[ci] Migrate serve and runtime_env compute configs to new schema

[sai-miduthuri]

Summary

Migrates 12 Anyscale compute config files from the legacy schema to the new SDK 2026 schema, and adds anyscale_sdk_2026: true to the 10 corresponding tests in release_tests.yaml.

Compute configs migrated (12 files)

Serve tests (release/serve_tests/):

• compute_tpl_32_cpu.yaml / compute_tpl_32_cpu_gce.yaml
• compute_tpl_8_cpu_autoscaling.yaml
• compute_tpl_gpu_node.yaml / compute_tpl_gpu_node_gce.yaml
• compute_tpl_single_node_16_cpu.yaml
• compute_tpl_single_node_32_cpu.yaml
• compute_tpl_single_node_gce.yaml

Runtime env tests (release/runtime_env_tests/):

• rte_small.yaml / rte_gce_small.yaml
• rte_minimal.yaml / rte_gce_minimal.yaml

Tests updated in release_tests.yaml (10 tests)

1. pytest_serve_scale_replicas → compute_tpl_single_node_32_cpu.yaml
2. pytest_serve_multi_deployment_1k_noop_replica → compute_tpl_32_cpu.yaml (+ GCE variation on compute_tpl_32_cpu_gce.yaml)
3. pytest_serve_autoscaling_load_test → compute_tpl_single_node_32_cpu.yaml
4. serve_controller_benchmark_haproxy → compute_tpl_8_cpu_autoscaling.yaml (+ GCE on compute_tpl_single_node_gce.yaml)
5. pytest_serve_microbenchmarks → compute_tpl_single_node_16_cpu.yaml (+ GCE)
6. pytest_serve_throughput_optimized_microbenchmarks → compute_tpl_single_node_16_cpu.yaml (+ GCE)
7. pytest_serve_router_benchmark → compute_tpl_8_cpu_autoscaling.yaml (shares the config with serve_controller_benchmark_haproxy — must be flipped together)
8. pytest_serve_resnet_benchmark → compute_tpl_gpu_node.yaml (+ GCE on compute_tpl_gpu_node_gce.yaml)
9. runtime_env_rte_many_tasks_actors → rte_small.yaml (+ GCE on rte_gce_small.yaml)
10. runtime_env_wheel_urls → rte_minimal.yaml (+ GCE on rte_gce_minimal.yaml)

All 10 tests have their anyscale_sdk_2026: true on the base cluster: block; GCE variations inherit via deep_update. None of these tests have a kuberay variation.

Schema changes applied

• cloud_id → cloud, ANYSCALE_CLOUD_ID → ANYSCALE_CLOUD_NAME
• Removed region: us-west-2 / GCE region + allowed_azs → zones
• Removed max_workers: N (top-level)
• head_node_type → head_node, worker_node_types → worker_nodes (preserved empty lists as worker_nodes: [] on single-node configs)
• min_workers → min_nodes, max_workers → max_nodes
• use_spot: false → market_type: ON_DEMAND
• advanced_configurations_json → advanced_instance_config
• Dropped head/worker name: fields (single worker group per config)
• compute_tpl_gpu_node.yaml / compute_tpl_gpu_node_gce.yaml: head has GPU — set explicit resources: {CPU: 16, GPU: 1} per the plan's head-schedulability rule
• compute_tpl_8_cpu_autoscaling.yaml: flattened head resources: {custom_resources: {proxy: 1, CPU: 0}} to {CPU: 0, proxy: 1}; workers' {custom_resources: {proxy: 1}} flattened to {CPU: 8, proxy: 1} — explicit CPU count because the new SDK treats worker_nodes[].resources as a full override rather than a merge with the instance's natural resources
• compute_tpl_32_cpu.yaml: workers' {custom_resources: {worker: 1}} flattened to {CPU: 32, worker: 1} (same full-override rationale)
• rte_small.yaml / rte_gce_small.yaml: set explicit head resources: {CPU: 4} because the corresponding test has wait_for_nodes.num_nodes: 4 (= head + 3 workers) — head must be schedulable

Test plan

• All 12 config files validated against ComputeConfig.from_yaml()
• CI passes with anyscale_sdk_2026: true flag on all 10 test entries

🤖 Generated with Claude Code

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, have a team admin enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 9865da2. Configure here.}

[gemini-code-assist]

Code Review

This pull request migrates cluster compute templates to the new SDK 2026 schema across several YAML configuration files. Key changes include renaming fields such as cloud_id to cloud, head_node_type to head_node, and worker_node_types to worker_nodes, as well as updating worker node parameters and adding the anyscale_sdk_2026: true flag to test definitions. Review feedback identifies a missing flag for the pytest_serve_router_benchmark test that will cause parsing errors and suggests consistent placement of the new flag within the YAML structure for better readability.

[sai-miduthuri]

All tests pass on baae41b except for 2 tests that have been set to manual due to test errors unrelated to these compute-config changes.

These tests were failing before this PR, as seen at #63231 , so leaving the tests without any further changes to get them passing.