Squash Odysseus development history

2026-09-17 13:42:21 +02:00 · 2026-09-11 06:04:19 +00:00
parent e5c99a5eee
commit 6ee6502010
2050 changed files with 538359 additions and 57745 deletions
@@ -0,0 +1,70 @@
+#!/usr/bin/env bash
+# Build patched wheels for Real-ESRGAN's unmaintained dependencies.
+#
+# basicsr / gfpgan / facexlib (xinntao, last released 2022) read their version
+# in setup.py with:
+#
+#     exec(compile(f.read(), version_file, 'exec'))
+#     return locals()['__version__']
+#
+# Python 3.13+ implements PEP 667: locals() inside a function returns an
+# independent snapshot that exec() can no longer mutate, so the read raises
+# `KeyError: '__version__'` and the sdist build fails. That is why the Cookbook
+# "install realesrgan" button dies on the python:3.14 image. The packages have
+# no fixed release, so we patch get_version() to exec into an explicit namespace
+# dict (works on every Python) and build wheels from the patched source.
+#
+# Usage: build-realesrgan-wheels.sh [OUTPUT_DIR]   (default: /wheels)
+set -euo pipefail
+
+OUT="${1:-/wheels}"
+mkdir -p "$OUT"
+
+work="$(mktemp -d)"
+trap 'rm -rf "$work"' EXIT
+cd "$work"
+
+# Pinned to the versions Real-ESRGAN 0.3.0 resolves to.
+SPECS="basicsr==1.4.2 gfpgan==1.3.8 facexlib==0.3.0"
+
+for spec in $SPECS; do
+  name="${spec%%==*}"
+  ver="${spec##*==}"
+  # pip download builds metadata (and trips the same bug), so fetch the raw
+  # sdist URL from the PyPI JSON API instead.
+  url="$(python - "$name" "$ver" <<'PY'
+import json, sys, urllib.request
+name, ver = sys.argv[1], sys.argv[2]
+data = json.load(urllib.request.urlopen(f"https://pypi.org/pypi/{name}/{ver}/json"))
+for f in data["urls"]:
+    if f["packagetype"] == "sdist":
+        print(f["url"]); break
+else:
+    sys.exit(f"no sdist found for {name}=={ver}")
+PY
+)"
+  echo ">> fetching ${name} ${ver}: ${url}"
+  curl -fsSL "$url" -o "${name}.tar.gz"
+  tar xzf "${name}.tar.gz"
+done
+
+echo ">> patching get_version()"
+python - <<'PY'
+import pathlib
+old_exec = "exec(compile(f.read(), version_file, 'exec'))"
+new_exec = "_ver_ns = {}\n        exec(compile(f.read(), version_file, 'exec'), _ver_ns)"
+old_ret = "return locals()['__version__']"
+new_ret = "return _ver_ns['__version__']"
+patched = 0
+for setup in pathlib.Path(".").glob("*/setup.py"):
+    s = setup.read_text()
+    if old_exec in s and old_ret in s:
+        setup.write_text(s.replace(old_exec, new_exec).replace(old_ret, new_ret))
+        print("   patched", setup)
+        patched += 1
+assert patched == 3, f"expected to patch 3 setup.py files, patched {patched}"
+PY
+
+echo ">> building wheels into ${OUT}"
+pip wheel --no-deps -w "$OUT" ./basicsr-* ./gfpgan-* ./facexlib-*
+ls -l "$OUT"
@@ -13,6 +13,8 @@ set -e

 PUID="${PUID:-1000}"
 PGID="${PGID:-1000}"
+GOSU_BIN="$(command -v gosu)"
+PYTHON_BIN="$(command -v python)"

 # Reuse an existing matching group/user if the host's UID/GID already
 # corresponds to one in /etc/passwd (e.g. when the image is rebuilt
@@ -24,29 +26,121 @@ if ! getent passwd "$PUID" >/dev/null 2>&1; then
    useradd -u "$PUID" -g "$PGID" -M -s /bin/sh -d /app odysseus
 fi

-# Repair ownership on every writable path the app touches at runtime.
-#
-# Bind-mounted dirs (/app/data, /app/logs) are the obvious ones, but
-# the app ALSO writes inside the image's own source tree at runtime:
-#   - services/cache/{search,content}/*  (search cache LRU)
-#   - services/search_analytics.json
-#   - services/search_engine_error.log
-#   - services/tts cache, etc.
-# These dirs were created as root during `docker build`, so dropping
-# to PUID:PGID would otherwise crash on the first import that tries
-# to mkdir them. Chown the whole /app tree — fast (<1s on this size)
-# and idempotent via the `-not -uid` filter so we only touch files
-# that need fixing.
-for dir in /app /app/data /app/logs; do
+ODY_USER="$(getent passwd "$PUID" | cut -d: -f1)"
+[ -z "$ODY_USER" ] && ODY_USER=odysseus
+
+# Docker-socket group plumbing for the explicit host-Docker overlay. When
+# opted in, the socket is owned by root:<host docker gid>. Add the app user
+# to that group and later call gosu by username so supplementary groups are
+# retained.
+DOCKER_SOCK="${DOCKER_SOCK:-/var/run/docker.sock}"
+if [ "${ODYSSEUS_ENABLE_HOST_DOCKER:-}" = "true" ] && [ -S "$DOCKER_SOCK" ]; then
+    SOCK_GID="$(stat -c '%g' "$DOCKER_SOCK" 2>/dev/null || echo '')"
+    if [ -n "$SOCK_GID" ] && [ "$SOCK_GID" != "0" ]; then
+        if ! getent group "$SOCK_GID" >/dev/null 2>&1; then
+            groupadd -g "$SOCK_GID" docker_host || true
+        fi
+        SOCK_GROUP="$(getent group "$SOCK_GID" | cut -d: -f1)"
+        if [ -n "$SOCK_GROUP" ]; then
+            usermod -aG "$SOCK_GROUP" "$ODY_USER" 2>/dev/null || true
+        fi
+    fi
+fi
+
+mount_root_for() {
+    awk -v target="$1" '$5 == target { print $4; exit }' /proc/self/mountinfo 2>/dev/null || true
+}
+
+is_broad_mount_root() {
+    case "$1" in
+        /|/home|/srv|/var|/usr|/opt|/tmp|/mnt|/media)
+            return 0
+            ;;
+    esac
+    return 1
+}
+
+repair_tree_ownership() {
+    dir="$1"
    if [ -d "$dir" ]; then
-        # `find ... -not -uid` keeps this O(touched-files), not
-        # O(everything), so terabyte-sized maildirs don't slow startup.
-        find "$dir" -not -uid "$PUID" -print0 2>/dev/null \
+        find "$dir" -xdev -not -uid "$PUID" -print0 2>/dev/null \
            | xargs -0 -r chown "$PUID:$PGID" 2>/dev/null || true
    fi
+}
+
+repair_app_tree_ownership() {
+    if [ -d /app ]; then
+        find /app -xdev \
+            \( -path /app/data -o -path /app/logs -o -path /app/.ssh -o -path /app/.cache -o -path /app/.local \) -prune \
+            -o -not -uid "$PUID" -print0 2>/dev/null \
+            | xargs -0 -r chown "$PUID:$PGID" 2>/dev/null || true
+    fi
+}
+
+repair_bind_mount_ownership() {
+    dir="$1"
+    if [ ! -d "$dir" ]; then
+        return
+    fi
+
+    mount_root="$(mount_root_for "$dir")"
+    if is_broad_mount_root "$mount_root"; then
+        echo "Skipping recursive ownership repair for $dir because it maps to broad host path $mount_root" >&2
+        chown "$PUID:$PGID" "$dir" 2>/dev/null || true
+        return
+    fi
+
+    repair_tree_ownership "$dir"
+}
+
+# Repair image-owned writable paths without walking into bind-mounted host
+# trees, then repair the app-owned mount roots separately.
+repair_app_tree_ownership
+for dir in /app/data /app/logs /app/.ssh /app/.cache/huggingface /app/.local; do
+    repair_bind_mount_ownership "$dir"
 done

+# Cookbook installs vllm/etc. via `pip install --user`, which pulls
+# nvidia-cuda-* wheels into /app/.local but does not set CUDA_HOME or
+# symlink /usr/local/cuda. vllm 0.22+ then crashes during engine init
+# when FlashInfer tries to JIT a sampler kernel ("Could not find nvcc",
+# then "CUDA compiler and toolkit headers are incompatible" on the
+# mixed cuda-nvcc 13.3 / cuda-runtime 13.0 wheel combo).
+#
+# Auto-set CUDA_HOME if a pip-installed nvcc is present, and disable the
+# FlashInfer JIT sampler — sampler only, no impact on attention path.
+# No-op when vllm isn't installed.
+#
+# Checked layouts (all are real pip-wheel install paths):
+#   nvidia/cu13        — nvidia-nvcc-cu13 (CUDA 13.x wheel style)
+#   nvidia/cu12        — nvidia-nvcc-cu12 (CUDA 12.x wheel style)
+#   nvidia/cuda_nvcc   — nvidia-cuda-nvcc-cu12 (older cu12 sub-package style)
+for cu in \
+    /app/.local/lib/python*/site-packages/nvidia/cu13 \
+    /app/.local/lib/python*/site-packages/nvidia/cu12 \
+    /app/.local/lib/python*/site-packages/nvidia/cuda_nvcc; do
+    if [ -x "$cu/bin/nvcc" ]; then
+        export CUDA_HOME="$cu"
+        break
+    fi
+done
+
+# Disable the FlashInfer JIT sampler unconditionally — it is sampler-only
+# and has no impact on the attention path, but requires nvcc + matching
+# CUDA headers at startup. Without this, vLLM crashes with "Could not find
+# nvcc" even when the GPU itself is fully visible to the container.
+export VLLM_USE_FLASHINFER_SAMPLER="${VLLM_USE_FLASHINFER_SAMPLER:-0}"
+
+# Make Cookbook-installed Python CLIs visible after `pip install --user`.
+# vLLM and helper scripts land here because /app is the non-root user's HOME.
+export PATH="/app/.local/bin:$PATH"
+
+# Run first-time setup as the app user so data/ files get the right ownership.
+# setup.py is idempotent — skips auth.json / .env if they already exist.
+# || true so a setup failure never prevents the container from starting.
+"$GOSU_BIN" "$ODY_USER" "$PYTHON_BIN" /app/setup.py || true
+
 # Drop root and run the actual app. `gosu` is preferred over `su` /
 # `sudo` because it cleans up the process tree (no extra shell layer)
 # so signals (SIGTERM from `docker stop`) reach uvicorn directly.
-exec gosu "$PUID:$PGID" "$@"
+exec "$GOSU_BIN" "$ODY_USER" "$@"
@@ -0,0 +1,19 @@
+# AMD ROCm GPU overlay. Enable by setting COMPOSE_FILE in .env:
+#   COMPOSE_FILE=docker-compose.yml:docker/gpu.amd.yml
+#   RENDER_GID=<numeric output of: getent group render | cut -d: -f3>
+#
+# Requires ROCm drivers on the host (kfd + DRI devices). The host user
+# running Docker must be in the `video` and `render` groups.
+#
+# This overlay only passes the host GPU through to the container.
+# The slim Odysseus image does not bundle ROCm userspace or inference
+# engines — install ROCm-compatible builds of vLLM / llama-cpp-python
+# via Cookbook -> Dependencies (or pip) before serving GPU models.
+services:
+  odysseus:
+    devices:
+      - /dev/kfd
+      - /dev/dri
+    group_add:
+      - video
+      - ${RENDER_GID:-render}
@@ -0,0 +1,34 @@
+# NVIDIA GPU overlay. Enable by setting COMPOSE_FILE in .env:
+#   COMPOSE_FILE=docker-compose.yml:docker/gpu.nvidia.yml
+#
+# Use scripts/check-docker-gpu.sh to diagnose GPU passthrough, optionally
+# install the NVIDIA Container Toolkit (Ubuntu/Debian), and write COMPOSE_FILE
+# to .env. The script is read-only by default — it installs nothing and never
+# edits .env unless explicitly asked.
+#
+# Requires the NVIDIA Container Toolkit on the host.
+#   Arch:    sudo pacman -S nvidia-container-toolkit
+#   Debian:  sudo apt install nvidia-container-toolkit
+#   Fedora:  sudo dnf install nvidia-container-toolkit
+# Then:
+#   sudo nvidia-ctk runtime configure --runtime=docker
+#   sudo systemctl restart docker
+# Verify with:
+#   docker info | grep -i nvidia
+#
+# This overlay only passes the host GPU through to the container.
+# The slim Odysseus image does not bundle CUDA userspace or inference
+# engines — install vLLM / llama-cpp-python / SGLang via
+# Cookbook -> Dependencies (or pip) before serving GPU models.
+services:
+  odysseus:
+    environment:
+      - NVIDIA_VISIBLE_DEVICES=all
+      - NVIDIA_DRIVER_CAPABILITIES=compute,utility
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - driver: nvidia
+              count: all
+              capabilities: [gpu]
@@ -0,0 +1,12 @@
+# High-trust host Docker access. Enable only when local Docker-daemon
+# management from Cookbook is required and you accept that raw socket access
+# grants broad control over the host Docker daemon.
+#   COMPOSE_FILE=docker-compose.yml:docker/host-docker.yml
+#   DOCKER_GID=<numeric host Docker group id>
+services:
+  odysseus:
+    volumes:
+      - /var/run/docker.sock:/var/run/docker.sock
+    group_add: ["${DOCKER_GID:-963}"]
+    environment:
+      - ODYSSEUS_ENABLE_HOST_DOCKER=true
@@ -0,0 +1,21 @@
+# High-trust host network access. Enable only when the Odysseus agent needs
+# host-native LAN/VPN/mDNS behavior that Docker bridge networking cannot
+# provide. Linux only; Docker Desktop does not provide equivalent host
+# networking semantics.
+#   COMPOSE_FILE=docker-compose.yml:docker/host-workspace.yml:docker/host-network.yml
+#   APP_PORT=7011
+services:
+  odysseus:
+    network_mode: host
+    ports: !reset []
+    environment:
+      - APP_PORT=${APP_PORT:-7011}
+      - APP_BIND=${APP_BIND:-0.0.0.0}
+      - SEARXNG_INSTANCE=${ODYSSEUS_HOST_NETWORK_SEARXNG_INSTANCE:-http://127.0.0.1:8080}
+      - CHROMADB_HOST=${ODYSSEUS_HOST_NETWORK_CHROMADB_HOST:-127.0.0.1}
+      - CHROMADB_PORT=${ODYSSEUS_HOST_NETWORK_CHROMADB_PORT:-8100}
+      - ODYSSEUS_CONTAINER_NETWORK_MODE=host
+    command:
+      - sh
+      - -c
+      - exec uvicorn app:app --host "$${APP_BIND:-0.0.0.0}" --port "$${APP_PORT:-7011}"
@@ -0,0 +1,11 @@
+# High-trust host workspace access. Enable only when the Odysseus agent should
+# work on a host directory outside the container's normal /app/data sandbox.
+#   COMPOSE_FILE=docker-compose.yml:docker/host-workspace.yml
+#   ODYSSEUS_HOST_WORKSPACE_DIR=/absolute/host/path
+#   ODYSSEUS_HOST_WORKSPACE_MOUNT=/host/workspace
+services:
+  odysseus:
+    volumes:
+      - ${ODYSSEUS_HOST_WORKSPACE_DIR:?set ODYSSEUS_HOST_WORKSPACE_DIR}:${ODYSSEUS_HOST_WORKSPACE_MOUNT:-/host/workspace}:rw,z
+    environment:
+      - ODYSSEUS_HOST_WORKSPACE_MOUNT=${ODYSSEUS_HOST_WORKSPACE_MOUNT:-/host/workspace}