Sail clusters and sizes
Databasin Sail cluster sizes from XSmall to XLarge: workers, task slots, coordinator options, access modes and how to pick the right size.
A Sail cluster is one coordinator and a set of workers, picked together as a size. You choose a size when you create the cluster and can change it later. Medium is the default, and the right place to start for most teams.
The sizes
| Size | Workers | Each worker | Recommended coordinator |
|---|---|---|---|
| XSmall | 1 | 8 vCPU, 64 GB | 2 vCPU, 4 GB |
| Small | 2 | 8 vCPU, 64 GB | 2 vCPU, 4 GB |
| Medium | 4 | 8 vCPU, 64 GB | 3 vCPU, 6 GB |
| Large | 4 | 16 vCPU, 128 GB | 4 vCPU, 8 GB |
| XLarge | 5 | 32 vCPU, 256 GB | 4 vCPU, 8 GB |
You can change the worker count for any size under Advanced, from 0 to 10. With 0, the coordinator runs every query itself. A cluster changed this way shows as Custom (based on Medium), or whichever size you started from.
Databasin Cloud prices, billed by the minute (Pricing has every rate):
| Size | Per hour, all running | Ready, workers at zero |
|---|---|---|
| XSmall | $1.82 | $0.27 |
| Small | $3.37 | $0.27 |
| Medium | $6.61 | $0.41 |
| Large | $12.94 | $0.54 |
| XLarge | $31.55 | $0.54 |
Each extra worker is $1.55 an hour on XSmall, Small and Medium, $3.10 on Large and $6.20 on XLarge.
Task slots: how much runs at once
Each worker runs several pieces of a query at the same time. We call those task slots:
| Worker | Task slots |
|---|---|
| 8 vCPU, 64 GB | 6 |
| 16 vCPU, 128 GB | 8 |
| 32 vCPU, 256 GB | 16 |
A cluster's parallelism is its workers times their slots. A Medium cluster runs 4 × 6 = 24 pieces at once; an XLarge runs 5 × 16 = 80. More slots help most with big scans, joins and aggregations. A small lookup query won't get faster on a bigger cluster.
The coordinator
The coordinator plans queries, checks access, answers metadata and runs the notebook service. Two settings matter:
- Keep the coordinator always on (checked by default). BI tools and SQL connect instantly, and the coordinator runs all the time. Uncheck it to let the coordinator scale to zero when idle; the first connection after a quiet period then waits while it starts.
- Coordinator size. Each size recommends one, and you can pick another:
| Coordinator | Offered when | Python notebook sessions |
|---|---|---|
| 2 vCPU, 4 GB | Always on or scales to zero | 2 |
| 3 vCPU, 6 GB | Always on or scales to zero | 8 |
| 4 vCPU, 8 GB | Always on or scales to zero | 8 |
| 4 vCPU, 16 GB (dedicated) | Scales to zero only | 20 |
| 8 vCPU, 32 GB (dedicated) | Always on or scales to zero | 50 |
| 8 vCPU, 64 GB (dedicated) | Always on or scales to zero | 100 |
| 16 vCPU, 128 GB (dedicated) | Always on only | 100 |
Pick a bigger coordinator when many people run Python notebooks at once. Advanced shows how many SQL sessions the coordinator you picked allows.
A dedicated coordinator needs its own node before it starts, so a cold start takes about 2–5 minutes. BI tools such as Tableau and Power BI may time out while it wakes. If people connect BI tools, keep it always on.
Workers
- Workers scale to zero (on by default). Workers stop about 30 minutes after the last request and start again for the next query, which waits for them. Turn it off to keep workers running so the first query never waits; they're then billed continuously.
- Scheduled starts (when the cluster scales to zero) wake it at times you choose and keep it warm for a window, so a morning dashboard refresh doesn't wait.
Advanced settings
| Setting | What it does |
|---|---|
| Worker count | Overrides the size's workers (0 to 10). |
| Max Python sessions | Python notebook sessions that run at once, up to what the coordinator holds. |
| Session idle timeout | Idle sessions close after this many minutes (default 30). |
| Access mode | Shared (default): everyone in the project can use the cluster; Python UDFs are off. Single user: one project member you choose; Python UDFs allowed. |
| Python packages | PyPI packages installed each time the cluster starts. Project admins set them. See Python and PySpark on Sail. |
Any change to the cluster itself (size, coordinator, workers or these settings) restarts it, so the edit dialog asks you to confirm with Restart and apply. Network access and scheduled starts change without a restart.
How to pick
- Most teams: Medium, coordinator always on, workers scaling to zero. Change it when you have a reason.
- Light use or small data: XSmall or Small. One or two workers handle everyday SQL on modest tables, and cost less while running.
- Heavy joins and big tables: Large or XLarge. Bigger workers bring more memory per worker and more task slots.
- Lots of notebook users: a bigger coordinator, sized by the Python sessions column above.
- Python UDFs: a single-user cluster for the person who needs them.
- Dashboards at a fixed time: scheduled starts, so the refresh finds workers already running.