PAAS runner can get stuck if interactions with the user-visible job log fail
The PAAS Runner watches the startup of jobs by querying the job trace (internal terminology for the job log that gets displayed in the Web UI), and issue a timeout if it is empty after a while (300s by default). It also appends to the trace in case of timeout.
But it is possible that the trace querying itself, or the subsequent reporting to the job trace fails. The exception catching there not being wide enough, the end result is the job staying in the internal queue for jobs that are in the process of being launched, quickly saturating it. At this point, the runner will not acquire new jobs and is completely stuck.