Apps with stored data
On a cluster, an app with a volume moves only while the cluster replicates its data, and otherwise waits for its server or comes back from a backup.
An app that keeps data on its server cannot simply start somewhere else: its files are not there. On a cluster it moves only when they already are.
| The app keeps | When its server stops |
|---|---|
| Nothing | It moves on its own. See When a server stops |
| A Volume, a File written in Storage, or a Compose app's own volume, with replication on | It moves on its own, with its data replicated there as it changed. See Replication |
| The same, with replication off (the default) | It waits for its server. You can bring it back from a backup |
| A Bind | It never moves |
The app's Overview has a Failover card that says which one applies.
With replication on
Every change of the app's data streams to the cluster's other servers as it happens, and a failover puts the copy already on the server that takes the app in place. Turning it on is one switch per cluster: see Replication.
It gets a backup
An app with stored data on a cluster also gets the same Failover backup as a database: every 15 minutes, to a bucket first, else another server of the cluster, never the server it runs on.
- It is created when the app is deployed on the cluster, or arrives because its server joined.
- An app that already has a backup schedule keeps its own.
It is what bringing the app back restores, and where the data a returning server still held is saved.
Without replication it waits
web waits for server-a if it stops: cluster eu does not copy its data. You can bring it back on another server from its latest backup.
Waiting loses nothing: when the server is back, the app carries on with every file it had.
Bring it back from a backup
While the server is stopped, the app's page says web is waiting for server-a and offers one button, naming the backup it would restore:
Bring web back on another server from the backup of 14:15
It needs Restore backups. The dialog asks "Bring web back from a backup?" and warns: "Anything it wrote after that backup is lost." Confirm with Bring it back.
Activity records Asked cluster eu to bring web back on another server from the backup of 14:15,
then by Deplo Restored web on server-b from the backup of 14:15.
When the click is refused
| It says | Why |
|---|---|
| "web's server may still be running it. Deplo never starts a second copy next to one that could be writing: try again in a minute." | About 25 seconds have not passed since the server went quiet. In the slower case it is longer |
| "web has no backup or copy another server can read." | Its backups are on its own server, or on a server outside the cluster |
| "No other server of the cluster has room for web." | Add a server or lower its memory |
| "No other server of the cluster is online to take web." | Bring one back |
| "The cluster's leader cannot reach its servers right now. Try again in a minute." | The leader cannot reach the servers, or too few of them answer it. See When a cluster is cut in two |
A move that fails later is an Activity entry: Could not bring web back on another server. web did not start on another server.
An app with a Bind never fails over
web never fails over: it uses a folder that lives on server-a, and only on that machine.
Replication does not copy it either. Keep the data in a Volume, or in a database, to change that.
When the server comes back
What the old server still held is saved as a backup before its copy is removed, where the app's backups already go: a bucket, or another server of the cluster. See Data left on the server.
Over the API: bringAppBack and App.failover. Over MCP: bring_app_back.
Did this page help you?