Build sim image via install-sim.sh, drop GHA buildx cache

.github/workflows/build-sim-image.yml

@@ -73,9 +73,8 @@ jobs:
     runs-on: ${{ matrix.runner }}
 
     steps:
-      # ``linux/amd64`` → ``linux-amd64``. Used as a key for the digest
-      # artifact and the buildx cache scope so different platforms
-      # don't trample each other's GHA cache.
+      # ``linux/amd64`` → ``linux-amd64``. Used as the key for the
+      # digest artifact so different platforms don't collide.
       - name: Sanitise platform name
         id: plat
         run: |
@@ -113,8 +112,6 @@ jobs:
           file: scripts/sim.Dockerfile
           platforms: ${{ matrix.platform }}
           push: false
-          cache-from: type=gha,scope=${{ steps.plat.outputs.key }}
-          cache-to: type=gha,mode=max,scope=${{ steps.plat.outputs.key }}
 
       # main / dispatch: push by digest. The image gets uploaded to the
       # registry under its content-addressable digest; no tag is set
@@ -128,8 +125,6 @@ jobs:
           file: scripts/sim.Dockerfile
           platforms: ${{ matrix.platform }}
           outputs: type=image,name=${{ steps.img.outputs.ref }},push-by-digest=true,name-canonical=true,push=true
-          cache-from: type=gha,scope=${{ steps.plat.outputs.key }}
-          cache-to: type=gha,mode=max,scope=${{ steps.plat.outputs.key }}
 
       - name: Stash digest for the merge job
         if: github.event_name != 'pull_request'

scripts/sim.Dockerfile

@@ -1,103 +1,44 @@
-# LLMServingSim simulator image (slim, multi-stage).
+# LLMServingSim simulator image.
 #
 # Built by .github/workflows/build-sim-image.yml on every push to main
 # that touches scripts/install-sim.sh, the astra-sim submodule, or this
 # file. Published to ghcr.io/<owner>/<repo>/sim.
 #
-# Stage 1 (builder) carries the C++ toolchain to compile ASTRA-Sim;
-# Stage 2 (runtime) carries only the python interpreter, the
-# simulator's runtime python deps, the compiled ASTRA-Sim binary, the
-# chakra package, the repo source, and the ``.git/`` tree (kept so
-# users can ``git log`` / ``git rev-parse`` / ``git submodule status``
-# inside a running container to identify the exact revision they're
-# debugging against). .dockerignore at the repo root strips
-# perf/results/venv/CMake artefacts from the context so neither stage
-# sees them.
+# Single stage: install-sim.sh does the full setup (apt deps, python
+# deps, ASTRA-Sim build, chakra package), so this Dockerfile just
+# bootstraps a base image and hands off. The multi-stage / pip-inline
+# version was hard to reason about when the GHA buildx cache served
+# stale layers; replacing it with a single ``RUN install-sim.sh`` keeps
+# the image build path identical to the bare-metal install path.
 #
-# install-sim.sh also pulls workload-generator / bench / power-model
-# training deps (transformers, datasets, scikit-learn, xgboost,
-# matplotlib). Those are NOT installed here: the sim image is for
-# `python -m serving` only, and workload generation lives in the vLLM
-# docker image (scripts/docker-vllm.sh).
+# The ``.git/`` tree (main repo + submodules) is intentionally KEPT in
+# the image so users can ``git log`` / ``git rev-parse`` / ``git
+# submodule status`` inside a running container to identify the exact
+# revision they're debugging against. ``git gc`` at the end compacts
+# the packs so the size impact stays reasonable, and
+# ``safe.directory=*`` is set system-wide because the .git/ trees were
+# created during the build as root and may now be inspected by a
+# different uid under enroot / docker --user.
 #
 # Local build:
 #   docker build -f scripts/sim.Dockerfile -t llmservingsim-sim .
 
-# ============================================================================
-# Stage 1: builder — toolchain for ASTRA-Sim native build
-# ============================================================================
-FROM ubuntu:24.04 AS builder
-
-ENV DEBIAN_FRONTEND=noninteractive
-
-RUN apt-get update -qq \
- && apt-get install -y -qq --no-install-recommends \
-        git ca-certificates \
-        build-essential cmake \
-        protobuf-compiler libprotobuf-dev \
- && rm -rf /var/lib/apt/lists/*
-
-# Same WORKDIR (/workspace) as the runtime stage. ASTRA-Sim's build.sh
-# creates legacy-compat symlinks (e.g. AnalyticalAstra/bin/AnalyticalAstra
-# → bin/AstraSim_Analytical_Congestion_Unaware) using ABSOLUTE paths
-# rooted at ${BUILD_DIR}. If the builder lives at /build but the
-# runtime stage mounts the tree at /workspace, those absolute symlinks
-# become dangling after the COPY — which broke
-# serving/__main__.py's hard-coded legacy binary path on both archs.
-# Keep the path identical so the symlinks resolve in both stages.
-WORKDIR /workspace
-COPY . /workspace
-
-# Init submodules, capture the commit SHA for runtime inspection,
-# build the analytical backend, then prune everything we don't need at
-# runtime: CMake build trees, downloaded _deps, object files,
-# __pycache__. ``.git`` trees are KEPT (main repo + submodules) so the
-# image can self-report its version via ``git rev-parse HEAD`` /
-# ``git submodule status``. ``git gc`` at the end compacts loose
-# objects + repacks so deep local clones don't bloat the image; CI
-# checkouts are already shallow.
-RUN git config --global --add safe.directory '*' \
- && git submodule update --init --recursive --depth 1 \
- && (git rev-parse HEAD 2>/dev/null || echo unknown) > /workspace/.git-rev \
- && bash astra-sim/build/astra_analytical/build.sh \
- && rm -rf astra-sim/build/astra_analytical/build/CMakeFiles \
-           astra-sim/build/astra_analytical/build/_deps \
-           astra-sim/build/astra_analytical/build/CMakeCache.txt \
-           astra-sim/build/astra_analytical/build/cmake_install.cmake \
-           astra-sim/build/astra_analytical/build/Makefile \
- && find astra-sim -name '*.o' -delete \
- && find astra-sim -name '*.a' -delete \
- && find . -type d -name __pycache__ -prune -exec rm -rf {} + \
- && git gc --quiet --prune=now --aggressive \
- && (git submodule foreach --recursive 'git gc --quiet --prune=now --aggressive' || true)
-
-# ============================================================================
-# Stage 2: runtime — slim, only what's needed to RUN the simulator
-# ============================================================================
 FROM ubuntu:24.04
 
 ENV DEBIAN_FRONTEND=noninteractive \
     PIP_BREAK_SYSTEM_PACKAGES=1 \
     PIP_NO_CACHE_DIR=1 \
     PIP_DISABLE_PIP_VERSION_CHECK=1
 
-# Runtime apt: python + pip + tls roots + git. No compilers, no
-# protobuf C++ runtime — the python protobuf wheel ships its own.
-# ``git`` is here so ``git rev-parse HEAD`` / ``git log`` / ``git
-# submodule status`` work against the bundled .git tree for debugging.
-# ``safe.directory=*`` because the .git/ trees were created during the
-# builder stage as root and may now be inspected by a different uid
-# under enroot / docker --user; without it git refuses with "dubious
-# ownership in repository".
-RUN apt-get update -qq \
- && apt-get install -y -qq --no-install-recommends \
-        python3 python3-pip python-is-python3 \
-        ca-certificates git \
- && rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/* \
- && git config --system --add safe.directory '*'
-
 WORKDIR /workspace
-COPY --from=builder /workspace /workspace
+COPY . /workspace
+
+RUN bash scripts/install-sim.sh \
+ && (git rev-parse HEAD 2>/dev/null || echo unknown) > /workspace/.git-rev \
+ && git gc --quiet --prune=now --aggressive \
+ && (git submodule foreach --recursive 'git gc --quiet --prune=now --aggressive' || true) \
+ && git config --system --add safe.directory '*' \
+ && find / -type d -name __pycache__ -prune -exec rm -rf {} + 2>/dev/null || true
 
 # Sanity-check that the legacy binary symlink resolves. ASTRA-Sim's
 # build.sh wires AnalyticalAstra/bin/AnalyticalAstra → the canonical
@@ -111,13 +52,4 @@ RUN ASTRA_BIN_LEGACY=/workspace/astra-sim/build/astra_analytical/build/Analytica
         exit 1; \
     }
 
-# Simulator runtime python deps. Narrower than scripts/install-sim.sh
-# on purpose — see the header comment.
-RUN pip3 install --quiet --no-input \
-        pyyaml pyinstrument msgspec \
-        pandas numpy rich protobuf \
- && pip3 install --quiet --no-input --no-deps \
-        /workspace/astra-sim/extern/graph_frontend/chakra \
- && find / -type d -name __pycache__ -prune -exec rm -rf {} + 2>/dev/null || true
-
 CMD ["/bin/bash"]