Skip to content

Tale v0.5.107

Latest

Choose a tag to compare

@github-actions github-actions released this 16 Nov 03:49
5b5aebd

Highlights

  • Sandbox sessions on Incus can spread over several hosts (experimental). The Incus sandbox backend reached exactly one host, so every session of a stack ran on that one node. The new SANDBOX_INCUS_ENDPOINTS takes a list of hosts, each with its own restricted client certificate, key and pinned server certificate, and optionally its own project, profile, network, storage pool, session cap and egress proxy. A new session goes to the host with the most room left by its session cap, its project's memory and disk limits and its pool's free space. A host marked drained takes no new session while its sessions keep running there. A host that does not answer is skipped for new sessions and asked again after 30 seconds, and the spawner reports healthy while at least one host answers. A session always resumes on the host that holds its workspace: the spawner records each session's host on a persistent volume, and while that host does not answer the session answers 503 session_unavailable rather than starting elsewhere with an empty workspace. A create that no host can take answers 429 session_quota when every host is full or drained, and 503 session_unavailable when none answers. On the Incus runtime, tale declares the hosts in runtime.incus.sandboxHosts and places each host's certificate files in sandbox-incus/<name>/ (Declare the container runtime, Declare the Incus runtime). Without the list nothing changes (#5483).
  • Configuration written right after an organization is created is kept. Creating an organization queued a job that emptied the organization's configuration folder and then seeded it. The job runs on a worker after the create has returned, so a provider, agent or governance file that a client or an automation wrote for the new organization in that window was deleted. A leftover folder of an earlier organization with the same slug is now removed during the create, before the organization exists, and the job only adds the catalog files and skill bundles the organization does not have yet. It never overwrites or removes a file (#5438).
  • tale logs --follow no longer fails when an Incus instance has just stopped. Right after an instance stops, Incus can briefly answer the read of its console with 500 "connection refused". The follow then failed instead of showing the console's last lines. It now asks again a poll apart, at most ten times, only for that answer; any other error still fails at once (#5488).
  • Security dependency update: @hono/node-server 2.1.8. The platform moves from 1.19.17 to 2.1.8. This fixes GHSA-jpvv-r467-5p4w (CVE-2026-126210, moderate), which affected Tale: a request path with a raw backslash, such as GET /public\..\ping, was resolved as /ping and reached that route. Such a request is now refused with 400 before any route sees it. It also fixes GHSA-rmxm-3fg6-px4f (CVE-2026-118663, moderate), which concerns serveStatic, which Tale does not use; v0.5.106 said a later release would adopt its fix in the 2.x line (#5434).

Upgrade notes

  • Platform: run the 0.5.107 images and upgrade the tale CLI with them. This release adds no database migration and does not change the API contract version. No configuration change is needed. A scaffold job that the previous release queued during a rolling deploy still runs as before.
  • @hono/node-server 2.x needs Node 20 or later; the platform image runs Node 22. Nothing changes in how the platform serves requests, except that a request path with a raw backslash is refused with 400.
  • Several sandbox hosts are opt-in. A sandbox spawner without SANDBOX_INCUS_ENDPOINTS, and a stack without runtime.incus.sandboxHosts, run exactly as before. With the list, SANDBOX_INCUS_ENDPOINT and its three certificate variables must be unset, and SANDBOX_INCUS_STATE_DIR (default /var/lib/tale-sandbox/incus) must be on a persistent volume; the tale CLI adds a sandbox-state volume for it. Every host must give a session the same shape, and each host needs its own routable sandbox network and egress proxy. A stack that moves from one host to the list keeps its sessions only through an entry with the one host's endpoint and project, and a deploy without such an entry is refused. Going back from the list to one host is refused while the record volume exists; tale deploy --clear-sandbox-placement deletes it on purpose.

API contract changes

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

What's Changed

Full Changelog: v0.5.106...v0.5.107