Managing servers
This page explains how to create and claim servers in the Felis panel, control when they start, stop and hibernate, and use the console, player management, the file manager, scheduled tasks and backups. It is written for server owners, Admins and the platform Owner.
Permissions
What you can do depends on your account role and on who owns the server. In the table below, "server owner" is the account that claimed or created the server; "Owner" is the platform Owner role, which has every Admin right.
| Action | Who |
|---|---|
| Create a server | Admin; an account with a server quota can create servers of its own, see Self-service creation |
| Change server settings | Admin; server owners can change their own servers, except the image |
| Claim an unowned server | Any account with a linked Minecraft account |
| Start a server | Server owner, Admin; other players as the autostart policy allows |
| Stop, restart, force stop, console, player management, files, scheduled tasks, backups and restore | Server owner, Admin |
| Give up a server | Server owner, Admin |
| Delete a server | Admin |
| Forwarding mode, emergency stop | Owner |
Admins see every server on the platform on the Servers page, with its owner, online players and autostart policy. Users see the servers they own and the servers they can claim. For the roles, see Roles.
Server list
The Servers page lists the servers you can see. You can search the list, filter it by status, and sort it by name, status or number of online players. The list also offers the following features.
Servers per page
The list shows 20 servers per page by default. To change this, select 50 or 100 under Servers per page above the list. The selected number is saved in the current browser and used the next time the page opens.
Tags
Tags group servers. Server owners and Admins add or remove tags in the Tags row of the Settings tab: type a tag, then press Enter or type a comma to add it.
- A server can have up to 6 tags. Each tag has 1 to 20 characters and cannot contain commas or control characters.
- Tags that differ only in letter case are treated as the same tag, and the spelling entered first is kept.
- Tags appear below the server name in the list, and only server owners and Admins can see them.
- When a server is claimed by another account, its existing tags are cleared.
- System services, such as the login space and the lobby, cannot have tags.
The Tags filter above the list shows every tag that appears in the current list. Select a tag to filter by it, and select it again to clear the filter. When several tags are selected, only servers that carry all of them are shown.
Batch start, stop and restart
Tick the checkbox at the start of a server's row, or the checkbox in the header to select every server on the current page. A batch action bar then appears above the list. Admins can select any server, and other users can select only the servers they own; system services cannot be selected. A batch action applies only to the selected servers within the current filtered results.
- Wake runs immediately.
- Stop and Restart first ask for one confirmation. The confirmation lists the servers with players online and their player counts. When no player is online, it states only how many servers will be affected.
- Servers are processed one at a time in list order, and the batch action bar shows the progress, for example “Processing server 2 of 5”.
When every server has been processed, the results appear below the batch action bar. Each server is listed as done or with the reason it failed, and a summary states how many completed and how many failed. Failed servers stay selected so that you can run the action again, and successful servers are deselected. Each server follows the same rules as a single action: for example, a server pending reclamation cannot start, and a server under maintenance refuses a restart. Batch actions do not include giving up or deleting servers.
Creating a server
A server an Admin creates is stopped and belongs by default to the Admin who created it. It can instead be created without an owner, in which case it belongs to no account until a user claims it; see Claiming a server. An account with a server quota can create servers it owns itself; see Self-service creation.
Open Servers and select New server.
Under Runs on, select Server software or Whitelisted image. Regular accounts can select server software only.
Fill in the fields:
Field Description Server software The core, game version, build or loader version, and Java version. See Server software. Image With Whitelisted image, an image from the image whitelist. See Images and Minecraft versions. Name 3–32 lowercase letters, digits or hyphens, with no hyphen at either end. login,lobby,admin,panel,console,api,felis,velocity,registry,internalandwwware reserved by the platform.Subdomain Same rules as the name. Players reach the server at <subdomain>.<root>.Display name (optional) The name shown in the panel and the lobby menu. Server owner Shown to Admins only; the current Admin by default. With Nobody — a player can claim it, the server belongs to no account until a user claims it. The Owner can also select Another account and choose the owner from the account list. Execution node Shown to Admins only, in a multi-node deployment: an online, approved node. See Multi-node and high availability. Memory The container memory limit in GiB: at least 1, decimals such as 1.5 allowed; through the API, at least 768 MiB when it is set or changed. The Java heap is sized from it. When an Admin creates the server, the default is the smaller of 4 and the memory that can still be allocated, rounded down to a multiple of 0.5 GiB, and the field shows how much can still be allocated; see below. For self-service creation, see Self-service creation. Storage The space the world files may use, in whole GiB: 10 by default. It can be changed after creation. See Disk usage and quota. Autostart policy Defaults to Wake list when an Admin creates the server, and to Owner only — wake from the panel for self-service creation. See Autostart policy. With server software, accept the Minecraft End User License Agreement (EULA).
Select Create.
A new server hibernates after 600 seconds with nobody online; see Idle handling.
The memory that can still be allocated is the node's allocatable memory less the memory already requested on it by the login space, the lobby, the control plane components and other servers. A game server requests its full memory limit, so a server whose memory exceeds that figure can still be created but cannot start until memory is freed; the field says so in a warning style. In a multi-node deployment the figure belongs to the selected node, or, before a node is selected, to the node with the most memory left. When the platform cannot read the figure, the memory defaults to 4 and the field does not show it.
A server with an owner gets scheduled restore points while it is in use, and its owner is warned before the world is reclaimed for inactivity; see World cleanup. A server without an owner gets no scheduled restore points until it is claimed, and has no account to warn. An owner chosen when an Admin creates the server is not limited by their quota; the server counts toward the owner's quota usage, with the same effect as Assigning a server. A server created without an owner charges no user's quota; quotas are charged when the server is claimed. A name that matches the leftover world volume of a deleted server is refused until that volume is removed or another name is chosen.
Self-service creation
New server appears on the Servers page, and the account can create servers it owns, when:
- It has a linked Minecraft account. See Linking your account.
- It has a server quota: the Owner set a quota for this account, or turned on the default quota under Self-service servers.
A self-created server runs server software only, and its resources count toward the account's quota. The panel shows the quota usage and what is left of it. The memory defaults to the smaller of 4 and what is left of the quota, rounded down to a multiple of 0.5 GiB, and a server whose memory or storage exceeds what is left cannot be created. Once the server count reaches its cap, New server is unavailable. In a multi-node deployment the server is placed on the approved node with the fewest user servers. Server owners can give up their servers; deleting a server remains an Admin action, see Giving up and deleting a server.
Images and Minecraft versions
A server that runs an image can use only enabled images from the image whitelist, which Admins maintain. See Image whitelist. Modpacks that players submit join the whitelist once they are reviewed and built. See Submissions.
- When a server is created, it is pinned to the build its image tag points to at that moment. When the tag later points to a new build, existing servers stay as they are.
- After you change the image on the Settings tab, the server uses the build the tag points to now; selecting the same tag again also selects its latest build. A Minecraft version change may upgrade the world on the next start, and upgraded chunks cannot be used by older versions. Before saving, tick I've backed up the world — change the image.
- The game proxy passes the player's identity to servers with modern forwarding. For Paper-based images, Felis writes the forwarding settings before every start. Other server software must read the environment variable
FELIS_FORWARDING_SECRETand enable Velocity modern forwarding itself, or use legacy forwarding. - The login space and the lobby run the platform's Minecraft version; the client version players need is shown on the Account page. A target server may need another version or mods, which its owner announces.
Claiming a server
A server that belongs to no account can be claimed. The claiming account becomes the server owner, with console, whitelist and backup management rights, and its Minecraft UUID is added to the server's access whitelist.
To claim a server:
- The account must have a linked Minecraft account. See Account linking.
- The claim must fit the account's quota. See Quotas.
- The server must not be pending release or deletion.
You can claim:
- In the panel: on Servers, select Claim on a claimable server and confirm.
- In game: join the server and run
/felis claim. - In the lobby menu: click the Claim & Start tile. See Joining.
For claim errors, see Account link and claim errors.
Server settings
Admins and server owners change the following directly on the server's Settings tab. The tab groups them into General, Server software, Resources, Start and run, Email notifications, Network and Advanced startup settings. Once something changes, Discard changes and Save Config appear at the bottom of the page; saving sends only the changed fields, and the rest keep their values. Before the page is closed or reloaded, the browser warns about unsaved changes.
- Display name and autostart policy.
- Server software. See Changing the core or version.
- Image (Admins only). See Images and Minecraft versions.
- Memory and CPU limit. Enter the CPU limit in cores (for example
1or1.5) or millicores (for example500m); leave it empty for no limit. - Storage. See Disk usage and quota.
- Idle handling. See Idle handling.
- Restarting after a crash and resuming after a host restart. See Crashes and host restarts.
- Email notifications. See Email notifications.
- JVM arguments, server arguments, environment variables and stop commands. See Advanced startup settings.
Raising the CPU, memory or storage of an owned server must fit the server owner's quota; lowering them is never held to it. The login space and the lobby are system services, which Admins manage under Login & lobby. See Customizing the login space and lobby.
Advanced startup settings
Server owners and Admins can select Show startup settings under Advanced startup settings on the Settings tab to read and change that server's JVM arguments, server arguments, environment variables and stop commands. Admins manage the login space and the lobby. Advanced startup settings are excluded from public server status and do not include platform-managed environment variables.
| Setting | Input and limits |
|---|---|
| JVM arguments | Enter one complete argument per item, up to 128 items and 1,024 bytes per item. Arguments cannot contain whitespace. Heap sizing options, including -Xmx and -Xms, and argument file options are rejected; the heap size is derived from the memory setting. |
| Server arguments | Each item is passed to the server as one argument, preserving spaces, order and empty strings. Up to 128 items and 2,048 bytes per item. |
| Environment variables | Enter names and values individually, up to 64 items. Names must be unique, use only English letters, digits and underscores, and cannot start with a digit; up to 128 bytes per name. Values may be empty and are limited to 4,096 bytes. |
| Stop commands | Commands run in the order entered, up to 16 items. Each item must be a single line of at most 1,024 bytes. Leading and trailing whitespace and one leading / are removed. |
All advanced settings together are limited to 64 KiB, measured in UTF-8 bytes. Environment variables cannot override FELIS_* or reserved runtime variables such as JAVA_MEMORY, JAVA_FLAGS, JAVA_TOOL_OPTIONS, RCON_PASSWORD, PATH and HOME. Removing an item and saving removes that value; unchanged settings keep their values.
Saved JVM arguments, server arguments and environment variables take effect on the next start or restart. Saving while the server is running or starting does not immediately restart it. On the next normal stop or explicit restart, stop commands are attempted over RCON in order after the world save attempt and before the process shuts down; the group has a total time limit of 30 seconds. Unavailable RCON, a command failure or a timeout does not prevent shutdown. Commands may run again after a failed attempt is retried. Force stop, emergency stop and automatic retries after a failed start skip these commands. Releasing, claiming or assigning the server clears its advanced startup settings; account migration preserves them.
Additional TCP and UDP ports
Server owners and administrators can add up to five entries under Extra ports in the Network section of the Settings tab. Select TCP or UDP and enter the internal port on which the plugin or mod listens. Internal ports range from 1 to 65535, and protocol/port pairs must be unique. Managed game, RCON and health-check TCP ports cannot be exposed.
The platform assigns external ports from 30000–32767. Once applied, each port shows a direct node address that can be copied. IPv6 addresses use brackets. Plugins listen on internal ports, and clients connect to external ports. For example, Simple Voice Chat defaults to UDP 24454, Dynmap to TCP 8123 and BlueMap to TCP 8100. Keep Simple Voice Chat's port set to the configured internal port, and set voice_host to the complete external address shown in the console. Open web maps at http://node-IP:external-port/.
Mappings can be changed while the server is running or stopped. Saving does not start or restart the game. Plugin configuration still takes effect as that plugin requires. The panel indicates pending configuration or an unavailable address until a direct address is confirmed. Wait for the confirmed address before configuring clients. Stopping a server retains its configured port allocations. Removing an entry revokes its mapping and releases the corresponding allocation.
Additional connections go directly to the node running the server, without using the existing Cloudflare Tunnel routes. The installer opens the default range in active firewalld or ufw configurations. Cloud security groups and NAT forwarding must also allow the relevant connections. Migration can change the direct node IP, so use the address currently shown in the console.
Disk usage and quota
The storage size selected when creating a server is also its disk quota, and it can later be changed on the Settings tab. The new storage must exceed the usage of the latest measurement. A server stopped because its disk was full can be started again once the storage is raised and the platform confirms a new measurement below the quota. The change only moves the disk quota the platform enforces; the world volume keeps the size requested at creation, so on a storage class that enforces capacity the storage should not exceed the volume's actual size. The Monitoring tab and the file page show usage, the quota and the latest measurement time, including while the server is stopped. Usage counts allocated bytes in the world volume, including worlds, plugins, configuration, logs and temporary files within that volume. Sparse files count only allocated space, hard links count once, and symbolic links are not followed.
At 90% of the quota, the console displays a warning. Server owners with problem notifications enabled also receive an email. At 100%, the platform stops the server through the normal shutdown process, records the DiskFull stop reason, and refuses starts, start retries and restarts. File management and downloads remain available for cleanup. Once the platform confirms a new successful measurement below the quota, the server can be started again; it does not start automatically.
The platform schedules read-only volume scans serially every minute. Updates may take longer with more servers, slow volume mounts or large directory trees. Protection takes effect after periodic detection, so writes may continue during measurement and normal shutdown. A failed, timed-out or incomplete scan preserves the last successful result. Measurements older than three minutes are marked stale. After cleanup, even a new measurement below the quota may still await confirmation. During this period, its timestamp is retained, current usage is shown as unavailable, and the startup restriction remains in effect. Usage without a successful measurement is unavailable rather than zero. Sampling failures do not release an existing full-disk startup block.
The file listing's available space is the free space on the underlying filesystem, separate from the server quota. Writes can fail when the underlying storage is full even if the server remains below its quota.
Start, stop and hibernation
A server is in one of these states:
| State | Meaning |
|---|---|
| Running | The server is ready and players can join. |
| Starting | Resources are being scheduled, files prepared and images pulled, or Minecraft is starting and being checked for readiness. |
| Stopping | The world is being saved and the server shut down. |
| Stopped | The server is shut down and uses no memory or CPU. |
| Hibernating | The world has been saved; the process is being frozen and its memory written to swap. |
| Hibernated | The process is frozen with its memory kept on disk and uses no CPU. Resuming continues from the frozen state. See Hibernation. |
| Timed out · retrying | The start timed out; the system is recreating the runtime instance and retrying. |
| Failed to start | The automatic retry limit has been reached. |
Starting
A server starts:
- When you select Wake at the top of the server page.
- When a player joins its address, as the autostart policy allows.
- When a player runs
/felis go <server>in game or clicks its tile in the lobby menu. - When a scheduled start task runs. See Scheduled tasks.
Every start is subject to the platform's running server limit and wake cooldown. See Startup and wake policy. A server pending release or deletion cannot start. While a backup, restore or file change is in progress, a start is refused with a maintenance notice.
While the server starts, its console page shows the current stage and diagnostics. When schedulable memory runs short, the page suggests reducing the server's memory, stopping other servers or adding node capacity. The default start budget is 300 seconds. After a timeout the system recreates the runtime instance up to 3 times, waiting longer before each attempt; if the server is still not ready, its state becomes Failed to start. Resolve the fault, then select Retry start. To investigate, see Server stuck in Starting and Server failed to start.
Stopping
Select Stop at the top of the server page. If players are online, the server sends them an in-game notice and disconnects them 30 seconds later before stopping; starting the server again within that window calls the stop off. When the online count cannot be read, the panel warns that players may still be connected and asks for confirmation.
Idle handling
A server whose online count stays at zero for the configured time hibernates or stops automatically, as selected; the next join wakes it according to the autostart policy. New servers hibernate after 600 seconds without players. Server owners and Admins turn Idle handling off on the Settings tab, choose how long the server waits without players, or choose the behaviour under Idle action. The online count is read over RCON; while it cannot be read, idle handling pauses and resumes once the count is readable again. The login space and the lobby never hibernate or stop when idle.
| Idle action | Behaviour |
|---|---|
| Hibernate (default) | Freezes the server process and writes its memory to swap. Resuming continues from the frozen state, with redstone circuits, entities and loaded chunks unchanged. A hibernated server uses no CPU and little memory. |
| Stop | Saves the world, ends the process and releases all memory and CPU. The world loads again on the next start. |
Hibernation
Hibernation has two phases, freezing and resuming. Each phase checks resources before it begins, and the server keeps its current state when they fall short:
- Before freezing, the platform turns off automatic world saving and writes the whole world to disk, then checks that the node's free swap can hold all of the server's memory. When swap is short, the world cannot be saved, or the memory cannot be fully written to swap, the process resumes and the server stops normally instead.
- Once frozen, the server's memory allocation shrinks to 128 MiB, so the node can give the rest to other servers.
- When resuming, the platform first reclaims the server's original memory allocation, then checks that the node's available memory can hold what is in swap. While memory is short, the process stays frozen and the console shows Resume is waiting for memory; resuming continues once memory is available. The process runs again only after all of its memory has been read back.
- Once the process runs again, automatic world saving is turned back on.
While a server is hibernated:
- The console shows the log from before the freeze and accepts no commands.
- The file manager is unavailable; resume or stop the server to manage its files.
- A backup reads the world saved before the freeze, and the server stays hibernated. Restoring a backup requires stopping the server.
- Stop ends the frozen process directly. The world was saved in full before the freeze, so stopping loses nothing that was saved.
- A scheduled stop stops the hibernated server, a scheduled restart skips it, and a scheduled start resumes it.
A frozen process exists only in the node's memory and swap. After a node restart, the platform keeps the world saved before the freeze and marks the server as stopped, with the reason in its lifecycle log. When a frozen process cannot resume, the platform starts the server again from the world saved before the freeze.
Hibernation relies on swap on the node. The installer creates a /swapfile as large as physical memory on nodes without swap; see What the platform uses. When the node's free swap cannot hold a server's memory, that server stops when idle instead.
A server that hibernates runs with the game's own hang watchdog turned off (max-tick-time set to -1, and Paper's watchdog disabled), so that the jump in time after resuming is not taken for a hang. The platform detects hangs instead: a server that does not answer RCON for 90 seconds is treated as crashed and handled according to Restarting after a crash.
Restarting
The top of the server page shows Restart while the server is running. Restarting saves and shuts down the current process, then starts the server again. The panel asks for confirmation when players are online or the player count cannot be read. Restart requests are refused during maintenance such as backups, restores or file changes.
Force stop
When the server process is unresponsive or a normal stop has not completed, the server owner or an Admin can select Force stop in the Danger zone of the Settings tab and confirm after reading the data loss warning. Force stop terminates the process immediately, without saving or waiting for the shutdown countdown or termination grace period. Online players will be disconnected. Changes since the last save may be lost.
After the request is accepted, the platform processes the stop and the panel continues updating the server state. The server cannot start again until the stop request has been processed. Force stop requires the operator; if the operator is unresponsive, the platform Owner uses the emergency stop below.
Emergency stop
When a normal stop does not respond, the Owner can use an emergency stop:
- On the server's Settings tab, select Emergency stop in the Danger zone.
- Type the server's name and verify your identity.
The emergency stop first saves the stop intent, then sets the workload's replicas directly to zero, bypassing the game process and the operator while keeping the normal termination grace period. World data is not deleted; unsaved changes may be lost. An accepted request does not prove that the process has exited: the panel keeps checking the game pod and shows Stopping until it has gone.
WARNING
An emergency stop disconnects every player on the server at once. When the control API is offline, the panel cannot act on the cluster: expand Host emergency shutdown procedure and run its commands on the host in order. Do not resolve a lost control connection by force-deleting worlds, pods or PVCs.
When the panel cannot read a server's state within 12 seconds, it shows the time of the last successful read and retries on its own. The state shown then is for reference only and does not prove that the game process is healthy.
Crashes and host restarts
Restarting after a crash
When the server process exits unexpectedly (including being killed for running out of memory) or stops responding, the platform restarts the server by default. Automatic restarts have a limit on crashes in a row, so a server that keeps crashing is not restarted forever:
- While the crashes in a row stay within the limit, the server is restarted.
- Once they exceed the limit, the server stops, and its console page shows why, together with the cause of the latest crash.
- A crash more than an hour after the previous one starts the count over.
Admins and server owners set this under Start and run on the Settings tab:
| Setting | Default | Description |
|---|---|---|
| Restart after a crash | On | When off, the platform stops the server once it detects that the process exited. The process may already have restarted briefly before the stop. |
| Restart at most | 3 times | From 1 to 20. |
A server stopped for crashing is not woken by players joining. Its owner or an Admin fixes the cause and selects Wake at the top of the server page, which also resets the crash count. Exiting the process with an in-game command such as /stop also counts as an unexpected exit; to stop a server, use Stop at the top of the server page.
The Activity list on the server's Monitoring tab shows the latest starts, stops, crashes, failed starts and host restarts.
Resuming after a host restart
After a power loss or a restart of the host, the platform services start with the host, and servers that were running before resume by default. With Resume after a host restart turned off on the Settings tab, the server stays stopped once the host is back and its console page shows why; select Wake when it is needed, and players joining wake it under its autostart policy. A process exit caused by a host restart does not count as a crash.
TIP
Whether the host powers on by itself after a power loss depends on the hardware or virtualization platform, for example the BIOS setting Restore on AC Power Loss. Unattended deployments need that setting turned on at the host.
The login space and the lobby always restart and resume after a host restart, whatever these settings say; the platform alerts the platform owners about their problems. See Monitoring and alerts.
Email notifications
When the platform has a mail relay, server events are mailed to the verified email of the server's owner; for an unclaimed server, to the platform owners. Admins and server owners turn on the notifications they want under Email notifications on the Settings tab:
| Notification | Default | Events |
|---|---|---|
| Problems | On | Crashes and the outcome of the restart, a stop after crashing, failed starts, and resuming or staying stopped after a host restart. |
| Server started | Off | The server finished starting. |
| Server stopped | Off | The server finished stopping, including an idle stop. |
After a crash, the platform waits up to 3 minutes and sends the outcome of the restart and any events in between in one mail; a stop after crashing is sent at once. Events not sent within an hour are dropped. These mails count toward the relay's hourly limit. The login space and the lobby send no such mail.
Autostart policy
The autostart policy decides who may wake a stopped or hibernated server. The server owner and Admins can always start it.
| Policy | Who can wake the server |
|---|---|
| Owner only — wake from the panel | Only the server owner and Admins. |
| Public — any player join wakes it | Any player who joins. |
| Wake list — previous players with wake permission can start the server | The server owner, Admins, and players on the wake list who hold wake permission. |
Server owners and Admins choose the policy under Autostart policy on the Settings tab. With the Wake list policy, a player is added to the list on their first join. The server owner or an Admin selects Revoke under Player management → Wake list to take the permission away and Allow again to restore it. Rejoining does not restore a revoked permission. Releasing, claiming or assigning the server clears the wake list and resets the autostart policy to Owner only; account migration clears the wake list and preserves the policy.
A player without permission to wake the server is told in game that they cannot start it. When the platform is at its running server limit, the player is told to try again later.
Console
The server's console page shows the live log and offers:
- The stream state (Live, Reconnecting, Stream ended) with Reconnect, Clear and Jump to latest. RCON connection lifecycle logs are hidden by default; show them with Show RCON connection logs. The saved log shown after the server stops works the same way. The platform keeps one RCON connection per server and reuses it, so connection entries appear only after the server starts or a connection drops.
- While the server starts or runs, its owner and Admins can use Server command to send console commands, for example
listorop <player>. - Links to player management, LuckPerms permissions, backups and restore, server files and scheduled tasks.
Sending commands
Enter a single-line command and press Enter or select Send command. The command is written to the server's standard input, as if typed on the server's own terminal: the log first shows the command echoed with a leading >, and its output follows in the same log stream without any RCON session entries. When standard input is unavailable and the command has not been delivered, the platform runs it over RCON instead, and the reply also appears in the log. Before the log starts, the panel shows Command sent. Check the console log for the result. above the input.
If the connection fails after sending begins, the panel reports an uncertain outcome and retains the input. Check the log before deciding whether to send it again; commands are never retried automatically. Input is unavailable before startup or during shutdown; send requests are refused during maintenance such as backups, restores and migrations.
Servers already running before an upgrade enable standard input at their next start or restart. The upgrade does not restart the game process for this change. Older instances can continue to use RCON when it is available.
Stopped server logs
Once the server has fully stopped, its owner and Admins can view the last 200 lines of logs/latest.log on the world volume and load earlier content. Each page displays at most 256 KiB; longer individual lines appear in segments. History is read-only. The live log stream connects automatically when the server starts.
An absent world volume or log file displays no history. If the file is replaced or modified, refresh to read it again and avoid mixing different versions. Existing file access restrictions apply during backups, restores and migrations.
Resource usage
Server owners and Admins can view current CPU, memory and disk usage and the last hour of history on the server's Monitoring tab. The wide server list also displays current usage. CPU percentages use the server's CPU limit: 500 millicores against a limit of two cores is 25%. Without a CPU limit, usage is shown in millicores. Memory shows the working set against the container memory limit, including memory outside the JVM heap.
The platform samples every 15 seconds and retains at most 240 points per server. Stopped servers and unavailable metrics show no current value; existing history remains for up to one hour. Missing metrics are never shown as zero usage. History is held in the API instance's memory and starts accumulating again after an instance restart or change. Sampling does not include disk usage or network traffic.
Player management
The Player management page works over a live connection to the server, so it is available only while the server runs, to the server owner and Admins.
- Whitelist: enter a player name (1–16 letters, digits or underscores) and select Add, or select Remove. While the whitelist is turned on in the game (console command
whitelist on), only listed players can join. - Online players: see who is online and select Kick to disconnect a player.
- Bans: enter a full player name and select Ban; the player is kicked and blocked from rejoining. Select Pardon in the ban list to lift a ban.
- Wake list: see Autostart policy.
When the panel cannot parse the server's reply, View raw server reply shows the raw output.
LuckPerms permissions
When the server runs the LuckPerms plugin, the LuckPerms Permissions page manages a player's groups and permission nodes:
- Select an online player on the left, or look up a player by name at the top.
- Select Add Parent Group or Remove Parent Group to manage parent groups.
- Enter a permission node, a value (grant or deny) and an optional world context, then select Add Permission or Remove Permission. Removal deletes the node in that world context whatever its value.
The page offers presets and keeps a Recent Actions History, from which Revert undoes a change. Changes by player name are reliable only for players who have joined since LuckPerms was installed.
Installed plugins and mods
Server owners and Admins can open the server's Plugins and mods tab to inspect .jar and .jar.disabled files in plugins/ and mods/. The page is available while running or fully stopped; wait for startup, shutdown or maintenance to finish before making changes.
The list shows the filename, size, enabled state, and the name, version, authors and loaders declared in readable manifests. It supports Bukkit, Paper, Fabric, Quilt, Forge, NeoForge and Velocity manifests, retaining multiple mod or loader declarations in one JAR. Files with unreadable metadata remain listed and can be disabled or deleted. Consult the server console for actual loading status.
- Enable and disable rename between
.jarand.jar.disabled. An existing destination is never replaced; resolve conflicting names through the file manager first. - Delete permanently removes the selected JAR after confirmation and keeps configuration files.
- Open configuration opens an existing configuration directory in the file manager. The action is absent when no matching directory was found.
- Upload JAR sends files to the plugin or mod directory through the existing upload queue and multipart transfer. The list refreshes after the batch finishes.
Restart the server to apply uploads, enables, disables and deletions made while running. Changes to world directories still require the additional acknowledgement, including when level-name is set to plugins or mods. Missing directories produce an empty list. Scans inspect at most 2000 entries per directory, return at most 500 files, and bound manifest and response sizes; a truncated list directs you to the file manager for remaining files.
Server files
The Server Files page browses and repairs files in the world volume, for the server owner and Admins. It is available while the server is running or fully stopped; wait for startup or shutdown to finish during a transition. A new server that has not started yet can use it as well, for example to put a modpack, a world or a custom jar in place before its first start.
During live access, the page warns that world changes may be overwritten by the server and that plugin and configuration changes usually take effect after a restart. Select the additional risk acknowledgement before modifying world, world_nether, world_the_end or the world directory named by level-name in server.properties. Extracting an archive while running also requires this acknowledgement. It applies only to the current page and resets after reopening the page or a change in running state.
You can:
- Browse folders, create files and folders, rename and delete. Select several entries to delete or download them together. Deletion is permanent; back up anything you need first. Each row's More menu also supports right-click and a long press on mobile.
- Copy or cut files and folders within the same server, then open the destination folder and paste. Review the conflict list before confirming replacement of existing files. Rename entries first when a file conflicts with a folder.
- Search file names in the current folder and its descendants. Search returns at most 500 results, descends at most 16 levels, and skips the
region,entitiesandpoichunk-data folders. Narrow the folder or query when results are truncated. - Edit text files up to 5 MiB online, preserving CRLF line endings. Download binary or oversized files, edit them elsewhere and upload them again. If a file changed after you opened it, the editor offers Load latest or Overwrite anyway.
- Upload files or folders using the upload buttons or by dragging them into the list. Folder uploads preserve relative paths, create directories and upload each file in turn. Files up to 64 MiB go up in one request; larger files are sent automatically in 32 MiB parts and resume after an interruption. For a file that already exists, choose Replace or Skip. An account can have at most 4 large uploads in progress, and an upload session idle for 6 hours is dropped.
- Download a public URL to a chosen server path. Only HTTP and HTTPS without a username or password are accepted. Private, loopback and link-local destinations, including redirects to them, are refused. The file is first fetched onto the platform's staging disk, then written to the server; its size is bounded by available staging space.
- Pack selected entries into a
.zipon the server, or extract.zip,.tar,.tar.gzand.tgzarchives into the current folder. Before extracting, Felis checks containment, links, file/folder conflicts and free space. When existing files would be replaced, the panel lists them for confirmation. Zip archives with file names in GBK encoding extract correctly. - Download a file, a folder as a
.zip, or selected entries together as a.zip. Each account may run 2 downloads at a time and 30 per hour; the platform runs at most 4 at a time. PNG, JPG, GIF and WebP images up to 5 MiB can be previewed through the same download channel and count toward these limits.
Upload landing, copies, moves, packing and extraction run in the background and continue after you close the page; Background operations shows their progress, and each may run for up to 2 hours. URL downloads wait for remote content to reach staging before starting the background landing operation. Backups, restores and migrations block file operations. Writes and downloads on a stopped server hold a maintenance lock that prevents startup until they finish. Live file operations take no admission lock and do not prevent shutdown.
A running server may recreate source or destination files during an operation. If rollback cannot finish, Background operations lists the retained directories. Review the contents of move-recovery-* or file-recovery-* before moving or deleting them. Concurrently created content is not overwritten automatically; protected backups require recovery by the host administrator.
Some files are protected:
config/paper-global.ymlholds the proxy forwarding secret that every server shares. Its source and downloads remain protected, and it is left out of folder downloads. Common settings can be edited through the restricted form below.- Downloading, packing or copying
server.propertiesredacts its credentials. server.properties,configandconfig/paper-global.ymlare managed by Felis and cannot be renamed or moved.
Configuration forms
Open server.properties in the root folder through file management to switch between Form and Text. The form groups common settings under game rules, players, world, network and performance, with boolean, integer, choice and text controls, field descriptions and validation errors. Use text mode for other settings.
Both modes share one draft, so switching preserves unsaved changes. The form updates only properties you change, retaining comments, order, unknown keys and line endings; opening the form does not add missing properties. Legacy properties such as PVP, Nether access and mob spawning appear only when their keys exist in the file, with descriptions of applicable versions. Platform-managed ports, authentication and RCON fields are read-only in the form. Forwarding and player limits for the login space are also managed by the platform.
The editor hides rcon.password and forwarding-secrets. Saving retains the current on-disk values for unchanged placeholders; these credentials do not need to be entered.
bukkit.yml, spigot.yml, config/paper-world-defaults.yml and purpur.yml also offer form and text modes. Each form lists existing common settings with compatible types, without inserting defaults or creating world overrides. Edits preserve comments, order and unknown data, although formatting may be normalized; unchanged files retain their source. Use text mode for complex structures such as anchors, aliases or multiple documents.
config/paper-global.yml uses a restricted form for existing common fields with compatible types, without inserting defaults; proxy forwarding and authentication remain managed by the platform. The form receives only editable fields and saves only those you change, retaining other settings and comments. After a conflict, Overwrite anyway reads the latest version before applying those changes.
If another process changed the file before saving, choose Load latest or Overwrite anyway. Configuration saved while running usually requires a restart to take effect; the existing file access and world-directory acknowledgement rules apply.
Importing a world
- Back up and stop the server.
- Compress the world folder into a
.zipand upload it to the root folder under Server Files. - Extract the archive and confirm that the world folder directly contains
level.dat. - In
server.properties, setlevel-nameto the name of the world folder. - Start the server.
For refused file operations, see File manager errors.
Scheduled tasks
The Scheduled tasks page runs a sequence of server operations on a schedule, for the server owner and Admins, with up to 20 tasks per server. It offers templates such as Restart every night at 04:00, Back up every night at 05:00 and Announce in chat every 30 minutes; select New task for a custom one.
A task consists of one to ten steps that run in order. Steps can be added, removed and reordered. The editor also offers three templates that fill in steps at once: Warn, then restart (console command → delay of 60 seconds → restart), Stop, back up, start (stop → backup → start) and Save, then back up (save-all → backup).
| Step | Server running | Server stopped |
|---|---|---|
| Console command | Runs the command over RCON; the leading / is optional. The command is one line of up to 1024 bytes. | The whole run is recorded as Skipped |
| Delay | Waits 1 to 3600 seconds | Waits 1 to 3600 seconds |
| Start | Goal already holds | Starts the server, subject to the running server limit |
| Stop | Stops the server; players are disconnected | Goal already holds |
| Restart | Stops and starts the server; players are disconnected | Goal already holds |
| Force stop | Ends the server process at once, without saving; players are disconnected | Goal already holds |
| Backup | Takes a live backup where possible; otherwise stops the server, backs it up and starts it again | Backs it up; the server stays stopped |
A task must contain at least one step that is not a delay, and at most one backup step.
Task settings:
- Execution mode: Once a day runs at a set time. Repeat runs at an interval counted from 00:00 each day: 1, 5, 10, 15 or 30 minutes, or 1, 2, 3, 4, 6, 8 or 12 hours. A task with a start, stop, restart, force stop or backup step repeats at most once an hour. One time only runs once at the chosen date and time, interpreted in the task's time zone. When the task is enabled, that time must be in the future, and the task switches itself off after it runs.
- Execution days: the weekdays to run on, with shortcuts for every day, weekdays and weekends. A one-time task does not use this setting.
- Time zone: an IANA time zone name (for example
Asia/Shanghai); defaults to the browser's. - Disable after N runs: from 1 to 10000. Leave it empty or enter
0for no limit. A run started by hand counts toward the limit; a missed run does not. When the limit is reached, the task switches itself off. When a task that is off is enabled again, its run count returns to zero. While the task is enabled, the limit must be greater than the number of runs already made. - Warn players: when a task has a stop, restart, force stop or backup step, it can post a Chinese and English warning in game chat 1, 5, 10, 15 or 30 minutes before the task starts, while the server runs. The warning goes out before the first step, and its wording follows the first stop, restart, force stop or backup step.
- Enabled: switch it off to keep the task without running it.
Each row in the task list shows the task's steps (for example Console command → Delay 60 seconds → Restart), the run time of a one-time task, and, for a task with a run limit, the runs made so far (for example Executed 3 of 10 times). While a task runs, its row shows the current progress, for example Step 2 of 3: Waiting for the server to stop…
How tasks run:
- felis-api checks every 15 seconds, so a task starts within about 15 seconds of its time. A run more than 10 minutes late (for example because felis-api was not running) is recorded as Missed, and the next one runs as usual.
- Steps run in order. When a step fails, the run is recorded as Failed and the remaining steps do not run. If the run stopped a server that was running, the server is started again after the failure. A command whose outcome is unknown is not sent again.
- After a start or restart that has steps behind it, the run waits up to 15 minutes for the server to be ready. After a stop or force stop that has steps behind it, the run waits for the server to stop completely.
- A start, stop, restart or force stop whose goal already holds (for example, a stop or restart on a server that is already stopped) is recorded as Skipped when the task consists of that step alone, and is passed over in a longer task. A command step on a server that is not running records the whole run as Skipped.
- The result shows as Succeeded, Skipped, Failed or Missed, with details.
- Run now also works on a task that is off. It skips the players' warning, disconnects online players right away, and keeps the next scheduled run. While a run is in progress, the task cannot be edited, deleted or run again.
- When the server changes owner, any run in progress ends at once as Skipped and its tasks are switched off; the new owner reviews and saves them to turn them back on.
- A time the clock skips at a daylight-saving change runs an hour early; a time it repeats runs once. A time zone the host does not know runs in UTC.
- Deleting a server deletes its tasks.
For skipped or failed runs, see Scheduled task errors.
Backups and restore
The Backups & restore page lists the server's world backups, for the server owner and Admins. A backup holds the server's entire data volume: worlds, configuration, plugins and mods.
Felis creates these backups automatically:
- Scheduled backups: every server played that day gets a daily restore point, taken live while it runs. By default 7 are kept per server, for 90 days.
- Reclamation archives: before a server is given up, deleted or reclaimed for inactivity, its world is archived.
Manual actions:
- Back up now: a stopped server is archived as it is. A running server is backed up live: saving pauses for a few seconds while the world is copied, and players stay connected; a live backup needs the server's RCON. While the server is starting or stopping, wait until it runs or has stopped. By default each server keeps its 5 latest manual backups for 30 days, and two manual backups must be at least 10 minutes apart; Admins are exempt from the interval.
- Restore this backup: stops the server (with a 30-second notice to online players) and overwrites the current world with the chosen backup. Tick Take a safety snapshot of the current world first (recommended): the restore starts only once the snapshot succeeds, and nothing is restored if it fails. To undo a restore, restore the Pre-restore snapshot; the last 3 pre-restore snapshots are kept.
- Download backup: downloads a
.tar.gz, checked during transfer against the sha256 recorded at backup time. - Export world: stop the server first. The current world is packaged as a
.tar.gzand downloaded automatically; keep the page open, as the server cannot start until the download ends. - Delete backup: the backup leaves the list at once and can no longer be restored. Its archive is removed at the next daily cleanup, and its off-site copy at the next sync.
A backup that fails read-back verification is marked Corrupt and cannot be restored intact. Recent operations lists backup, restore and export tasks with the reason for any failure. When the backup store is full, manual backups by server owners are refused.
WARNING
Without a safety snapshot, a restore overwrites the current world directly; existing data that has no other backup cannot be recovered.
For retention, live backups and restore points in detail, see World backups, Live backups and Daily restore points. For failed backups, see Backup failures.
Forwarding mode
The forwarding mode decides how the game proxy passes the player's identity to a server. The Owner sets it under Forwarding mode in the Network section of the server's Settings tab; the login space and the lobby do not offer it.
| Mode | Use |
|---|---|
| Modern forwarding | The default. Servers on 1.13 or newer use modern forwarding, signed with the proxy's secret. |
| Legacy forwarding (BungeeCord) | Older backends behind ViaVersion. |
- On the server's Settings tab, select Change forwarding in the Forwarding mode row.
- Tick or clear Use legacy forwarding and select Save.
- Verify your identity. The proxy applies the change at its next server-list refresh, within 15 seconds.
WARNING
A legacy-forwarded server trusts any identity that reaches its port. Felis lets only the proxy reach game servers; do not expose the server's port in any other way.
The change runs through the host node operations service and is recorded in the audit log. The servers named in FELIS_LEGACY_FORWARDING_SERVERS at install time (default legacy18) always use legacy forwarding.
Giving up and deleting servers
A server owner can give up a server; Admins can also delete it:
| Action | Result |
|---|---|
| Give up | The world is archived and its storage deleted; the server is released for other accounts to claim. |
| Delete | The world is archived, and its storage and the server record are deleted; the name and subdomain can be reused. |
- On the server's Settings tab, select Give up or Delete in the Danger zone.
- Type the server's name and confirm.
The server then stops and is processed at the next daily reclaim run; until then it stays stopped and cannot be started. Archives are kept for 90 days by default. Before the run, select Cancel request on the console to withdraw it; a deletion requested by an Admin can be cancelled only by an Admin. A server that was given up counts toward the account's quota until it has been processed.
With world cleanup enabled, worlds nobody has played for more than 15 days are backed up and then deleted. See World cleanup.
Quotas
The Owner can set a quota for each account, limiting how many servers it owns and their total CPU, memory and storage; with Self-service servers on, accounts without their own quota use the default quota. Quotas are checked:
- When a server is self-created, at the selected size.
- When a server is claimed, at the server's actual size.
- When the CPU or memory of an owned server is raised.
- When an account receives servers through an account migration.
An action that would exceed the quota is refused. Lowering a quota restricts later actions only and leaves existing servers alone. To set quotas, see Quotas.