[BUG] MTPLX shows degraded connection|loses contact with server|stops in middle of tasks

[GratefulDave]

mtplx doctor --json

I am on Mac M5/128GB using oh-my-pi mostly on port 18791. I am persistently having degraded connection, lost connection with socket.... not sure if it is an MTPLX issue or omp. I am generally not using more than ~50GB of cache.

Exact command

MTPLX Mac App.

Model path or repo id

~/.mtplx/models/

Chip, RAM, macOS version

M5/128GB/

[NorthernMan54]

I had something similar, but it was failing due to this - #194 in the logs

I think the platform needs #202

[youssofal]

2.4.0 adds the telemetry to chase this properly: /health now reports restore_verified, restore_failures and stale_leases_reconciled, and a watchdog logs loudly if any request bookkeeping is held while the engine sits idle. if you can update and paste /health output plus the server log tail from the next time the connection degrades, that narrows whether this is the same wedge class we fixed in #201 or a different one.