build(backend): cache torch wheel + pip downloads across rebuilds

Splits torch+torchvision into its own RUN layer so edits to
requirements.txt don't invalidate the ~200MB CPU-only torch download.
Adds buildkit cache mounts on both pip-install layers so even when a
layer is invalidated (or buildkit evicts it) the wheel is reused from
the on-disk pip cache instead of refetching from download.pytorch.org.

Triggered by two consecutive deploy failures where pytorch.org timed
out mid-download (2026-05-13 ~21:41 and possibly ~22:47).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
claudio
2026-05-13 23:08:10 +02:00
parent 4e1af0f356
commit 6915c30911

View File

@@ -1,3 +1,4 @@
# syntax=docker/dockerfile:1.7
FROM python:3.12-slim
# Install system dependencies
@@ -22,13 +23,23 @@ RUN apt-get update && apt-get install -y \
WORKDIR /app
# Copy requirements first for better caching
# Install PyTorch CPU-only FIRST, in its own layer, so open-clip-torch
# doesn't pull the full CUDA build (~7 GB). CPU inference is all we need
# — the heavy lifting happens through ONNX Runtime.
#
# Two cache wins here:
# 1. Its own RUN layer means edits to requirements.txt don't force a
# re-pull of the ~200MB torch wheel.
# 2. The buildkit cache mount keeps pip's download cache on disk
# across builds even when the layer itself is invalidated, so a
# torch-version bump or a builder cache eviction still reuses the
# wheel from local cache instead of re-fetching from pytorch.org.
RUN --mount=type=cache,target=/root/.cache/pip \
pip install torch torchvision --index-url https://download.pytorch.org/whl/cpu
COPY requirements.txt .
# Install PyTorch CPU-only FIRST so open-clip-torch doesn't pull the full
# CUDA build (~7 GB). CPU inference is all we need — the heavy lifting
# happens through ONNX Runtime.
RUN pip install --no-cache-dir torch torchvision --index-url https://download.pytorch.org/whl/cpu \
&& pip install --no-cache-dir -r requirements.txt
RUN --mount=type=cache,target=/root/.cache/pip \
pip install -r requirements.txt
# Copy application code
COPY . .