build: remove grpcio-tools build time dependency

[dtrifiro]

grpcio-tools is not available as a wheel for some platforms, making building vllm-tgis-adapter require building grpcio-tools from scratch.

Since this is not always possible (e.g. bazel not available on rhel), and is quite time consuming, we just include the generated code in the repo.

Note: this could (should?) also be added as a pre-commit hook, although this might slow down development a bit, so I left this as a gha step, for now.

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 63.46%. Comparing base (178d4ba) to head (3a03dd5).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #230   +/-   ##
=======================================
  Coverage   63.46%   63.46%           
=======================================
  Files          28       28           
  Lines        1768     1768           
  Branches      219      219           
=======================================
  Hits         1122     1122           
  Misses        541      541           
  Partials      105      105           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[R3hankhan123]

@dtrifiro , if we build grpcio-tools from scratch it will take alot of time as suggested in the pr, but for s390x we install the package with one env variable. Using this variable installation will take max 5 mins
GRPC_PYTHON_BUILD_SYSTEM_OPENSSL=1 python3 -m pip install grpcio-tools --no-cache-dir

Note: grpcio-tools took only 1-2 mins to build rest of time is taken by grpcio

[R3hankhan123]

after this vlllm-tgis-adapter is getting installed successfully after on s390x

[R3hankhan123]

@dtrifiro one more observation I have made for s390x is that during pip install vllm_tgis_adapter it takes the wheel that is already present on pypi and builds only grpcio and hftransfer

[R3hankhan123]

These packages are being installed

STEP 5/8: RUN --mount=type=cache,target=/root/.cache/pip     --mount=type=cache,target=/root/.cache/uv     --mount=type=bind,from=build,source=/workspace/vllm,target=/workspace/vllm     HOME=/root uv pip install vllm-tgis-adapter==0.6.2
Using Python 3.12.5 environment at: /opt/vllm
Resolved 117 packages in 248ms
Uninstalled 2 packages in 3ms
warning: Failed to hardlink files; falling back to full copy. This may lead to degraded performance.
         If the cache and target directories are on different filesystems, hardlinking may not be supported.
         If this is intentional, set `export UV_LINK_MODE=copy` or use `--link-mode=copy` to suppress this warning.
Installed 21 packages in 358ms
 + accelerate==1.3.0
 + cachetools==5.5.2
 + deprecated==1.2.18
 + googleapis-common-protos==1.68.0
 + grpcio==1.70.0
 + grpcio-health-checking==1.70.0
 + grpcio-reflection==1.70.0
 + hf-transfer==0.1.8
 - importlib-metadata==8.6.1
 + importlib-metadata==8.5.0
 + opentelemetry-api==1.30.0
 + opentelemetry-exporter-otlp==1.30.0
 + opentelemetry-exporter-otlp-proto-common==1.30.0
 + opentelemetry-exporter-otlp-proto-grpc==1.30.0
 + opentelemetry-exporter-otlp-proto-http==1.30.0
 + opentelemetry-proto==1.30.0
 + opentelemetry-sdk==1.30.0
 + opentelemetry-semantic-conventions==0.51b0
 + opentelemetry-semantic-conventions-ai==0.4.2
 - prometheus-client==0.21.1
 + prometheus-client==0.21.0
 + vllm-tgis-adapter==0.6.2

[dtrifiro]

@joerunde could you take a look? I don't see any potential issues with doing this apart from the annoyance of having to commit the generated .pyi in the repo and having to re-generate those every time generation.proto is updated.

Note: I'm not a big fan of introducing generated code in the repo and/or manual steps in the build process, so the alternative of using GRPC_PYTHON_BUILD_SYSTEM_OPENSSL=1 to build this package for s390x, might be good enough