Vertex AI Notebook Execution Job
Does ZopNight manage Vertex AI Notebook Execution Job?
Vertex AI notebook execution jobs run a notebook non-interactively and bill for the machine held during each run. A daily schedule on an oversized machine repeats its waste every day, which is why ZopNight lists execution jobs via the live aiplatform API across 32 regions and attributes their recurring compute spend.
Rules that fire on Vertex AI Notebook Execution Job
No active rule family targets Vertex AI Notebook Execution Job today. Rules that used to are retired, and retired rules publish no pages and fire no findings. Scheduling and permissions coverage are unaffected.
A notebook execution job runs a notebook non-interactively on managed compute, billed for the run duration. Scheduled executions on oversized machines quietly repeat their waste daily.
Paying per scheduled notebook run
An execution job provisions the machine named in its configuration, runs the notebook top to bottom, and releases the compute when the run ends. Billing covers exactly that window. The economics differ from an interactive notebook: there is no idle time inside a run, so cost hygiene comes down to machine specification and run frequency. A schedule converts any misconfiguration into an annuity: the same oversized machine, the same unnecessary GPU, every day at the same hour.
Execution jobs and their recurring spend
ZopNight inventories execution jobs via the live aiplatform API across 32 Vertex regions and attributes their recurring compute spend. That live listing corrects the stale state Cloud Asset Inventory reports for transient job types. Execution jobs are not schedulable in ZopNight, and would gain nothing from it: each run already has a defined end. The recurring schedule that creates the runs is the thing to review, not any individual run.
When a daily run repeats its waste daily
Look for executions configured with a GPU because the interactive notebook they were copied from had one, though the batch path never touches it; schedules that survive their consumers, refreshing a report nobody opens anymore; and notebooks that grew over time, so a run that once took minutes now takes hours on the same cadence. Multiply any of these by 30 runs a month and small sloppiness compounds into a real line item.
Notebook executions in the console
Google Cloud console → Vertex AI → Colab Enterprise → Executions lists runs with duration and machine configuration. Sort by frequency and duration to find the schedules whose cost quietly grew while nobody was reading the output.