:hedgehog: PostHog is the leading platform for building self-driving products. Our developer tools – AI observability, analytics, session replay, flags, experiments, error tracking, logs, and more – capture all the context agents need to diagnose problems, uncover opportunities, and ship fixes. Steer it all from Slack, web, desktop, or the MCP.
Hobby: moving object storage off MinIO
PostHog's hobby stack used two object stores: MinIO on objectstorage:19000 for general storage, and SeaweedFS on seaweedfs:8333 for session replay v2.
MinIO stopped publishing Docker images, so the objectstorage service now runs SeaweedFS as well.
SeaweedFS cannot read the MinIO on-disk format, and there is no in-place conversion.
The service starts on a new objectstorage-data volume, and the MinIO data stays where it is, on the objectstorage volume.
What becomes unavailable
Until you copy the objects across, PostHog cannot read:
- exported assets (CSV, PNG, PDF, video)
- uploaded media, such as images in notebooks
- error tracking symbol sets, so stack traces stay unsymbolicated
- session recordings that are still in MinIO
Postgres and ClickHouse are untouched. Events, persons, insights, dashboards, feature flags and experiments all survive the upgrade.
Option 1: copy the objects after the upgrade
This is the usual path.
bin/upgrade-hobby asks you to acknowledge the change before it pulls the new version, and it keeps the old volume.
Upgrade first, wait for the stack to come up, then run this from the directory that holds your docker-compose.yml:
./posthog/bin/migrate-storage-hobby --dry-run # see what would be copied
./posthog/bin/migrate-storage-hobby # copy it
The script starts a temporary MinIO container on the stack's network, mounted on the old volume, and copies bucket contents through the running services. It copies:
- the whole
posthogbucket intoobjectstorage, so a prefix that no list here mentions still comes across - the
ai-blobsbucket intoobjectstorage - session recordings into
seaweedfs - symbol sets into
seaweedfsas well, because cymbal reads them there while Django reads them fromobjectstorage
The query cache is the one thing left behind, because PostHog rebuilds it.
The copy skips objects that already exist in the destination, so it is safe to run again after a failure. It exits non-zero if any object failed, and it only suggests deleting the old volume after a complete run.
Check your data in PostHog before you free the disk space:
docker volume rm <project>_objectstorage
The script prints the volume name it read.
It picks that volume by name (<project>_objectstorage) and refuses to guess when a host holds more than one, so pass --volume <id> if you run several installs on one machine.
Upgrading through the installer instead
bin/hobby-installer does not run bin/upgrade-hobby, so it cannot show that prompt.
Its preflight checks warn instead, whenever the legacy volume is still on the host, and ask you to confirm before the upgrade continues.
The warning clears once you copy the objects across and delete the old volume.
Upgrade rendering checks
The hobby smoke test starts the Go installer, which copies the Compose file directly.
It does not execute bin/upgrade-hobby or its environment substitution step.
The separate Test hobby upgrade rendering CI job exercises bin/helpers/render-hobby-compose.sh, which the upgrade script calls.
It checks that deployment values render while container shell variables and Compose defaults stay intact.
Option 2: stay on your current version
Answer N at the upgrade prompt.
Nothing changes, and you keep a running MinIO.
Copy the posthog bucket out with any S3 client against http://127.0.0.1:19000, with access key object_storage_root_user and secret object_storage_root_password, then upgrade.
This option matters most if you cannot pull quay.io/minio/minio, which option 1 needs.
If the copy does not work
The temporary container reads a pinned MinIO release from Quay. MinIO archived its source repository in October 2025, so that image can disappear. If the pull fails, the objects are still on the volume. Mount it on any host that can still run MinIO, start MinIO against it, and copy the bucket out with an S3 client.
Why the anonymous identity is off in hobby
Dev and CI register an anonymous SeaweedFS identity with Admin rights, so unsigned clients keep working.
Hobby does not, because Caddy proxies /posthog/* straight to objectstorage.
An anonymous identity there makes every object readable and writable from the internet.
Only the posthog identity is registered, so SeaweedFS rejects unsigned requests and the presigned URLs Django hands out keep working.