docker-autoscaler: Jobs queue on existing instances instead of scaling up when capacity limit reached

Summary

GitLab Runner docker-autoscaler executor not scaling properly with multiple instances when capacity is reached. When concurrent jobs exceed current capacity, jobs queue instead of triggering new instance creation, regardless of max_instances setting.

Steps to reproduce

1. Configure runner with:

   • capacity_per_instance = 2
   • max_instances = 2
   • idle_count = 1

2. Submit multiple concurrent jobs (4+)

.gitlab-ci.yml

test-job:
  script:
    - echo "Starting long running job..."
    - sleep 300
    - echo "Job complete"

Alternative reproduction with multiple runners:

1. Configure manager with:

   • concurrent = 8
   • capacity_per_instance = 2
   • max_instances = 4
   • idle_count = 3
   • idle_time = "1h"

2. Configure runners with:

   • concurrent = 2

3. Submit 6 jobs simultaneously

Expected: Jobs should distribute across 3 instances (2 jobs per instance) Actual: Jobs queue on existing instances and run sequentially despite available max_instances headroom

Actual behavior

All jobs queue up on first instance instead of triggering autoscaling to create new instances.

Expected behavior

When first instance reaches capacity (2 jobs), autoscaler should create second instance to handle additional jobs while maintaining idle_count=1.

Relevant logs

job log

Dec 04 03:44:54 - Job #8538068739 starts
Dec 04 03:45:05 - Job #8538069349 starts
[Both jobs assigned to same instance i-06e9991af2c40bc92]
Dec 04 03:42:03 - Failed to process runner error=failed to update executor: reserving taskscaler capacity: no capacity: no immediately available capacity

Environment

• GitLab Runner 17.6.0
• docker-autoscaler executor
• AWS Fleeting plugin

config.toml

concurrent = 2
[[runners]]
executor = "docker-autoscaler"
[runners.autoscaler]
capacity_per_instance = 2
max_instances = 2
idle_count = 1

Possible fixes

The taskscaler needs to properly factor in instance utilization when making scaling decisions.