Do the weekly check
This lesson is one habit. It takes fifteen seconds, and it is the difference between adjusting course on a Tuesday and explaining a bill in arrears.
Open your organization. You land on the dashboard.
Answer four questions
Do this now, on your own dashboard. Each answer is somewhere on the page.
1. How much of this billing period's compute have you used?
Find the plan and usage panel. Note the figure and the period it covers.
2. Is that ahead of or behind where the calendar is?
Two-thirds through the month with 90% of the budget spent is a finding. Two-thirds through with 40% spent is not. The absolute number tells you nothing without that comparison, which is why the habit is weekly rather than a glance whenever someone worries.
3. How many workspaces are there, and is every one of them in use?
An abandoned workspace still holds storage. If you cannot name who uses one, that is worth chasing.
4. Is anything mid-provision right now?
A workspace still deploying is not usable. Sending someone to it produces a confusing bug report rather than an error message.
Now check the thing the dashboard does not volunteer
Are start times slow?
If sessions or clusters take a long time to become available while usage looks healthy, you are probably hitting a concurrency limit rather than a capacity one:
| What it bounds | How it fails | |
|---|---|---|
| Capacity | total compute this period | work refused, or a larger bill |
| Concurrency | how much runs at once | work accepted but queued |
The second gets misdiagnosed constantly, because it does not look like a limit. It looks like the platform being slow, and gets reported that way.
If you have heard "Eugo feels sluggish this week" from anyone, check concurrency before you believe it.
Set the habit up so you do not rely on remembering
Put a fifteen-second recurring reminder on a day you are already at your desk. Answering four questions weekly costs you a minute a month.
The alternative is not "check it when it matters." It is finding out afterward.
Before you optimize anything
The dashboard tells you how much. It does not tell you which workloads, and that is what makes the number actionable.
Tuning a job responsible for 3% of the bill is effort with nothing to show for it. Establish which workloads dominate first, then look at those. The cost optimization checklist is the working form of that order.
If a team is always at a limit
The plan does not fit how they work.
That is a sizing decision, not a discipline problem. Treated as the latter, you get people running smaller jobs than they should, or waiting on capacity, which costs more in salary than the upgrade would.
What you now know
- The four questions, and that question 2 is the one that makes the others mean anything.
- That "the platform is slow" is often a concurrency limit wearing a disguise.
- Why attribution comes before optimization.
Next: inviting members and assigning roles. Getting people into a workspace with the right access, in the right order.
Video: Reading the dashboard (same material, with a transcript).