mutants'
+      continue-on-error: true
+      uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
+      with:
+        name: anvil-mutants-${{ runner.os }}-${{ runner.arch }}-${{ github.run_attempt }}
+        path: |
+          ${{ runner.temp }}/anvil-pr-mutants*.log
+          ${{ github.workspace }}/mutants.out/
+        retention-days: 3
+        if-no-files-found: warn
+
     # Reporting is supplemental: run after success or failure, but never let
     # an API outage determine the authoritative workflow-job result.
     - name: Publish supplemental Anvil commit status

crates/cargo-anvil/docs/design/checks.md

@@ -294,7 +294,7 @@ available.
 
 | Check     | Invocation                                                                  | Source |
 |-----------|-----------------------------------------------------------------------------|--------|
-| `mutants` | Resolves the base ref, writes `git diff <base>` against the working tree to a temporary unified-diff file, then runs `cargo mutants --in-diff <file> --no-shuffle --jobs 0`. Self-skips on aarch64-pc-windows-msvc where cargo-mutants doesn't build; other ARM legs run normally. | oxidizer-github |
+| `mutants` | Resolves the base ref, writes `git diff <base>` against the working tree to a temporary unified-diff file, then runs `cargo mutants --in-diff <file> --no-shuffle --jobs 0 --caught --unviable`. Self-skips on aarch64-pc-windows-msvc where cargo-mutants doesn't build; other ARM legs run normally. | oxidizer-github |
 
 The mutants check requires a base ref: locally the recipe resolves `BASE_REF` (if set), then `origin/main`, then `origin/master`, then errors out. GitHub passes `${{ github.event.pull_request.base.sha }}` as `BASE_REF`; on ADO the shared resolver reads `$(System.PullRequest.TargetBranch)` from the environment.
 

crates/cargo-anvil/docs/design/github.md

@@ -683,6 +683,27 @@ workflows. Anvil does not emit or support public per-group action paths.
 
 ### Failure attribution and commit statuses
 
+PR mutation jobs print caught and unviable mutations as they complete. On Linux,
+the shared action also streams memory, disk, process-resource usage and the
+latest cargo-mutants debug record every 30 seconds. It disables core dumps for
+the mutation group: intentionally crashing mutations still fail normally, but
+do not generate unused core files. On ephemeral GitHub-hosted Linux runners it
+also temporarily replaces piped crash-handler routing with file routing:
+Linux ignores the core-size limit for piped handlers, and repeated aborts can
+exhaust memory with Ubuntu's `apport` processes. The original routing is
+restored on exit. Configuration failure stops the group; restoration failure
+emits a warning without changing the test result. Self-hosted runners do not
+change this host-wide setting. Other groups retain their existing limits.
+
+After success or failure, mutation jobs attempt to upload the group log,
+resource samples and `mutants.out` as an OS/architecture-specific artifact with
+three-day retention. Artifact publication is supplemental and cannot override
+the mutation exit status. A runner that disappears before publication can still
+lose its local files; streamed samples improve diagnosis but do not guarantee
+recovery after runner loss. The resource sampler is stopped when the run step
+exits, including on a failed mutation check. Probe and resource-log write failures
+emit warnings without preventing later samples or changing the group's result.
+
 GitHub fixes workflow job names before a job runs, so a matrix check named
 `PR Job / Check Group: Fast Checks (linux)` cannot rename itself after
 discovering that `anvil-license-headers` failed. The checks UI groups that name

crates/cargo-anvil/templates/github/run-group-action.yml

@@ -61,6 +61,59 @@ runs:
         # Capture the group result before propagating failure so the
         # best-effort reporter can consume both outputs.
         log="$RUNNER_TEMP/anvil-$ANVIL_GROUP.log"
+        resource_pid=""
+        if [[ "$ANVIL_GROUP" == "pr-mutants" && "$RUNNER_OS" == "Linux" ]]; then
+          # Mutants intentionally crash. Core files are not used by this check.
+          ulimit -c 0
+          (
+            sleep_pid=""
+            trap 'if [[ -n "$sleep_pid" ]]; then kill "$sleep_pid"; fi; exit 143' TERM
+            while true; do
+              {
+                sample_time="$(date -u +%FT%TZ)" || echo "::warning::Mutation clock probe failed"
+                echo "[anvil-resources] $sample_time"
+                free -m || echo "::warning::Mutation memory probe failed"
+                df -m . || echo "::warning::Mutation disk probe failed"
+                ps -eo pid,ppid,rss,pcpu,comm --sort=-rss | sed -n '1,12p' || echo "::warning::Mutation process probe failed"
+                if [[ -f mutants.out/debug.log ]]; then
+                  tail -n 1 mutants.out/debug.log || echo "::warning::Mutation debug-log probe failed"
+                fi
+              } 2>&1 | tee -a "$RUNNER_TEMP/anvil-$ANVIL_GROUP-resources.log" || echo "::warning::Mutation resource log write failed"
+              sleep 30 &
+              sleep_pid=$!
+              wait "$sleep_pid"
+              sleep_pid=""
+            done
+          ) &
+          resource_pid=$!
+          stop_resources() {
+            if ! kill "$resource_pid" 2>/dev/null; then
+              echo "::warning::Mutation resource sampler exited before cleanup"
+            fi
+            wait "$resource_pid" || {
+              sampler_status=$?
+              if [[ "$sampler_status" != 143 ]]; then
+                echo "::warning::Mutation resource sampler exited with status $sampler_status"
+              fi
+            }
+          }
+          core_pattern=""
+          restore_core_pattern=false
+          cleanup_mutation() {
+            stop_resources
+            if [[ "$restore_core_pattern" == true ]]; then
+              sudo -n sysctl -w "kernel.core_pattern=$core_pattern" || echo "::warning::Could not restore mutation core-dump routing"
+            fi
+          }
+          trap cleanup_mutation EXIT
+          if [[ "${RUNNER_ENVIRONMENT:-}" == "github-hosted" ]]; then
+            # Linux ignores ulimit for piped handlers such as Ubuntu's apport.
+            # Only ephemeral hosted runners may change this host-wide setting.
+            core_pattern="$(sysctl -n kernel.core_pattern)"
+            restore_core_pattern=true
+            sudo -n sysctl -w kernel.core_pattern=core
+          fi
+        fi
         set +e
         just "anvil-$ANVIL_GROUP" 2>&1 | tee "$log"
         # Preserve Just's status rather than tee's pipeline status.
@@ -74,6 +127,18 @@ runs:
         echo "exit_code=$status" >> "$GITHUB_OUTPUT"
         exit "$status"
 
+    - name: Upload mutation diagnostics
+      if: always() && inputs.group == 'pr-mutants'
+      continue-on-error: true
+      uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
+      with:
+        name: anvil-mutants-${{ runner.os }}-${{ runner.arch }}-${{ github.run_attempt }}
+        path: |
+          ${{ runner.temp }}/anvil-pr-mutants*.log
+          ${{ github.workspace }}/mutants.out/
+        retention-days: 3
+        if-no-files-found: warn
+
     # Reporting is supplemental: run after success or failure, but never let
     # an API outage determine the authoritative workflow-job result.
     - name: Publish supplemental Anvil commit status

crates/cargo-anvil/templates/justfiles/anvil/checks/mutants-diff.just

@@ -52,7 +52,7 @@ anvil-mutants-diff: anvil-mutants-diff-validate-prereqs anvil-impact
     # not a behaviour change for it.
     git diff "$base" --output=$diff_path
     if ($LASTEXITCODE -ne 0) { exit $LASTEXITCODE }
-    & cargo {{_anvil_stable_toolchain_args}} mutants --in-diff $diff_path --no-shuffle --jobs 0 {{_anvil_mutants_config_arg}}
+    & cargo {{_anvil_stable_toolchain_args}} mutants --in-diff $diff_path --no-shuffle --jobs 0 --caught --unviable {{_anvil_mutants_config_arg}}
     if ($LASTEXITCODE -ne 0) { exit $LASTEXITCODE }
 
 # --- pr-mutants members ---