Pods
Roadmap
This is what's planned after 1.0, roughly in order. Nothing here is a promise, and the order can change.
Volume backups and restores
Pods exports volumes (see Backups), but doesn't restore them yet. Next:
- lpod.
lpod volume import NAME --stdin [--stop] --jsonreads an archive intopodman volume import, creates the volume first when it doesn't exist, and refusesdeviceand NFS volumes, likevolume export. - Restore. An import starts from an archive Pods made, or one the owner uploads. It replaces the volume's contents, so the UI asks for confirmation, shows which applications it stops, and offers to export the current contents first. It runs in
pods-queuelike an export, streaming the archive over SSH from the backups disk, and sends the same kind of notification.
Open questions:
- Plain bind mounts (such as
Volume=/var/mnt/photos:/photos) aren't Podman volumes, sopodman volume exportdoesn't cover them. They may need a separatetarpath, or stay out of scope. - Databases are safer to back up with their own dump tool, named in
pods.yml(see More of pods.yml). - Another disk, such as an S3-compatible one, for the archives.
- Scheduled backups, with the same retention.
Proxy
Still open for the optional per-host proxy (see Proxy):
- More DNS providers for the wildcard mode, each a
--withline incontainers/runtimes/proxy/Containerfile, a name inProxySettingsData::DnsProvidersand its token variable in lpod. - Routes to sleeping services that aren't
via: main. These still get a 502, such as Mailpit or the RustFS console opened on their own. Caddy could start the app's socket service on a 502 and retry.
More of pods.yml
- Backups.
backupslists the volumes to leave out of "export every volume", and a dump and restore command for a database, run instead of exporting its volume. See Volume backups and restores. - Linking to Git. Pick the source from the repository's
pods.ymlin the "Link to Git" dialog, as the install page does, instead of typing its path. - Timers. Edit a timer's schedule on the application page, kept over updates like a drop-in.
Installing applications from a checkout
Pods installs Quadlets from Git. Development setups, such as a Laravel application's containers/quadlets/development, build their image from a checkout on the host and mount it into the containers. lpod quadlet fetch should read a directory on the host as well as a Git repository, so Pods can install those, and show the build's progress while the first start builds the image.
Later
- Invited users. The owner invites users and gives each a role per environment (viewer, operator or admin), using spatie/laravel-permission. Host routes are already scoped per environment and go through policies, so this doesn't need a rewrite.
- Streaming logs for applications, containers and the journal, instead of polling.
- Live host state for changes made outside Pods, such as a unit that crashed or finished starting. The overview lists the last day of
podman events, but pages still poll for these. A watcher ofpodman eventsper host could broadcast them instead, though it would keep an on-demand Pods awake. - More of the application catalog: brand logos, and more of lerd.sh's services in foxws/quadlets.
- Application metadata on the Applications list: the label, description and icon of the stored
pods.yml, as the application page shows them. - Pagination for long lists. Once a host's lists get long,
DataTableand the catalog grid can show a page at a time withUPagination. This saves rendering, not host calls: lpod still returns the whole list. - Show the commands. Each action can show the
lpod …commands it runs, with a copy button, so they can be run on the host by hand. Multi-step flows (install, update) show the commands in order, and a secret or a password shows as a placeholder on stdin, never its value. - Labels on resources. Tags, a color and a description for applications, containers, pods, images, volumes, networks and secrets, to filter the lists by. These live on the host, so Pods keeps them in a table of its own keyed by environment, kind and name (not a container's ID, which changes when systemd recreates it), tagged with spatie/laravel-tags, and forgets the rows of resources that are gone, like
pods:sync. A description of its own shows instead of the one inpods.yml. - A terminal. A shell in a container (
podman exec -it) or on the host, in the browser with xterm.js. lpod's JSON contract doesn't fit a two-way stream, and Octane workers shouldn't hold one open, so this would be a package of its own: a small bridge between a WebSocket and a PTY over SSH, run as its own Quadlet, that Pods hands a short-lived signed token per session. It needs the invited users' roles first (only an admin of an environment gets a shell), an idle timeout, and a log of who opened which shell. - Compose to Quadlet. A converter, possibly AI-assisted. Pods itself stays Quadlet-only: no Docker and no Podman Compose.
