Dov/antithesis poc upstream

Cargo.toml

@@ -263,6 +263,7 @@ ahash = { version = "0.8.12", default-features = false }
 aho-corasick = "1.1.4"
 allocation-counter = "0"
 anyhow = "1.0.102"
+antithesis_sdk = "0.2.8"
 array-concat = "0.5.5"
 arrayvec = "0.7.6"
 arrow = { version = "57", default-features = false }

bin/ci-builder

@@ -18,6 +18,9 @@ set -euo pipefail
 
 NIGHTLY_RUST_DATE=2026-05-06
 
+# Allow overriding the container runtime (e.g. MZ_DEV_CI_BUILDER_RUNTIME=podman).
+DOCKER="${MZ_DEV_CI_BUILDER_RUNTIME:-docker}"
+
 workdir=$(pwd)
 cd "$(dirname "$0")/.."
 
@@ -128,10 +131,14 @@ gid=$(id -g)
 [[ "$gid" -lt 500 ]] && gid=$uid
 
 build() {
+    local cache_args=()
+    if [[ "$DOCKER" != "podman" ]]; then
+        cache_args+=(--cache-from=materialize/ci-builder:"$cache_tag")
+        cache_args+=(--cache-to=type=inline,mode=max)
+    fi
     # shellcheck disable=SC2086 # intentional splitting of build args string
-    docker buildx build --pull \
-        --cache-from=materialize/ci-builder:"$cache_tag" \
-        --cache-to=type=inline,mode=max \
+    "$DOCKER" buildx build --pull \
+        "${cache_args[@]}" \
         $docker_build_args \
         --tag materialize/ci-builder:"$tag" \
         --tag ghcr.io/materializeinc/materialize/ci-builder:"$tag" \
@@ -181,13 +188,13 @@ case "$cmd" in
         build "$@"
         ;;
     exists)
-        docker manifest inspect "$image_registry"/ci-builder:"$tag" &> /dev/null
+        "$DOCKER" manifest inspect "$image_registry"/ci-builder:"$tag" &> /dev/null
         ;;
     tag)
         echo "$tag"
         ;;
     push)
-        docker login ghcr.io -u materialize-bot --password "$GITHUB_GHCR_TOKEN"
+        "$DOCKER" login ghcr.io -u materialize-bot --password "$GITHUB_GHCR_TOKEN"
         build --push "$@"
         ;;
     run)
@@ -274,6 +281,7 @@ case "$cmd" in
                 --env AZURE_SERVICE_ACCOUNT_PASSWORD
                 --env AZURE_SERVICE_ACCOUNT_TENANT
                 --env GCP_SERVICE_ACCOUNT_JSON
+                --env ANTITHESIS_GCP_SERVICE_ACCOUNT_JSON
                 --env GITHUB_TOKEN
                 --env GITHUB_GHCR_TOKEN
                 --env GPG_KEY
@@ -372,20 +380,26 @@ case "$cmd" in
             )
         fi
         if [[ "$(uname -s)" = Linux ]]; then
-            args+=(
-                --user "$(id -u):$(stat -c %g /var/run/docker.sock)"
-            )
+            if [[ "${MZ_DEV_CI_BUILDER_RUNTIME:-docker}" == "podman" ]]; then
+                args+=(--userns=keep-id)
+            else
+                args+=(
+                    --user "$(id -u):$(stat -c %g /var/run/docker.sock)"
+                )
+            fi
 
             if [[ $secrets == "true" ]]; then
                 # Allow Docker-in-Docker by mounting the Docker socket in the
                 # container. Host networking allows us to see ports created by
                 # containers that we launch.
                 args+=(
-                    --volume "/var/run/docker.sock:/var/run/docker.sock"
                     --network host
                     --env "DOCKER_TLS_VERIFY=${DOCKER_TLS_VERIFY-}"
                     --env "DOCKER_HOST=${DOCKER_HOST-}"
                 )
+                if [[ -S /var/run/docker.sock ]]; then
+                    args+=(--volume "/var/run/docker.sock:/var/run/docker.sock")
+                fi
 
                 # Forward Docker configuration too, if available.
                 docker_dir=${DOCKER_CONFIG:-$HOME/.docker}
@@ -431,14 +445,22 @@ case "$cmd" in
         image="$image_registry/ci-builder:$tag"
         # Try downloading the image a few times in case of registry flakiness
         if [[ "${CI:-}" ]]; then
-            if ! docker inspect "$image" > /dev/null 2>&1; then
-                docker pull "$image" || (sleep 3 && docker pull "$image") || (sleep 3 && docker pull "$image") || sleep 3
+            if ! "$DOCKER" inspect "$image" > /dev/null 2>&1; then
+                "$DOCKER" pull "$image" || (sleep 3 && "$DOCKER" pull "$image") || (sleep 3 && "$DOCKER" pull "$image") || sleep 3
             fi
         fi
-        docker run "${args[@]}" "$image" eatmydata "${docker_command[@]}"
+        if [[ "$DOCKER" == "podman" ]]; then
+            # --userns=keep-id already maps the host UID/GID into the
+            # container, so autouseradd is unnecessary.  Override the
+            # entrypoint to skip it.
+            args+=(--entrypoint eatmydata)
+            "$DOCKER" run "${args[@]}" "$image" "${docker_command[@]}"
+        else
+            "$DOCKER" run "${args[@]}" "$image" eatmydata "${docker_command[@]}"
+        fi
         ;;
     root-shell)
-        docker exec --interactive --tty --user 0:0 "$(<"$cid_file")" eatmydata ci/builder/root-shell.sh
+        "$DOCKER" exec --interactive --tty --user 0:0 "$(<"$cid_file")" eatmydata ci/builder/root-shell.sh
         ;;
     *)
         printf "unknown command %q\n" "$cmd"

ci/builder/Dockerfile

@@ -399,6 +399,11 @@ ENV CARGO_HOME=/cargo
 RUN mkdir /cargo && chmod 777 /cargo
 VOLUME /cargo
 
+# Antithesis coverage instrumentation library (used when --antithesis is passed)
+RUN curl -sSL https://antithesis.com/assets/instrumentation/libvoidstar.so \
+      -o /usr/lib/libvoidstar.so \
+    && ldconfig
+
 # Stage 3: Build a lightweight CI Builder image for console/playwright jobs.
 FROM ubuntu:noble-20260324 AS ci-builder-console
 

ci/mkpipeline.py

@@ -121,6 +121,12 @@ def main() -> int:
         type=Sanitizer,
         choices=Sanitizer,
     )
+    parser.add_argument(
+        "--antithesis",
+        action="store_true",
+        default=ui.env_is_truthy("CI_ANTITHESIS"),
+        help="enable Antithesis coverage instrumentation",
+    )
     parser.add_argument(
         "--priority",
         type=int,
@@ -166,6 +172,7 @@ def get_hashes(arch: Arch) -> tuple[str, bool]:
             arch=arch,
             coverage=args.coverage,
             sanitizer=args.sanitizer,
+            antithesis=args.antithesis,
         )
         deps = repo.resolve_dependencies(image for image in repo if image.publish)
         check = deps.check()
@@ -209,6 +216,7 @@ def fetch_hashes() -> None:
                 args.coverage,
                 args.sanitizer,
                 lto,
+                args.antithesis,
             )
             trim_ci_glue_exempt_steps(pipeline)
         else:
@@ -218,9 +226,11 @@ def fetch_hashes() -> None:
                 args.coverage,
                 args.sanitizer,
                 lto,
+                args.antithesis,
             )
     truncate_skip_length(pipeline)
     handle_sanitizer_skip(pipeline, args.sanitizer)
+    handle_antithesis_skip(pipeline, args.antithesis)
     increase_agents_timeouts(pipeline, args.sanitizer, args.coverage)
     prioritize_pipeline(pipeline, args.priority)
     switch_jobs_to_aws(pipeline, args.priority)
@@ -240,6 +250,7 @@ def fetch_hashes() -> None:
         args.coverage,
         args.sanitizer,
         lto,
+        args.antithesis,
     )
     add_nightly_deploy_dependency(pipeline, args.pipeline)
     remove_dependencies_on_prs(pipeline, args.pipeline, hash_check)
@@ -328,6 +339,21 @@ def handle_sanitizer_skip(pipeline: Any, sanitizer: Sanitizer) -> None:
                 step["skip"] = True
 
 
+def handle_antithesis_skip(pipeline: Any, antithesis: bool) -> None:
+    if antithesis:
+        pipeline.setdefault("env", {})["CI_ANTITHESIS"] = "1"
+
+        for step in steps(pipeline):
+            if step.get("antithesis") == "skip":
+                step["skip"] = True
+
+    else:
+
+        for step in steps(pipeline):
+            if step.get("antithesis") == "only":
+                step["skip"] = True
+
+
 def increase_agents_timeouts(
     pipeline: Any, sanitizer: Sanitizer, coverage: bool
 ) -> None:
@@ -711,6 +737,7 @@ def trim_tests_pipeline(
     coverage: bool,
     sanitizer: Sanitizer,
     lto: bool,
+    antithesis: bool = False,
 ) -> None:
     """Trim pipeline steps whose inputs have not changed in this branch.
 
@@ -731,6 +758,7 @@ def trim_tests_pipeline(
         profile=mzbuild.Profile.RELEASE if lto else mzbuild.Profile.OPTIMIZED,
         coverage=coverage,
         sanitizer=sanitizer,
+        antithesis=antithesis,
     )
     deps = repo.resolve_dependencies(image for image in repo)
 
@@ -917,6 +945,7 @@ def add_cargo_test_dependency(
     coverage: bool,
     sanitizer: Sanitizer,
     lto: bool,
+    antithesis: bool = False,
 ) -> None:
     """Cargo Test normally doesn't have to wait for the build to complete, but it requires a few images (ubuntu-base, postgres), which are rarely changed. So only add a dependency when those images are not on Dockerhub yet."""
     if pipeline_name not in ("test", "nightly"):
@@ -933,6 +962,7 @@ def add_cargo_test_dependency(
         profile=mzbuild.Profile.RELEASE if lto else mzbuild.Profile.OPTIMIZED,
         coverage=coverage,
         sanitizer=sanitizer,
+        antithesis=antithesis,
     )
     composition = Composition(repo, name="cargo-test")
     deps = composition.dependencies
@@ -1090,6 +1120,8 @@ def remove_mz_specific_keys(pipeline: Any) -> None:
             del step["coverage"]
         if "sanitizer" in step:
             del step["sanitizer"]
+        if "antithesis" in step:
+            del step["antithesis"]
         if "ci_glue_exempt" in step:
             del step["ci_glue_exempt"]
         if (

ci/nightly/pipeline.template.yml

@@ -65,6 +65,29 @@ steps:
         branches: "main"
         skip: "currently broken"
 
+      - id: build-x86_64-antithesis
+        label: ":rust: Build x86_64 (Antithesis)"
+        # Regenerate the antithesis compose YAML before building so the
+        # `antithesis-config` image's fingerprint captures the same
+        # materialized fingerprint we're about to publish — otherwise
+        # Antithesis would try to pull a stale `materialized:mzbuild-…`
+        # whenever the committed YAML lagged behind source changes.
+        command: bin/ci-builder run stable ci/test/build-antithesis.sh
+        inputs:
+          - "*"
+        depends_on: []
+        timeout_in_minutes: 90
+        agents:
+          queue: l-builder-linux-x86_64
+        env:
+          CI_ANTITHESIS: "1"
+        # Antithesis-flavored images get distinct mzbuild fingerprints, so
+        # they coexist with regular GHCR tags. The build is x86_64-only —
+        # Antithesis runs amd64 sandboxes.
+        sanitizer: skip
+        coverage: skip
+        antithesis: skip
+
       - id: build-rust-latest-beta
         label: "Build with Latest Rust Beta"
         command: bin/ci-builder run stable ci/test/rust-beta-build.sh