The leader
The part of a cluster that moves apps off a server that stopped - where it runs, how to move it, and how it is updated.
Every cluster has a leader: it decides which server runs what, and it is what moves apps off a server that stopped. You never install or update it by hand.
- It starts on Deplo's own machine, the moment the cluster is created.
- With the leader on Deplo's machine and that machine down, nothing fails over.
- Failover works while Deplo is down only when both hold: the leader runs on a server of the cluster, and the cluster has three servers or more, more than half of them up.
- The leader keeps leading only while enough servers answer it: more than half of the cluster's servers, or exactly half of them and Deplo itself. Otherwise it moves nothing. See When a cluster is cut in two.
Reading the banners
Open the cluster under Settings -> Servers. The server the leader runs on shows Leader. When the leader runs on Deplo's own machine, no server shows it.
No banner means the leader runs and enough servers answer it. Otherwise the banner Failover is not working right now says why:
| It says | What to do |
|---|---|
| "Only 1 of 3 servers answers this leader, so nothing moves. Failover works again once 2 do." | Bring the servers back. See When a cluster is cut in two |
| "The leader is installed but not running." or "The leader is not installed yet." | Try again starts it |
| "Deplo could not reach the server the leader runs on." | Try again, or bring that server back |
| "Deplo's own machine is not connected as a server yet, so there is nowhere to start the leader." | Connect Deplo's own machine as a server |
| "This leader is older than this Deplo accepts. Deplo updates it on its own." | Nothing: wait |
Deplo checks every 30 seconds, and at once after you move or start the leader.
Run the leader on another server
Why: so that failover keeps working while Deplo's own machine is down. That takes three servers or more. The confirmation says which case you are in:
| Cluster | It says |
|---|---|
| Three servers or more | "Failover keeps working while Deplo's own machine is down." |
| One or two servers | "Under three servers, failover still needs Deplo up wherever the leader runs." |
Even with three servers, a leader on another server keeps failover working only as long as it still reaches more than half of the cluster's servers.
Open the cluster
Settings -> Servers, then the cluster's folder.
Make leader
In the menu of the server that should run it, Make leader. The confirmation reads Run the leader on server-b?
- An offline server cannot be picked: "server-b is offline. Bring it back online first."
- A server with no proxy carries a note: "server-b has no proxy, so a leader there still needs Deplo's machine up to move anything."
- A server that cannot run a leader yet is refused: "The server agent on this server is too old to run a cluster leader. Deplo updates it on its own: try again in a few minutes."
Confirm
There is no command to run: that server installs the leader on its own. Toast: "The leader now
runs on server-b". Activity: Moved the leader of cluster eu to server-b.
To move it back, open the menu of the server that shows Leader and pick Move leader to Deplo's machine. When Deplo's own server is in the cluster, use Make leader on it instead. Instance admins only.
Pick a server that answers on port 443
The leader reaches the other servers over the cluster's private network, directly wherever two servers can reach each other, so it keeps leading without Deplo's machine.
A server that runs Deplo's proxy (it answers on 443) also runs a second cluster relay beside the leader, for the pairs of servers that cannot reach each other directly. Every server connects to it too.
A server with no proxy runs the leader alone. A pair that can only meet through a relay then still needs Deplo's machine, and the confirmation says so before you pick it.
When Cloudflare refuses Deplo's token
When the panel is served through Cloudflare Tunnel, servers reach the relay on Deplo's machine through that tunnel. Deplo adds the route to it, which takes the panel's Cloudflare API token.
If Cloudflare refuses that token, the banner Failover may stop working says: "Cloudflare refused Deplo's change, so servers cannot reach the relay on Deplo's machine and failover stops when the leader's server does. Give Deplo a working Cloudflare API token:", then Cloudflare's reason. With a backup address on, servers still reach the relay through it, and the banner says so.
To fix it, Fix in Cloudflare beside the banner:
Create the token
Create the token in Cloudflare opens Cloudflare's form with the three permissions filled
in, and a dated name like Deplo panel 2026-11-14. Continue to summary, Create Token,
then copy it.
Paste it and save
Paste it under New token, then Save. Deplo keeps it only if Cloudflare accepts it for the tunnel, its DNS zone and the relay route. Otherwise it says why and keeps the old token.
Delete the old token
Deplo now uses the new token, the route is back and the banner goes away. Open Cloudflare tokens lists your tokens: delete the Deplo panel token that is not the new one. Deplo does not delete it for you.
Activity: Gave the panel's Cloudflare Tunnel a new API token. Instance admins only.
When the leader's server stops answering
Deplo's own machine leads in its place, on its own, 25 to 40 seconds after the leader's server stopped answering.
-
The banner Deplo's own machine leads this cluster says: "Until server-b answers again."
-
Activity:
server-b stopped answering, so Deplo's machine leads cluster eu until it answers again, thenserver-b answers again, so it leads cluster eu again. -
Nothing runs twice during a change of hands, and a change of hands never causes a failover by itself: a leader that takes over moves nothing of a server that answers it.
-
For a server it has not heard from at all, it starts that server's databases and apps with a Volume elsewhere about 25 seconds after it started leading, once the servers that are left say no extra time it was given can still be running. In two cases it waits about 6 minutes 21 seconds instead. See The slower case.
-
Anything that was waiting for room is not started a second time. The new leader first hears from every server, and leaves each app with the server that still has it.
The limits. With the leader's server offline and Deplo's machine down at the same time, nothing fails over. Apps keep running where they are. A server that stops in that window is not replaced until either one is back. The same holds for a leader that reaches too few servers: with half of them or fewer answering it and Deplo out of reach, it moves nothing.
When Deplo's own server is in the cluster
Deplo's own server can be a member of a cluster like any server: drag it onto the cluster, its apps and databases come along, and it can be removed. This is how a first cluster starts with one rented server.
Deplo itself, its database and its proxy are never part of the cluster. Nothing places, moves, stops or empties them.
While Deplo's own server is a member and the leader runs on it, the banner Failover may stop working says:
If Deplo's own server stops, its Apps cannot move. Run the leader on another server to fix this.
The leader is what moves apps off a server that stopped, and here it stops with that server. Make leader on any other server of the cluster makes the warning go away.
On a cluster of one or two servers the warning reads differently, because moving the leader alone does not fix it:
If Deplo's own server stops, its Apps cannot move. Fixing this takes three servers, with the leader on another one.
Removing Deplo's own server leaves the leader and the relay running there.
Updates
Nobody updates a leader or a relay by hand. After a Deplo update, Deplo brings each one to the version it ships with, on its own: one server at a time, a leader on another server first, Deplo's own machine last. It only ever moves forward.
- A leader update costs no failover. Before it restarts, the leader gives every server of its clusters five more minutes, so nothing stops and nothing moves while it is away. Five minutes is the most a server ever accepts.
- A server whose server agent is still being updated gets that update first, then the leader's.
- Activity, by Deplo:
Updated the cluster leader and relay on server-b to v0.3.0.
When an update fails
Deplo tries again on its own, waiting 1, 2, 4 and up to 60 minutes between tries. Until it works, the banner Failover may stop working shows "Deplo could not update the leader on server-b to v0.3.0:", then the reason, then "It retries on its own."
Two reasons you can meet:
- "A server of cluster eu is not answering, and a leader is not restarted while its Apps may be moving." Deplo waits until that server answers, or 45 minutes have passed.
- "It took the new version but did not start on it. The other servers wait until it does." No other leader is updated until this one runs again.
A leader too old for this Deplo
The banner Failover is not working right now says: "This leader is older than this Deplo accepts. Deplo updates it on its own." When that leader runs on another server, Deplo's own machine leads the cluster meanwhile, and the banner Deplo's own machine leads this cluster says: "Until server-b is updated."
There is no update button, and none is needed. Try again is only offered for a leader that is not running.
Good to know
- A server that runs a cluster's leader cannot be removed from Deplo: "Move the leader of cluster eu to another server first."
- After moving Deplo to another machine, the new machine starts its own leader, and a leader on another server keeps leading.
- What each leader needs to restart an app (its last deployed version and variables) is kept sealed for that leader alone. Moving the leader hands it over to the new one.
- Deleting the last cluster removes the leader and the relay from Deplo's own machine.
- A leader left on a server that was offline when its cluster was deleted, or when the leader moved away, is removed as soon as that server answers.
- A server of a cluster takes orders only from the leader Deplo named for that cluster.
- Over the API:
moveClusterLeader(id, serverId),Cluster.leaderandCluster.leaderChoices, instance admin only. There is no MCP tool for it.
Did this page help you?
Move work between servers, remove a server, delete a cluster
Put an app or database on one server of a cluster, take a server out of a cluster without downtime, or delete the cluster and keep everything running.
Update
Deplo updates itself at night, or in one click, with its server agents. Pick when, read the update log, choose stable or canary releases, and recover an update that did not take.