Intermittent runner communication loss on AWS Fargate (ephemeral, ~2% failure rate)

Describe the bug

An ephemeral self-hosted runner hosted on AWS Fargate (Ubuntu 24.04) intermittently loses
communication with the GitHub Actions server shortly after a job starts. This does not happen
consistently — it occurs roughly once every 50 job runs. When it fails, the Fargate container
logs are truncated immediately after the "Running job" line, and the runner never produces
a completion log. The GitHub UI reports a communication loss error.

To Reproduce

This is an intermittent issue and cannot be reliably reproduced on demand. General setup:

1. Register an ephemeral self-hosted runner on AWS Fargate (Ubuntu 24.04, arm64).
2. Trigger a workflow that dispatches a job to the runner.
3. Approximately 1 out of 50 runs, the runner stops responding after picking up the job.

Expected behavior

The runner should complete the job and produce the following log sequence:

Job completed with result: Succeeded
√ Removed .credentials
√ Removed .runner
Runner listener exit with 0 return code, stop the service, no retry needed.
Exiting runner…
Runner exited with code: 0

Runner Version and Platform

• Runner version: actions-runner-linux-arm64-2.311.0
• OS: Ubuntu 24.04 (ARM64), running inside AWS Fargate
• Runner mode: Ephemeral (--ephemeral)

What's not working?

The runner loses communication with the GitHub server after picking up a job.

GitHub UI error:

The self-hosted runner lost communication with the server. Verify the machine is running
and has a healthy network connection. Anything in your workflow that terminates the runner
process, starves it for CPU/Memory, or blocks its network access can cause this error.

Notes:

• Memory and disk usage are logged every 5 seconds; no anomalies observed around failure time.
• No signs of OOM, CPU starvation, or network disruption in Fargate metrics.
• The runner container exits without producing a completion log or cleanup output.

Job Log Output

Failing run — Fargate container log ends abruptly at:
2026-05-21 01:03:53Z: Running job: ci-cd / dev-backend-ci-cd
(No further output. Container exits.)

Successful run — Fargate container log continues normally:
2026-05-21 01:03:53Z: Running job: ci-cd / dev-backend-ci-cd
2026-06-11 00:17:26Z: Job ci-cd / dev-backend-ci-cd completed with result: Succeeded
√ Removed .credentials
√ Removed .runner
Runner listener exit with 0 return code, stop the service, no retry needed.
Exiting runner…
Runner exited with code: 0

Runner and Worker's Diagnostic Logs

Diagnostic logs from the _diag folder were collected and reviewed. However, the logs only
provide more detailed output up to the following line — nothing is logged after this point:

2026-05-21 01:03:53Z: Running job: ci-cd / dev-backend-ci-cd

The diagnostic logs confirm the runner picked up the job successfully, but provide no
information about what caused the runner to stop responding afterward. The root cause
of the hang/exit after job dispatch remains unidentified from the available logs.

Start by comparing the failing and successful AWS Fargate container logs and the collected _diag logs around the "Running job" line. Investigate the runner's behavior after job dispatch, using the stated Ubuntu 24.04 ARM64 ephemeral setup and intermittent reproduction details. Done means identifying the communication-loss cause and restoring completion and cleanup logging for affected runs.