Containerfile

# Build llmproxy from source.
#
#   podman build -t llmproxy .
#
# Run with GCP Vertex AI (single-backend mode via env vars):
#
#   podman run --rm -p 8080:8080 \
#     -v $HOME/.config/gcloud:/gcp-creds:ro,z \
#     -e GOOGLE_APPLICATION_CREDENTIALS=/gcp-creds/application_default_credentials.json \
#     -e LLMPROXY_BACKEND=gcp-vertex \
#     -e GOOGLE_CLOUD_PROJECT=my-project \
#     -e VERTEX_LOCATION=us-east5 \
#     llmproxy
#
# Uses BuildKit cache mounts for fast incremental rebuilds (cargo
# target dir + registry survive across builds on the same host).
# Pattern borrowed from github.com/bootc-dev/bootc.

FROM registry.access.redhat.com/ubi10/ubi:latest AS builder

RUN dnf -y install cargo openssl-devel && dnf clean all

WORKDIR /build
COPY Cargo.toml Cargo.lock ./
COPY crates/ crates/

# Fetch dependencies (cached across builds via the registry mount).
RUN --mount=type=cache,target=/build/target \
    --mount=type=cache,target=/root/.cargo/registry \
    cargo fetch

# Build with cache mounts. The target dir persists across builds so
# incremental compilation kicks in. --network=none proves all deps
# were fetched above. Copy the binary out of the cache mount into a
# stable location before the layer closes.
RUN --network=none \
    --mount=type=cache,target=/build/target \
    --mount=type=cache,target=/root/.cargo/registry \
    cargo build --release && \
    cp target/release/llmproxy /usr/local/bin/llmproxy && \
    strip /usr/local/bin/llmproxy

# --- Integration test runner ---
#
# Build: podman build --target integration-runner -t llmproxy-integration .
# Run:   podman run --rm llmproxy-integration
#
# Compiles the integration test binary and bundles it with the proxy
# binary so tests can start the proxy and talk to it.
FROM builder AS integration-builder

RUN --mount=type=cache,target=/build/target \
    --mount=type=cache,target=/root/.cargo/registry \
    cargo test -p llmproxy-proxy --test integration --no-run --message-format=json 2>/dev/null \
    | python3 -c "import sys,json; [print(m['executable']) for line in sys.stdin if (m:=json.loads(line)).get('executable')]" \
    | while read -r exe; do cp "$exe" /usr/local/bin/llmproxy-integration; done

FROM registry.access.redhat.com/ubi10/ubi-minimal:latest AS integration-runner

RUN microdnf -y install openssl && microdnf clean all

COPY --from=integration-builder /usr/local/bin/llmproxy /usr/local/bin/llmproxy
COPY --from=integration-builder /usr/local/bin/llmproxy-integration /usr/local/bin/llmproxy-integration

CMD ["/usr/local/bin/llmproxy-integration"]

# --- Runtime ---
FROM registry.access.redhat.com/ubi10/ubi-minimal:latest

RUN microdnf -y install openssl && microdnf clean all

COPY --from=builder /usr/local/bin/llmproxy /usr/local/bin/llmproxy

EXPOSE 8080

ENTRYPOINT ["/usr/local/bin/llmproxy"]