Tailwind Traders triggers roughly 300 short job tasks an hour. Each task waits four to six minutes while VMs are acquired. The workspace must stay on classic compute because the tasks need a custom container image and an init script that reaches a private endpoint. Every run must remain isolated and billed at the jobs rate. You need to cut the start-up wait. What should you do?
- A.
Point every task at a single all-purpose cluster that is kept running continuously.
- B.
Enable Photon acceleration on every job cluster and switch the clusters to a compute-optimized node type so that each short task finishes its work sooner.
- C.
Enable autoscaling on each job cluster and raise the maximum worker count so that additional worker nodes are requested from Azure sooner.
- D.
Create an instance pool with the required node type and a minimum idle instance count, then configure the job clusters to draw nodes from the pool.
Show answer
Answer: D
An instance pool keeps warm, pre-acquired VMs on standby so job clusters can claim them in seconds while still being separate, jobs-rate clusters.
- A. A shared always-on all-purpose cluster removes isolation between runs and is billed at the higher interactive DBU rate, violating two stated constraints.
- B. Photon and node family affect execution speed after the cluster is running, not the time spent waiting for Azure to allocate virtual machines.
- C. Autoscaling adjusts worker count on a cluster that is already up; it cannot shorten the initial VM acquisition and a higher ceiling can lengthen it.
- D. Pools hold pre-acquired, optionally pre-loaded VMs, so job clusters start in seconds while keeping per-run isolation and jobs-rate billing.