Upgrading

The version of record is one line: the image in docker-compose.yml. If the sample on this page drifts from it, CI fails (scripts/check-doc-refs.sh calls scripts/check-pin.sh).

    image: docker.io/chrislusf/seaweedfs:4.46

Steps

1. Change the version in docker-compose.yml and in the sample on this page
2. Run scripts/verify.sh. It checks the behaviours the consumers depend on one at a time, in throwaway containers that use etc/’s entrypoint as it stands — reading and writing with credentials, the shape of anonymous Read, a wrong secret key, presigned URLs, the key length limit (255 bytes), bucket isolation, and writing after the bucket has been emptied and the store restarted. The exit code is the number of reds
3. Tag it, and bump the submodule in each consumer. Recreate the dev store

sh scripts/verify.sh          # the version docker-compose.yml pins
sh scripts/verify.sh 4.23     # a version you name

verify.sh needs container (Apple Container), aws (AWS CLI v2) and curl. It publishes S3 on 127.0.0.1:18333 (VERIFY_PORT changes it), so it runs with other stacks up. When it is done — including when it dies partway — it removes the container and the volume.

Do not go back below 4.27

On 4.26 and earlier, the startup check misreads a volume whose last record is a delete as corrupt and loads it read-only (upstream #9563, fixed in 4.27). On a stack with only one volume behind the bucket, emptying the bucket and restarting is by itself enough to stop writes. sh scripts/verify.sh 4.23 goes red on that one line and nothing else.