* feat: add domain virtual responses with a compact editor and bounded edge matching
* fix: keep traffic working during rolling edge script updates
* refactor: centralize edge management transport
* feat: add platform migrations and refresh the Cloud experience
* fix(ci): run MongoDB 8 imports on compatible runners
* docs(web): add Railway and Vercel migration pages
* feat(dashboard): add shared dotenv text editing and top add action
* fix(dashboard): wrap text editor actions on narrow screens
* fix(edge): serve a static site's 404.html for missing paths
* fix(routing): make static SPA fallback explicit and preserve defaults
* test(routing): update static fallback checks and run HTTP coverage
* fix(i18n): complete inbound alias guidance across locales
* test(monitoring): isolate the session in stream fixtures
---------
Co-authored-by: Hydra <abdelhamedmo59@gmail.com>
* fix(deploy): keep loopback-port deploys working without a host channel
On a raw compose install with no host channel, every deploy failed after
the image build with "Cannot read Openship edge routes: neither Docker nor
bare OpenResty inventory is available", even though the deploy log says
host-side edge/routing steps are skipped. The loopback-port preflights read
the strict edge inventory through the demoted host executor, which refuses
every call.
Fixes#1625
When the host executor is one we demoted for having no usable channel, the
preflights now get an inventory they can read — an empty one — so
allocation rests on durable claims alone, exactly like the unscanned
live-port pick. A real conflict still fails loudly at the Docker bind.
Covers both the compose deploy preflight and the single-container
preflight (plus its port-conflict alternative path). The route-apply
safety gate is untouched: refusing a loopback route on an unreadable edge
stays deliberate there.
Test: new case in compose-host-channel-notice drives a loopback-port
compose deploy with the real demoted executor and asserts the edge proxy
handed to the preflight resolves to an empty inventory; it failed with
the reported error before this change and passes after.
* fix(deploy): inventory socket edge routes before allocating ports
* test(deploy): share edge inventory fixtures across allocation paths
* test(routing): resolve inventory from the simulated edge config
---------
Co-authored-by: Hydra <abdelhamedmo59@gmail.com>
* fix(deploy): retain Compose containers until replacement activation
* fix(deploy): wait for exact Docker removal before replacement
* test(deploy): model confirmed removal in legacy daemon fixtures
* fix(deploy): finish Compose health gates before retiring predecessors
* test(docker): model completed removal in GPU adoption fixture
* test(backups): identify the mail size-history fixture correctly
Merge the updated Swarm discovery parent and regenerate MCP references. Match the wizard to the existing self-hosted controller gate so hosted BYOC cannot stage unsupported route passwords. Cover hosted BYOC, self-hosted servers, cloud targets, and cluster targets with the real editor.
Releases recorded before runtimeMode was persisted have no meta.runtimeMode.
The deploy runtime already falls back to the target's default runtime for
them, but the role configuration check required an explicit "docker", so
projects created before v0.9.3 could never configure roles. The dashboard
hid the control for the same reason.
Extract the runtime's fallback into resolveSnapshotRuntimeMode and use it in
both runtime resolvers and the role configuration check. The dashboard now
offers the control unless the project is known to be bare, Cloud or cluster;
the API's answer stays authoritative.
Closes#1307
A disabled service is never deployed, but the preview isolation check still rejected it for host networking, host PID mode, un-namespaced volumes or host bind mounts, so one disabled helper service could block previews for the whole project.
Skip the isolation guards for services with enabled === false, and drop their host bind mounts from the copied preview row instead of throwing. Enabled services (including enabled: undefined) are still refused, and the check runs again on every preview deployment, so enabling a copied row later is still guarded.