Skip to content

Tale v0.5.106

Choose a tag to compare

@github-actions github-actions released this 14 Nov 12:48
b68d753

Highlights

  • Run Tale on Incus (experimental). A workspace or a managed deployment can now run as Incus instances on one Linux host instead of Docker Compose containers, with no Docker daemon on that host. The services, images and settings stay the same: the tale CLI compiles the same Compose definitions into instances in one restricted Incus project, and the sandbox sessions run in a project of their own. Choose it in the reviewed declaration, never per command: runtime.engine: "incus" with a runtime.incus block in a workspace's tale.json (tale init --runtime incus writes it for a new workspace) or in a managed deployment specification. tale incus contract prints what the node must hold as JSON for your own tooling, tale incus bootstrap makes it on a host you administer alone, and tale incus check reads the node back and refuses what differs. The host needs Incus 7.2 or later (not a cluster member), Linux 5.12 or later with idmapped mounts, a zfs, btrfs or lvm storage pool and an x86_64 CPU; the page lists every prerequisite (#5423, #5424, #5427, #5430).
    • A workspace on Incus deploys with tale deploy, flips colours blue-green through a traffic switch that admits each new replica without a reload, and resumes an interrupted flip on the next deploy. --plan-only and --plan let you review each scope's plan before it is applied. Backups are native Incus snapshot sets taken under a freeze of well under a second, and a restore copies into a fresh volume beside the old one, so the snapshots newer than the restored one stay available (tale restore --resume and --abort finish or undo an interrupted restore).
    • Move a running Docker workspace with tale runtime migrate --to incus, at the same version and with the same .env. Each step is journaled and the command resumes after a stop. The Docker stack stays stopped so that tale runtime migrate --to docker can roll back, until --finish removes it. Users see downtime while the data is archived and restored, so plan a maintenance window.
    • A managed deployment on Incus gets its first deployment through tale deploy prepare and tale deploy --bundle, as on Docker.
    • Limits: the runtime is experimental and Docker Compose stays the default and supported runtime. Upgrading a managed deployment that already runs on Incus is still refused, because its recovery snapshot cannot be taken there yet. tale dev and connected sandbox devices run on Docker only, an Incus cluster is refused, and the stack commands on Incus refuse on Windows. The current limits list the rest.
  • Stop and start a whole workspace in order, on Docker too. The new tale stop stops the edge and both colours first, then the sandbox services, then the database, the object store and the resolvers, each within its own grace period. tale start starts them the other way round, each step healthy before the next. Run tale stop before a shared host shuts down, so the workers can finish their drain. A workspace's tale backup export and tale backup import also carry one snapshot as a single file to another host or runtime (Backups and restore).
  • A Docker deployment no longer waits five minutes for an idle sandbox. Every tale deploy that recreated the sandbox spawner on Docker waited out the whole five-minute drain budget, even when no session ran: the CLI read a field the spawner no longer sends. It now reads the spawner's session count and goes on as soon as no session runs. An answer it cannot read still counts as unknown, with one warning, and the deploy waits as before (#5429).
  • A restarted database waits for its own start before it reports ready. After a restart that keeps the container's filesystem, such as docker restart, the database reported ready from the previous start's marker as soon as Postgres answered, before this start's init scripts and migrations had run, so its health check and whatever waits on it could go ahead too early. The marker is now cleared at each start and set again only when that start's init has finished (#5421).
  • Security dependency updates. The platform and the session manager move to hono 4.13.16, which fixes several advisories. Tale is likely affected by one of them, GHSA-h7mg-cq63-vhpq (CVE-2026-126358, medium): the default router could skip wildcard middleware on some request paths, and Tale's backend guards routes with such middleware, for example the organization access check. This was not exploited in a test. The others concern hono features Tale does not use. fast-xml-parser moves to 5.12.0 (GHSA-474x-hrwr-fgx4, low, only with the stopNodes option, which Tale does not set) (#5426, #5419). @hono/node-server's advisory GHSA-rmxm-3fg6-px4f concerns serveStatic, which Tale does not use; its fix is in the 2.x line, which a later release adopts.

Upgrade notes

  • Platform: run the 0.5.106 images and upgrade the tale CLI with them, which carries the sandbox drain fix, tale stop, tale start and the Incus runtime. This release adds no database migration and does not change the API contract version. No configuration change is needed.
  • Docker deployments are unchanged. A workspace whose tale.json has no runtime, or "engine": "docker", runs on Docker Compose as before. The images gained what Incus needs (an init process with /init, an admission probe at GET /api/admission on the platform and GET /admission on the backend, and settings that only an Incus stack sets), and none of it changes how they run under Compose: their entrypoints and /api/health are the same. A runtime.engine value the CLI does not know is now refused instead of being read as Docker, and so is an incus block without "engine": "incus".
  • To try Incus, follow Run Tale on Incus. Set runtime.engine to incus by hand only in a workspace that has never been deployed. A workspace that already runs on Docker declares the target in runtime.next and moves with tale runtime migrate --to incus, so that its data comes along. Run the migration with the CLI of the version the workspace runs, and keep an off-host backup first. TALE_RUNTIME only makes a command refuse when the declaration names the other engine; it never selects one.

API contract changes

None in this range. The contract stays at 3.53.0: 163 operations.

What's Changed

Full Changelog: v0.5.105...v0.5.106