Large R2 bucket storage (storage cost)
Part of the Workers, Pages & API tokens check · fix arrives as a guide
What it is
An R2 bucket holds a large volume of stored data.
Why it matters
R2 bills on storage, so a bucket that only grows is a line item that only grows. Buckets accumulating build artifacts, logs, or old uploads are the usual cause, and nothing prompts a review.
How R2 charges — and where the surprise lives
R2 bills three meters: stored gigabytes per month, Class A operations (writes, lists) and Class B operations (reads). Egress is free — the reason teams pick R2 — which means the bill is almost entirely a function of how much you store and how often you touch it. Storage cost is a ratchet: every object you never delete bills again every month, so a bucket that only ever grows produces a bill that only ever grows.
The two classic silent accumulators
Buckets that get large without anyone deciding to store that much usually contain one of these:
1. Build artifacts and backups on a schedule with no expiry —
nightly dumps, per-deploy bundles, log exports. Each one small;
three years of them, not.
2. Incomplete multipart uploads — a client that crashed mid-upload
leaves its parts behind, invisible in object listings but billed
as storage, forever, unless a rule aborts them.See what a bucket actually holds
# object count and total size (dashboard: R2 → bucket → Metrics)
wrangler r2 bucket info YOUR_BUCKET
# spot-check the oldest objects — are they still worth paying for?
wrangler r2 object list YOUR_BUCKET --limit 20Fix it manually — lifecycle rules
R2 lifecycle rules delete or expire objects on a schedule, per bucket, optionally scoped by prefix (R2 → bucket → Settings → Object lifecycle rules). The two rules almost every bucket wants:
Rule 1: abort incomplete multipart uploads after 7 days
Rule 2: expire objects under a dated prefix past its useful life,
e.g. backups/ after 90 days, build-artifacts/ after 30Keep what you would actually restore from; a backup nobody would dare restore is storage cost wearing a safety blanket.
How lumioguard fixes it
The scan reads bucket sizes and lifecycle configuration on every run and flags growth with no expiry rules behind it. The fix arrives as a guide with the lifecycle rules matched to the prefixes actually present in the bucket.
Run them all on your app
Connect your repo and your live services with read-only scopes. The first scan is free, and nothing changes without your approval.