> ## Documentation Index
> Fetch the complete documentation index at: https://docs.lagerdata.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Version 0.34.4

> August 5, 2026

## <u>Features</u>

* **`lager diagnose` now covers USB hub nets.** They used to fall through to a
  generic "check the role command yourself", which is no help when the question
  is why the hub will not answer. The new section reports what host-side tools
  structurally cannot: the vendor SDK's own view of the hub. A hub can be
  enumerated by the kernel, held by no other process, and still invisible to
  hub discovery — and in that state every other signal looks healthy.

  It also reports the device's USB device number against the rest of the bench.
  The kernel assigns those in order, so a device far above its neighbours has
  re-enumerated many times since they did. On the bench that prompted this
  work, the failing hub sat at 93 while every other instrument was in the 60s;
  that one number was the most useful fact in the investigation and nothing
  surfaced it.

## <u>Bug Fixes</u>

* **The box no longer restarts its HTTP service for a USB hub it cannot reach.**
  A self-restart repairs exactly one thing: a USB handle the process orphaned
  across a re-enumeration, which only a fresh process can reopen. The Acroname
  driver keeps no such handle — it opens, operates and disconnects on every
  call — so there was never anything on that path for a restart to repair. It
  fired anyway, because the check asked only whether the device was still
  enumerated, and the kernel keeps a device node for hardware that has stopped
  answering on the wire.

  Observed on a two-hub bench: a hub that would not open triggered the restart
  twice, and each respawned process failed identically seconds later. Nothing
  was fixed, and every other in-flight operation — UART sessions, running
  scripts, hardware calls — was dropped to do it. Drivers now declare whether
  they hold a USB context between operations, and those that do not are
  skipped. The pyvisa and HID paths, where a session really does persist, are
  unchanged.

* **`lager diagnose` reported the wrong instrument on a bench with two of the
  same model.** The device lookup matched on vendor and product id and returned
  the first hit. It now prefers an exact serial match, falling back to
  vendor/product so a device with an unreadable serial is not lost.

* **`lager diagnose`'s kernel-log section has never worked.** It shelled out to
  a command the box image does not ship, so the field has read
  "unavailable" on every box since it shipped. It now reads the kernel log
  directly and, where the container is not permitted to, says so and points at
  where the history actually lives.

## <u>Improvements</u>

* **A USB hub that will not open now tells you what to do about it.** The box
  already worked out which of three faults it was looking at — nothing from
  this vendor on the bus, the hub's serial present but not answering, or other
  devices present and none of them the one addressed — then flattened all three
  into a single line with vendor return codes appended. The terminal showed a
  wall of codes and no sentence saying whether to check a cable, a power
  switch, or the net's address.

  Unknown entries in `lager nets state` now carry a machine-readable
  `reason_code` alongside the human reason, and the footnote adds one remedy
  per affected group — red when the fault is hardware, yellow when the bench is
  more likely in a normal state. `--json` output carries
  `live_state_reason_code`. The box always sends a complete, self-sufficient
  human reason, so an older CLI, or a newer one seeing a code it does not
  recognise, renders exactly as before.

  A hub the kernel has enumerated but that will not answer is now logged as an
  error rather than a warning. That case is always hardware and always worth
  acting on; a hub that is simply absent is a normal bench state.

* **An Acroname hub open is retried once when the bus says the hub is there.** A
  hub caught mid-re-enumeration is on the bus a beat before discovery will
  return it, so an operation landing in that window failed outright. The YKUSH
  driver has always retried once for this reason; this one did not. The retry is
  gated on the bus check, so a hub that is genuinely absent still costs exactly
  one attempt, and it is suppressed on the polling path so the whole-bench state
  sweep's timing is unchanged.

## <u>Installation</u>

To install this version:

```bash theme={null}
pip install lager-cli==0.34.4
```

To upgrade from a previous version:

```bash theme={null}
pip install --upgrade lager-cli
```

## Resources

[View Release on PyPI](https://pypi.org/project/lager-cli/0.34.4/)