[anfimovdm]

Resolves AlmaLinux/build-system#563
Depends on AlmaLinux/alts#236

Summary

Deployment side of sending AlmaLinux 10 x86_64 test tasks only to x86-64-v3 capable workers, plus two ALTS scheduler deployment fixes found while rolling it out.

Changes

• Route AlmaLinux 10 x86_64 tasks to x86-64-v3 workers
  • playbooks/templates/alts_service_config.yaml.j2: renders queue_arch_overrides for the scheduler only (it's a scheduler setting). Versions are quoted so they stay strings.
  • inventories/alts/group_vars/scheduler.yml: enables the override for AlmaLinux 10, x86_64 → x86_64_v3.
  • inventories/alts/hosts.yml: the example x86_64 worker also consumes docker-x86_64_v3-0, so AlmaLinux 10 tasks have a consumer there.
• Separate scheduler broker and Redis hosts
  • The scheduler config used rabbitmq_host/alts_redis_host, the same variables as the workers. When the worker-facing addresses are passed as extra vars, they override the scheduler's group_vars.
  • The scheduler now uses scheduler_rabbitmq_host/scheduler_redis_host (renamed in group_vars/scheduler.yml); the workers keep the old names.
• ansible-core 2.19
  • roles/alts_scheduler/tasks/main.yml: Create scheduler working directory failed because its when: evaluated a string as a boolean.
• Terraform install on workers
  • roles/alts_terraform/tasks/main.yml: the role downloaded and unpacked Terraform on every run, and in --check mode failed with Source '/tmp/terraform.zip' does not exist because get_url doesn't download in check mode. It now reads the installed version and downloads, unpacks and cleans up only when Terraform is missing or a different version; the unpack is skipped in check mode.

Upgrade notes

• Inventories that set rabbitmq_host/alts_redis_host for the scheduler need to rename them to scheduler_rabbitmq_host/scheduler_redis_host.
• Add docker-x86_64_v3-0 to celery_queues only on workers whose CPU supports x86-64-v3 (grep -c avx2 /proc/cpuinfo is non-zero). With the override enabled, AlmaLinux 10 x86_64 tasks wait in RabbitMQ until such a worker consumes the queue.

Rollout

1. Deploy Schedule AlmaLinux 10 x86_64 tests only on x86-64-v3 capable workers alts#236 with the celery playbook to the workers, x86-64-v3 capable ones with the new queue.
2. Check the new queue has a consumer.
3. Run the scheduler playbook.

Testing

• Rendered the template with the scheduler and worker settings. The scheduler gets scheduler_rabbitmq_host/scheduler_redis_host and queue_arch_overrides; the worker gets rabbitmq_host/alts_redis_host and no override, also with the worker-facing addresses passed as extra vars.
• ansible-inventory --host resolves the renamed variables and queue_arch_overrides for the scheduler and the new queue for the worker.
• alts_scheduler.yml in --check --diff mode against production got past the working directory task and showed the expected config diff. Not yet applied.
• alts_celery.yml passes --syntax-check with the Terraform change; the role itself hasn't been re-run yet.