s3du

s3du

du for S3. Scans bucket in parallel, groups by prefix and storage class, keeps everything in a compressed radix-tree snapshot you can re-open without re-paying the list cost.

Install

# The browser
go install github.com/ochaton/s3du@latest

# Companion: parallel batched delete by prefix (the rm -rf for S3)
go install github.com/ochaton/s3du/cmd/s3rmrf@latest

Or build from source:

git clone https://github.com/ochaton/s3du && cd s3du
go build -o s3du .

Run

# Scan a bucket and drop into the TUI
s3du -bucket my-bucket -i

# Re-open the snapshot without re-scanning (default path resolved automatically)
s3du -bucket my-bucket -load -i

# Snapshot + print stats only
s3du -snapshot path -load -stats

# Tune parallelism, override region
s3du -bucket my-bucket -workers 64 -region us-west-2

AWS credentials come from the env/profile chain (AWS_PROFILE, ~/.aws/credentials, IAM role, etc). The bucket's region is auto-detected via HeadBucket when -region is left blank.

Snapshot

Each scan writes a binary snapshot under the platform's user-cache dir (os.UserCacheDir):

Linux : ~/.cache/s3du/<bucket>@<region>/tree.snap
macOS : ~/Library/Caches/s3du/<bucket>@<region>/tree.snap

Re-open with -load -snapshot <path> — no re-scan, full TUI / -stats / -i flows work against the loaded snapshot.

TUI keybindings

Navigation       Sort                 Display
↑/k  ↓/j         s  by size           g  cycle bar widget
Home  End/G      n  by name           ?  show this help
PgUp PgDn        C  by object count
Enter / l        $  by monthly cost
Bksp / h         t  toggle dirs first
                                       q  quit

g cycles: no bar → bar only (default) → bar + percentage → percentage only. Bar is normalised to the listing total (entries sum to 100%).

Logs

Structured slog logs are off by default — pass -log <path> to enable file logging. SDK retry events (SlowDown, adaptive throttling) surface through the file logger so you can diagnose a scan that goes quiet.

Regenerating demo.gif

The README animation comes from a fully synthetic snapshot — no real bucket data. To re-record:

brew install vhs ffmpeg ttyd   # one-time
go install .                   # current s3du on $PATH
go run ./internal/cmd/gen-demo # writes /tmp/s3du-demo.snap
vhs demo.tape                  # writes demo.gif

The synthetic dataset has ~7 000 objects across datasets/, logs/, archive/, reports/, and misc/ prefixes, mixing STANDARD, STANDARD_IA, INTELLIGENT_TIERING, GLACIER, DEEP_ARCHIVE, and ONEZONE_IA storage classes. Total ~370 GiB.