Longhorn UI · Developed by us

Show the shards of a sharded V2 volume

Longhorn's V2 data engine can now shard a volume. It splits the data into data and parity pieces (erasure coding) and puts each piece on a disk on a different node, so no node holds a full copy. The UI didn't know about this: a sharded volume's page showed an empty Replicas card and the list showed "0 / 1". We added a read-only view of the shards. As the Longhorn maintainers asked, sharded volumes are still set up outside the UI.

Built on a branch · no PR yet Issue 11689 ↗ Backend issue 1061 ↗ Branch ↗

What changes for the user

A sharded volume's page has a Shards card instead of an empty Replicas card. There is one tile per shard, with its state, node, disk, size and whether it holds Data or Parity.
Volume Details shows how the data is split: Data Layout, Data Chunks, Parity Chunks and Strip Size. The help on Parity Chunks says how many shards can fail at the same time.
A failed shard turns red, and hovering it shows when it failed. A replacement shard turns yellow and shows its rebuild progress.
The volume list shows "3 / 3 shards" for a sharded volume instead of "0 / 1". Replicated volumes look the same as before.
Nothing here changes a volume. There are no new buttons or forms.

Needs a decision

Column name: the list column is still called "Replicas (Running/Desired)", and for a sharded volume we write "shards" after the numbers. Is that enough, or should the column get a new name?
Replica actions: a sharded volume's menu still offers replica actions such as Update Replicas Count, which don't apply to it. Should we hide them in this change or in a follow-up?

The flow

The volume page shows the shards

The empty Replicas card becomes a Shards card, and Volume Details says how the data is split.

Volume page before and after: empty Replicas card becomes a Shards card with three tiles; four new rows in Volume Details — This volume has 2 data shards and 1 parity shard, each on a different node. Before, the page didn't say where its data was. Hovering a shard's title shows its full name, and hovering the disk shows the disk's ID.

Problems are easy to spot

The tiles use the same colours as replica tiles: blue when healthy, red when failed, yellow while rebuilding.

Shards card with one healthy, one failed and one rebuilding shard with a progress bar — Our test volume stayed healthy. To show these states we changed the shard data in the browser and didn't break a disk on the shared cluster. With one parity shard, this volume can lose one shard and keep working.

The volume list counts shards

In the optional Replicas column, a sharded volume shows its healthy shards out of the total.

Volume list before and after: 0 / 1 becomes 3 / 3 shards for the sharded volume — Before, "0 / 1" made a healthy sharded volume look like it had lost its only replica. The other volumes in the list don't change.

Video (optional)

0:00 Before · 0:19 After · 0:45 Failed and rebuilding shards (simulated) · 1:00 Volume list

Notes

The issue is assigned to another Longhorn UI developer, so we haven't opened a PR yet.
Not covered yet: the Node page's disk replica count leaves out shards, and shard events don't show in the volume's event list.
Creating a sharded volume or changing its layout stays outside the UI. The Longhorn maintainers asked for display only, for now.