Running MinIO AIStor in Podman

A single node on a laptop, with diagrams. Four drives, a license, and the two traps that catch everyone.

AIStor is MinIO’s object store for AI workloads. The container image is public. The license is not. You can pull and start the server in a minute, but it will refuse every S3 call until you feed it a license file.

Here is the whole shape of it before any commands.

client mc, aws-sdk :9000 :9001 podman container minio server S3 API + WebUI minio.license read only certs/ TLS, optional erasure set /mnt/data1 /mnt/data2 /mnt/data3 /mnt/data4 one volume each One node, four drives, two mounted files.

What you need

Podman, and a license file from MinIO. Nothing else. The image lives on a public registry, so there is no podman login step.

Pull the imagecopy_all
podman pull quay.io/minio/aistor/minio:RELEASE.2026-08-07T18-34-35Z

That is about 392 MB.

Trap one: drives on the root filesystem

Point the server at four directories on your normal disk and it will reject all four. Every drive goes offline and nothing serves.

What that looks likecopy_all
Error: Drive /tmp/d3 returned an unexpected error:
       drive is part of root drive, will not be used,
       please investigate - drive will be offline

MinIO checks whether a drive sits on the root filesystem, and refuses the ones that do. Give it a volume per drive instead.

Make four volumescopy_all
for i in 1 2 3 4; do podman volume create aistor-data$i; done

A bind mount from the host works too. A plain directory inside the container does not.

Run it

Start a single nodecopy_all
podman run -d --name aistor \
  -p 9000:9000 -p 9001:9001 \
  -v aistor-data1:/mnt/data1 \
  -v aistor-data2:/mnt/data2 \
  -v aistor-data3:/mnt/data3 \
  -v aistor-data4:/mnt/data4 \
  -v "$HOME/minio/minio.license:/minio.license:ro" \
  -e MINIO_ROOT_USER=admin \
  -e MINIO_ROOT_PASSWORD='choose-something-long' \
  quay.io/minio/aistor/minio:RELEASE.2026-08-07T18-34-35Z \
  server /mnt/data{1...4} --console-address :9001 --license /minio.license

Two things in that command are worth a second look.

/mnt/data{1...4} is MinIO’s own shorthand, not your shell’s. It expands to four drives. You can go up to {1...64}.

--console-address :9001 is not optional. The console binds to port 0 by default, which means a random port. Name it or you will not find it.

What a good start looks like

First runcopy_all
INFO: Found unformatted pool - pool-0 (will attempt formatting with quorum: 3)
INFO: Successfully formatted pool pool-0 (quorum: 3)
INFO: First-time cluster setup completed successfully:
INFO:   Pool-0: 4/4 drives formatted and online
MinIO AIStor Server
API: http://127.0.0.1:9000
WebUI: http://127.0.0.1:9001

Four of four drives, online. The pool formats once and remembers.

Trap two: no license, no service

Leave the license out and the server still starts. It looks healthy. It is not.

The line that matterscopy_all
WARN: No valid license found, running in offline mode.
      All S3 operations are denied.
      Please install a valid license using 'mc license update' to restore service.

This is the one that wastes an afternoon. The server is up, the console answers, the drives are formatted, and every single bucket call fails. Read the startup log.

server starts drives formatted no license license offline mode all S3 calls denied serving buckets, objects, UI mc alias set The server starts either way. Only one branch serves data.

Where everything lives

What Where Note
S3 API :9000 the default, set with --address
WebUI :9001 set it, or you get a random port
Drives /mnt/data1 to /mnt/data4 one volume each
License /minio.license path passed to --license
Certs /root/.minio/certs default, set with --certs-dir

Drives and parity

The four drives form one erasure set. MinIO splits each object into shards and spreads them across every drive in the set. Some shards are parity.

object split on write shard shard shard shard /mnt/data1 /mnt/data2 /mnt/data3 /mnt/data4 one erasure set lose a drive, the object survives One object, four shards, one shard per drive.

The formatting log says quorum: 3. Three of the four drives have to agree before the pool will format. The same idea protects reads later.

Adding TLS

Drop a certificate and key into a directory and mount it.

Certs layoutcopy_all
certs/
├── public.crt
└── private.key
Mount itcopy_all
-v "$HOME/minio/certs:/root/.minio/certs" \
-e MINIO_SERVER_URL=https://localhost:9000 \
-e MINIO_BROWSER_REDIRECT_URL=https://localhost:9001

Set both URLs. The console builds links from them. Miss them and it will send your browser to http:// and the redirect will fail.

Check it

Point mc at itcopy_all
mc alias set local http://localhost:9000 admin 'choose-something-long'
mc mb local/first-bucket
mc ls local

If mc mb fails and the server looks fine, go back and read the startup log for the license warning.

Clean up

Remove the node and its drivescopy_all
podman rm -f aistor
for i in 1 2 3 4; do podman volume rm aistor-data$i; done

The volumes hold the formatted pool. Remove them and the next start formats a fresh one.


Everything above came from the image itself: minio server --help for the flags, and a throwaway container for the two error messages. The full documentation is at docs.min.io.