Borg Priorities

The priority of a job helps define how the scheduler treats it. Ranges of priorities that share similar properties are referred to as tiers: • Free tier: jobs running at these lowest priorities incur no internal charges, and have no Service Level Objectives (SLOs). 2019 trace priority <= 99; 2011 trace priority bands 0 and 1. • Best-effort Batch (beb) tier: jobs running at these priorities are managed by the batch scheduler and incur low internal charges; they have no associated SLOs. 2019 trace priority 110–115; 2011 trace priority bands 2–8. • Mid-tier: jobs in this category offer SLOs weaker than those offered to production tier workloads, as well as lower internal charges. 2019 trace priority 116–119; not present in the 2011 trace. • Production tier: jobs in this category require high availability (e.g., user-facing service jobs, or daemon jobs providing storage and networking primitives); internally charged for at “full price”. Borg will evict lower-tier jobs in order to ensure production tier jobs receive their expected level of service. 2019 trace priority 120–359; 2011 trace priority bands 9–10. • Monitoring tier: jobs we deem critical to our infrastructure, including ones that monitor other jobs for problems. 2019 trace priority >= 360; 2011 trace priority band 11. (We merged the small number of monitoring jobs into the Production tier for this paper.)

2.5 Priority, quota, and admission control
What happens when more work shows up than can be accommodated? Our solutions for this are priority and quota.
Every job has a priority, a small positive integer. A highpriority task can obtain resources at the expense of a lowerpriority one, even if that involves preempting (killing) the
latter. Borg defines non-overlapping priority bands for different uses, including (in decreasing-priority order): monitoring, production, batch, and best effort (also known as
testing or free). For this paper, prod jobs are the ones in the
monitoring and production bands.

Leave a Reply