self-hosted runner on odroid xu4 doesn't always start the job

Describe the bug
I have a self-hosted runner on odroid xu4 ARM single board computer and it quite often does not start the job.
I also have self-hosted runner on raspberry pi 3 and there is no such problem with it.

To Reproduce
Reproducibility ~50%

Steps to reproduce the behavior:

1. Set up the self-hosted runner on odroid xu4
2. Make a commit to the repo to start the workflow
3. The job which is supposed to be run by odroid xu4 remains in "Starting workflow run..." state for about 10 minutes, then it fails.

Note: The runner itself in that time is marked as "Active".

Looks like the runner picks the job, but does not start it, since when raspberry pi 3 runner gets freed it does not pick that stalled job, so it looks like it is considered already picked by odroid xu4.

Expected behavior
The runner on odroid xu4 board should pick up the job and execute it.

Runner Version and Platform

runner version 2.278.0

OS: Armbian Buster (Linux)

What's not working?

The job is not started.

Job Log Output

Runner and Worker's Diagnostic Logs

Worker_20210622-121039-utc.log

Start with the attached Worker_20210622-121039-utc.log and the reproduction steps for the odroid xu4 on Armbian Buster. Compare the stalled job behavior with the raspberry pi 3 runner and verify that the odroid runner reliably picks up and executes workflow jobs instead of timing out.