Capacity planning

Open Capacity to see three answers first: which pools fill up soon and when, who is filling them (the production whose mapped datasets grew most over the last 24 hours), and how much room is left in total against your warning line. This is runway to your warning threshold, not necessarily to a full pool.

List ranks pools by the shortest time to the warning threshold. Each row shows a status against your real thresholds — Critical at or past the hard threshold, Warning past the warning threshold or reaching it within your planning lead time (7 days by default), Soon reaching it within the forecast horizon (14 days by default), OK beyond that, and No forecast when history is not ready (never shown as OK) — plus % used against the warning line, the day it is reached, 24-hour growth, and free space. A long tail of OK pools collapses into one line; View all expands it. The side column names the largest growing dataset on each pool that needs action. Select a system to open its Capacity tab for the full chart and estimates.

Map (?view=map) shows where the space is: one framed box per pool sized by space used and coloured by pool status, split into its largest datasets (each counted by the space it holds itself, so nested datasets are not double counted) with everything else as other. Labels appear only where they fit. Click a box to see that pool's datasets, their growth, and any production mapping. This is dataset-level accounting only; Pillar never crawls files or inspects media.

The planning estimate is the headline number: a robust trend fit over a trailing window of history, shown with a scenario range — the spread of plausible threshold-crossing dates, not a promise of one exact day. It is deliberately slow to move so a single busy day doesn't whipsaw your plan.

When the last 24 hours of ingest is running elevated, or materially faster than the planning trend, the tab also shows a second, separately labelled “if recent ingest continues” estimate. Treat it as an early warning on top of the planning estimate, not a replacement for it — it reacts fast and can be noisy.

A pool needs at least 7 days of reasonably regular, fresh history before Pillar will plan a runway on it. Until then — or when history goes stale, has a gap large enough to hide a real outage, or shows no sustained growth — the pool shows a data-quality status (Collecting history, Data stale, Gaps in history, No sustained growth) and a short reason instead of a number. The capacity chart still draws actual usage history and your threshold lines in that state; it withholds only the projection, because a projection from too little or too broken a signal is worse than none.

Some Agents can report a pool's total size but not how much of it is used — today, a generic Linux host (md RAID or plain ext4/XFS) whose Agent container has no access to the host filesystems. Pillar shows that pool's allocation as Unknown (data quality Usage unknown) rather than 0%, produces no runway or projection for it, raises no capacity finding from it, and leaves it out of the facility used/total figures, which note how many pools were excluded.

An unavailable estimate does not mean unlimited headroom. Thresholds already exceeded always alert, regardless of forecast readiness — a pool over its warning or hard percentage is flagged from the raw allocation reading alone, independent of any history requirement.

A Facility Admin or Organization Admin can change capacity, planning lead time, and rapid-growth settings in Settings → Capacity & risk. Set the planning lead time to how long your facility actually needs to purchase, migrate, or archive — that is what drives the warning severity on a tightening pool and the fleet and Capacity page Warning status. Context Mapping can add project and department labels to help identify growth contributors.

A warning threshold is a planning policy. It is not a universal point at which ZFS performance changes. Hardware, fragmentation, workload, and maintenance all affect behavior.

1. Confirm the pool and the datasets responsible for recent growth — a dataset breakdown marked “explanatory only” means it doesn't reconcile with the pool's own usage and should be treated as directional, not exact.
2. Agree on an archive, expansion, or ingest plan with the storage owner; Pillar does not move or delete media for you.
3. Recheck capacity after the change and review capacity-forecast routing in Alerts & integrations.