From d344dc454e457286bef9789ec3731a33b9686300 Mon Sep 17 00:00:00 2001
From: Waleed Latif <walif6@gmail.com>
Date: Mon, 26 Oct 2026 16:54:08 -0700
Subject: [PATCH] chore(docker): pin Node and prep images for a read-only root
 filesystem

---
 docker/app.Dockerfile      | 26 +++++++++++++++++++++++---
 docker/pii.Dockerfile      |  9 +++++++++
 docker/realtime.Dockerfile |  5 +++++
 3 files changed, 37 insertions(+), 3 deletions(-)

diff --git a/docker/app.Dockerfile b/docker/app.Dockerfile
index 6a67f2332fd..65e8395a667 100644
--- a/docker/app.Dockerfile
+++ b/docker/app.Dockerfile
@@ -9,6 +9,14 @@ FROM oven/bun:1.4.4-slim AS base
 # `isolated-vm` pin in apps/sim/package.json — Node 24 (ABI 137) requires
 # isolated-vm 6.x.
 #
+# NODE_VERSION pins the exact NodeSource build: this layer is cached, so an
+# unpinned `nodejs` changed only when an unrelated edit forced a rebuild. The
+# `-1nodesource1` suffix selects the `nodejs` package, never `nsolid`, which
+# Provides `nodejs` at the same version numbers. build-base and deps inherit
+# this stage, so isolated-vm's node-gyp build targets the same Node (and ABI)
+# the runner executes it with.
+ARG NODE_VERSION=24.22.1
+#
 # Only what the running container needs belongs here. ffmpeg backs the media
 # subprocess adapter; python3 is the node-gyp interpreter and is kept because
 # build-base inherits from this stage. The compiler toolchain lives in build-base
@@ -24,7 +32,8 @@ RUN --mount=type=cache,target=/var/cache/apt,sharing=locked \
     apt-get update && apt-get install -y --no-install-recommends \
     python3 curl ca-certificates bash ffmpeg \
     && curl -fsSL https://deb.nodesource.com/setup_24.x | bash - \
-    && apt-get install -y nodejs \
+    && apt-get install -y --no-install-recommends "nodejs=${NODE_VERSION}-1nodesource1" \
+    && [ "$(node --version)" = "v${NODE_VERSION}" ] \
     && sed -i -E 's#^mozilla/(emSign_ECC_Root_CA_-_G3|Comodo_AAA_Services_root)\.crt$#!&#' /etc/ca-certificates.conf \
     && update-ca-certificates \
     && ! grep -qF -e 'l2D4kF501KKaU73yqWjgom7C12yxow+ev+to51byrvLjKzg6CYG1a4XXvi3tPxq3' -e 'CUfvO6wIBHxcmbHtRwfSAjEAnbpV/KlK6O3t5nYBQnvI+GDZjVGLVTv7jHvrZQnD' /etc/ssl/certs/ca-certificates.crt
@@ -234,9 +243,20 @@ COPY --from=builder --chown=nextjs:nodejs /app/apps/sim/lib/execution/sandbox/bu
 # anonymizer, docker/pii.Dockerfile), reached over the network via PII_URL —
 # no Python/Presidio in this image.
 
-# Create .next/cache directory with correct ownership
+# Writable paths for a read-only root filesystem. Declaring them as volumes
+# keeps writes working when the root is mounted read-only: ECS copies the
+# image's contents and ownership into a bind volume mounted at a VOLUME path.
+# - /tmp (mode 1777): os.tmpdir() scratch for ffmpeg and audio processing,
+#   knowledge base imports, the Pi agent, and libvips (sharp) spill files.
+# - apps/sim/.next/cache: the Next image optimizer's disk cache
+#   (.next/cache/images) and the data cache for `fetch` calls with
+#   `next.revalidate`.
+# Keep the VOLUME line after every step that writes to these paths: the
+# classic builder discards later build-time changes to a declared volume.
 RUN mkdir -p apps/sim/.next/cache && \
-    chown -R nextjs:nodejs apps/sim/.next/cache
+    chown -R nextjs:nodejs apps/sim/.next/cache && \
+    chmod 1777 /tmp
+VOLUME ["/tmp", "/app/apps/sim/.next/cache"]
 
 # Switch to non-root user
 USER nextjs
diff --git a/docker/pii.Dockerfile b/docker/pii.Dockerfile
index 677fd9c1b6f..799c1115c6c 100644
--- a/docker/pii.Dockerfile
+++ b/docker/pii.Dockerfile
@@ -41,6 +41,15 @@ RUN groupadd -g 1001 pii && \
 COPY --chown=pii:pii apps/pii/server.py ./
 COPY --chown=pii:pii --chmod=755 docker/pii-entrypoint.sh ./
 
+# Read-only root filesystem prep. Python skips writing __pycache__ next to
+# server.py (site-packages is already byte-compiled by pip at install time),
+# and /tmp (mode 1777), Python's tempfile scratch, stays writable because ECS
+# copies the image's /tmp into a bind volume mounted at a VOLUME path. The
+# VOLUME line stays below the model download, which stages wheels in /tmp.
+ENV PYTHONDONTWRITEBYTECODE=1
+RUN chmod 1777 /tmp
+VOLUME ["/tmp"]
+
 USER pii
 
 # Listen on 5001. Runs as its own ECS service (separate task), reached via PII_URL;
diff --git a/docker/realtime.Dockerfile b/docker/realtime.Dockerfile
index 02c1b23efd6..1b3631c184a 100644
--- a/docker/realtime.Dockerfile
+++ b/docker/realtime.Dockerfile
@@ -44,6 +44,11 @@ RUN addgroup -g 1001 -S nodejs && \
 COPY --from=deps --chown=nextjs:nodejs /app ./
 COPY --from=pruner --chown=nextjs:nodejs /app/out/full/ ./
 
+# /tmp stays writable under a read-only root filesystem: ECS copies the image's
+# /tmp (mode 1777) into a bind volume mounted at a VOLUME path.
+RUN chmod 1777 /tmp
+VOLUME ["/tmp"]
+
 USER nextjs
 
 EXPOSE 3002