Merge branch 'master' into codex/fix-ray-client-specific-server-fork-race

Author: Jerry2003826

.buildkite/rllib.rayci.yml

@@ -1,25 +1,30 @@
 group: rllib tests
 depends_on:
   - forge
-  - ray-core-build(python=3.10)
+  - ray-core-build(*)
   - ray-dashboard-build
 steps:
   # builds
-  - name: rllibbuild
+  - name: rllibbuild-multipy
+    label: "wanda: rllibbuild-py{{array.python}}"
     wanda: ci/docker/rllib.build.wanda.yaml
-    depends_on: oss-ci-base_ml-multipy(python=3.10)
+    tags: cibase
+    array:
+      python:
+        - "3.10"
+        - "3.12"
     env:
-      PYTHON: "3.10"
+      PYTHON: "{{array.python}}"
       BASE_TYPE: "ml"
       BUILD_VARIANT: "build"
       RAYCI_IS_GPU_BUILD: "false"
-    tags: cibase
+    depends_on: oss-ci-base_ml-multipy($)
 
   - name: rllibgpubuild
     wanda: ci/docker/rllib.build.wanda.yaml
-    depends_on: oss-ci-base_gpu-multipy(python=3.10)
+    depends_on: oss-ci-base_gpu-multipy(python=3.12)
     env:
-      PYTHON: "3.10"
+      PYTHON: "3.12"
       BASE_TYPE: "gpu"
       BUILD_VARIANT: "gpubuild"
       RAYCI_IS_GPU_BUILD: "true"
@@ -37,9 +42,9 @@ steps:
         --only-tags env,evaluation,models,offline,policy,utils,algorithms,callbacks,core
         --except-tags gpu,multi_gpu,manual
         --test-env RAY_USE_MULTIPROCESSING_CPU_COUNT=1
-        --build-name rllibbuild-py3.10
-        --python-version 3.10
-    depends_on: rllibbuild
+        --build-name rllibbuild-py3.12
+        --python-version 3.12
+    depends_on: rllibbuild-multipy(python=3.12)
 
   - label: ":brain: rllib: examples"
     tags: rllib
@@ -52,18 +57,18 @@ steps:
         --only-tags examples
         --except-tags gpu,multi_gpu,manual,examples_use_all_core
         --test-env RAY_USE_MULTIPROCESSING_CPU_COUNT=1
-        --build-name rllibbuild-py3.10
-        --python-version 3.10
+        --build-name rllibbuild-py3.12
+        --python-version 3.12
       # Tests all examples without gpu or multi_gpu
       - bazel run //ci/ray_ci:test_in_docker -- //rllib/... rllib
         --workers "$${BUILDKITE_PARALLEL_JOB_COUNT}" --worker-id "$${BUILDKITE_PARALLEL_JOB}"
         --only-tags examples_use_all_core
         --except-tags gpu,multi_gpu,manual
         --test-env RAY_USE_MULTIPROCESSING_CPU_COUNT=1
-        --build-name rllibbuild-py3.10
-        --python-version 3.10
+        --build-name rllibbuild-py3.12
+        --python-version 3.12
         --skip-ray-installation # reuse the same docker image as the previous run
-    depends_on: rllibbuild
+    depends_on: rllibbuild-multipy(python=3.12)
 
   - label: ":brain: rllib: learning tests"
     tags: rllib
@@ -76,18 +81,18 @@ steps:
         --only-tags learning_tests
         --except-tags gpu,multi_gpu,learning_tests_use_all_core,manual
         --test-arg --framework=torch
-        --build-name rllibbuild-py3.10
-        --python-version 3.10
+        --build-name rllibbuild-py3.12
+        --python-version 3.12
       # learning tests without a gpu but use all cores
       - bazel run //ci/ray_ci:test_in_docker -- //rllib/... rllib
         --workers "$${BUILDKITE_PARALLEL_JOB_COUNT}" --worker-id "$${BUILDKITE_PARALLEL_JOB}"
         --only-tags learning_tests_use_all_core
         --except-tags gpu,multi_gpu,manual
         --test-arg --framework=torch
-        --build-name rllibbuild-py3.10
-        --python-version 3.10
+        --build-name rllibbuild-py3.12
+        --python-version 3.12
         --skip-ray-installation # reuse the same docker image as the previous run
-    depends_on: rllibbuild
+    depends_on: rllibbuild-multipy(python=3.12)
 
   - label: ":brain: rllib: gpu tests"
     tags:
@@ -104,8 +109,8 @@ steps:
         --except-tags multi_gpu,manual
         --test-env=RAY_USE_MULTIPROCESSING_CPU_COUNT=1
         --test-env=RLLIB_NUM_GPUS=1
-        --build-name rllibgpubuild-py3.10
-        --python-version 3.10
+        --build-name rllibgpubuild-py3.12
+        --python-version 3.12
     depends_on: rllibgpubuild
 
   - label: ":brain: rllib: multi-gpu tests"
@@ -123,8 +128,8 @@ steps:
         --gpus 4
         --only-tags multi_gpu
         --except-tags manual
-        --build-name rllibgpubuild-py3.10
-        --python-version 3.10
+        --build-name rllibgpubuild-py3.12
+        --python-version 3.12
     depends_on: rllibgpubuild
 
   - label: ":brain: rllib: doc tests"
@@ -139,23 +144,40 @@ steps:
         --only-tags doctest
         --except-tags gpu,manual
         --parallelism-per-worker 2
-        --build-name rllibbuild-py3.10
-        --python-version 3.10
+        --build-name rllibbuild-py3.12
+        --python-version 3.12
       # doc examples
       - bazel run //ci/ray_ci:test_in_docker -- //doc/... rllib
         --except-tags gpu,post_wheel_build,timeseries_libs,doctest
         --parallelism-per-worker 2
-        --build-name rllibbuild-py3.10
-        --python-version 3.10
+        --build-name rllibbuild-py3.12
+        --python-version 3.12
         --skip-ray-installation # reuse the same docker image as the previous run
       # documentation test
       - bazel run //ci/ray_ci:test_in_docker -- //rllib/... rllib
         --only-tags documentation
         --parallelism-per-worker 2
+        --build-name rllibbuild-py3.12
+        --python-version 3.12
+        --skip-ray-installation # reuse the same docker image as the previous run
+    depends_on: rllibbuild-multipy(python=3.12)
+
+  - label: ":brain: rllib: python 3.10 component testing ({{matrix.worker_id}})"
+    if: build.pull_request.labels includes "continuous-build" || pipeline.id == "0189e759-8c96-4302-b6b5-b4274406bf89" || pipeline.id == "018f4f1e-1b73-4906-9802-92422e3badaa"
+    tags: rllib_directly
+    instance_type: large
+    commands:
+      - bazel run //ci/ray_ci:test_in_docker -- //rllib/... rllib
+        --workers 4 --worker-id {{matrix.worker_id}} --parallelism-per-worker 3
+        --only-tags env,evaluation,models,offline,policy,utils,algorithms,callbacks,core
+        --except-tags gpu,multi_gpu,manual
+        --test-env RAY_USE_MULTIPROCESSING_CPU_COUNT=1
         --build-name rllibbuild-py3.10
         --python-version 3.10
-        --skip-ray-installation # reuse the same docker image as the previous run
-    depends_on: rllibbuild
+    depends_on: rllibbuild-multipy(python=3.10)
+    matrix:
+      setup:
+        worker_id: ["0", "1", "2", "3"]
 
   - label: ":brain: rllib: flaky component & examples tests"
     key: rllib_flaky_tests_02
@@ -170,27 +192,27 @@ steps:
         --only-tags env,evaluation,models,offline,policy,utils,algorithms,callbacks,core
         --except-tags learning_tests,examples,documentation,gpu,multi_gpu,manual
         --test-env RAY_USE_MULTIPROCESSING_CPU_COUNT=1
-        --build-name rllibbuild-py3.10
-        --python-version 3.10
+        --build-name rllibbuild-py3.12
+        --python-version 3.12
 
       # flaky examples
       - bazel run //ci/ray_ci:test_in_docker -- //rllib/... rllib  --run-flaky-tests --parallelism-per-worker 3
         --only-tags examples
         --except-tags multi_gpu,gpu,manual,examples_use_all_core
         --test-env RAY_USE_MULTIPROCESSING_CPU_COUNT=1
-        --build-name rllibbuild-py3.10
-        --python-version 3.10
+        --build-name rllibbuild-py3.12
+        --python-version 3.12
         --skip-ray-installation # reuse the same docker image as the previous run
 
       # flaky examples use all core
       - bazel run //ci/ray_ci:test_in_docker -- //rllib/... rllib  --run-flaky-tests --parallelism-per-worker 3
         --only-tags examples_use_all_core
         --except-tags gpu,multi_gpu,manual
         --test-env RAY_USE_MULTIPROCESSING_CPU_COUNT=1
-        --build-name rllibbuild-py3.10
-        --python-version 3.10
+        --build-name rllibbuild-py3.12
+        --python-version 3.12
         --skip-ray-installation # reuse the same docker image as the previous run
-    depends_on: rllibbuild
+    depends_on: rllibbuild-multipy(python=3.12)
     soft_fail: true
 
   - label: ":brain: rllib: flaky learning tests"
@@ -206,9 +228,9 @@ steps:
         --only-tags learning_tests
         --except-tags gpu,multi_gpu,manual
         --test-arg --framework=torch
-        --build-name rllibbuild-py3.10
-        --python-version 3.10
-    depends_on: rllibbuild
+        --build-name rllibbuild-py3.12
+        --python-version 3.12
+    depends_on: rllibbuild-multipy(python=3.12)
     soft_fail: true
 
   - label: ":brain: rllib: flaky gpu tests"
@@ -225,8 +247,8 @@ steps:
         --except-tags multi_gpu,manual
         --test-env=RAY_USE_MULTIPROCESSING_CPU_COUNT=1
         --test-env=RLLIB_NUM_GPUS=1
-        --build-name rllibgpubuild-py3.10
-        --python-version 3.10
+        --build-name rllibgpubuild-py3.12
+        --python-version 3.12
     depends_on: rllibgpubuild
     soft_fail: true
 
@@ -244,7 +266,7 @@ steps:
         --gpus 4
         --only-tags multi_gpu
         --except-tags manual
-        --build-name rllibgpubuild-py3.10
-        --python-version 3.10
+        --build-name rllibgpubuild-py3.12
+        --python-version 3.12
     depends_on: rllibgpubuild
     soft_fail: true

.buildkite/serve.rayci.yml

@@ -303,4 +303,21 @@ steps:
     commands:
       - bazel run //ci/ray_ci:test_in_docker -- //... serve --run-flaky-tests --parallelism-per-worker 3
         --python-version 3.10 --build-name servebuild-py3.10
+        --except-tags gpu
     depends_on: servebuild-multipy(python=3.10)
+
+  - label: ":ray-serve: serve: flaky gpu tests"
+    key: serve_flaky_gpu_tests
+    tags:
+      - serve
+      - python
+      - gpu
+      - flaky
+      - skip-on-premerge
+    instance_type: gpu
+    soft_fail: true
+    commands:
+      - bazel run //ci/ray_ci:test_in_docker -- //... serve --run-flaky-tests
+        --python-version 3.10 --build-name docgpubuild-py3.10
+        --only-tags gpu
+    depends_on: docgpubuild

.github/CODEOWNERS

@@ -131,3 +131,9 @@
 /.github/workflows/ @ray-project/ray-ci
 
 /.gemini/ @edoakes @ray-project/ray-ci
+
+# TPU
+/python/ray/util/tpu.py @andrewsykim @edoakes @ryanaoleary
+/python/ray/tests/test_tpu.py @andrewsykim @edoakes @ryanaoleary
+/python/ray/tests/accelerators/test_tpu.py @andrewsykim @edoakes @ryanaoleary
+/python/ray/_private/accelerators/tpu.py @andrewsykim @edoakes @ryanaoleary

README.rst

@@ -48,7 +48,7 @@ Install Ray with: ``pip install ray``. For nightly wheels, see the
 `Installation page <https://docs.ray.io/en/latest/ray-overview/installation.html>`__.
 
 .. _`Serve`: https://docs.ray.io/en/latest/serve/index.html
-.. _`Data`: https://docs.ray.io/en/latest/data/dataset.html
+.. _`Data`: https://docs.ray.io/en/latest/data/data.html
 .. _`Workflow`: https://docs.ray.io/en/latest/workflows/
 .. _`Train`: https://docs.ray.io/en/latest/train/train.html
 .. _`Tune`: https://docs.ray.io/en/latest/tune/index.html

ci/ci.sh

@@ -214,8 +214,9 @@ build_dashboard_front_end() {
       if [[ -z "${BUILDKITE-}" || "${OSTYPE}" != linux* ]]; then
         if [[ -d "${HOME}/.nvm" ]]; then
           set +x  # suppress set -x since it'll get very noisy here
+          export TMPDIR="${TMPDIR:-/tmp}"
           . "${HOME}/.nvm/nvm.sh"
-          NODE_VERSION="14"
+          NODE_VERSION="20"
           nvm install $NODE_VERSION
           nvm use --silent $NODE_VERSION
         fi

ci/docker/llm.build.Dockerfile

@@ -1,6 +1,42 @@
 # syntax=docker/dockerfile:1.3-labs
 
 ARG DOCKER_IMAGE_BASE_BUILD=cr.ray.io/rayproject/oss-ci-base_build-py3.11
+FROM $DOCKER_IMAGE_BASE_BUILD AS haproxy-builder
+
+RUN <<EOF
+#!/bin/bash
+set -euo pipefail
+
+apt-get update -y
+apt-get install -y --no-install-recommends \
+    build-essential \
+    ca-certificates \
+    curl \
+    libc6-dev \
+    liblua5.3-dev \
+    libpcre3-dev \
+    libssl-dev \
+    zlib1g-dev
+
+# Install HAProxy from source.
+# Fetched from ray-project/haproxy-release (a GitHub release mirror) because
+# www.haproxy.org's wildcard TLS cert expired 2026-04-17 and the release tarball
+# disappeared from the upstream download site. Integrity is enforced by sha256
+# verification. Drop this mirror and switch back to www.haproxy.org once the
+# cert is renewed and the tarball is republished.
+HAPROXY_VERSION="2.8.20"
+HAPROXY_SHA256="c8301de11dabfbf049db07080e43b9570a63f99e41d4b0754760656bf7ea00b7"
+HAPROXY_BUILD_DIR=$(mktemp -d)
+curl --retry 5 --retry-all-errors --connect-timeout 20 --max-time 300 \
+     -sSfL -o "${HAPROXY_BUILD_DIR}/haproxy.tar.gz" \
+     "https://github.com/ray-project/haproxy-release/releases/download/${HAPROXY_VERSION}/haproxy-${HAPROXY_VERSION}.tar.gz"
+echo "${HAPROXY_SHA256}  ${HAPROXY_BUILD_DIR}/haproxy.tar.gz" | sha256sum -c -
+tar -xzf "${HAPROXY_BUILD_DIR}/haproxy.tar.gz" -C "${HAPROXY_BUILD_DIR}" --strip-components=1
+make -C "${HAPROXY_BUILD_DIR}" TARGET=linux-glibc USE_OPENSSL=1 USE_ZLIB=1 USE_PCRE=1 USE_LUA=1 USE_PROMEX=1 -j$(nproc)
+make -C "${HAPROXY_BUILD_DIR}" install
+rm -rf "${HAPROXY_BUILD_DIR}"
+EOF
+
 FROM $DOCKER_IMAGE_BASE_BUILD
 
 ARG RAY_CI_JAVA_BUILD=
@@ -10,6 +46,17 @@ SHELL ["/bin/bash", "-ice"]
 
 COPY . .
 
+COPY --from=haproxy-builder /usr/local/sbin/haproxy /usr/local/sbin/haproxy
+
+RUN <<EOF
+#!/bin/bash
+set -euo pipefail
+
+apt-get update -y && apt-get install -y --no-install-recommends liblua5.3-0 libpcre3
+rm -rf /var/lib/apt/lists/*
+mkdir -p /etc/haproxy /run/haproxy /var/log/haproxy
+EOF
+
 RUN <<EOF
 #!/bin/bash
 

ci/env/install-dependencies.sh

@@ -163,8 +163,15 @@ install_node() {
 
     (
       set +x # suppress set -x since it'll get very noisy here.
+      # nvm's error-recovery path in nvm.sh references $TMPDIR. The caller
+      # runs with `set -u`, so an unset TMPDIR (launchd-spawned shells don't
+      # inherit one) crashes nvm with "TMPDIR: unbound variable" whenever a
+      # binary download fails and nvm falls back to source.
+      export TMPDIR="${TMPDIR:-/tmp}"
       . "${HOME}/.nvm/nvm.sh"
-      NODE_VERSION="14"
+      # Node 14 EOL'd April 2023; nodejs.org removed the darwin-arm64
+      # prebuilt and the URL now 404s. 20 is the current LTS line.
+      NODE_VERSION="20"
       nvm install $NODE_VERSION
       nvm use --silent $NODE_VERSION
       npm config set loglevel warn  # make NPM quieter