Piotr Mlocek addd0132be fix(supervisor): keep polling startup and readiness intact without apply ...

- Close the startup session when the gateway does not enable apply and
  reconnect after the workload starts, as polling supervisors always have.
  Retry transient connection failures while preparing the startup session.
- Mark apply-capable supervisors ready only on SupervisorRuntimeReady, in
  any delivery mode, and resend it after a reconnect without apply.
- Resume polling in a stream-started runtime whenever its current session
  does not apply configuration, such as after a rollback to poll mode.
- Leave restart bookkeeping to runtime readiness instead of admission.
- Keep the owner reconciler local to each replica and suppress unchanged
  snapshots for shadow sessions.

Signed-off-by: Piotr Mlocek <pmlocek@nvidia.com>

2026-10-06 13:55:16 -07:00

..

src

fix(supervisor): keep polling startup and readiness intact without apply

2026-10-06 13:55:16 -07:00

Cargo.toml

fix(deps): upgrade russh to address Dependabot alert 40 (#4116)

2026-10-02 14:12:25 +00:00