Struct EpCommandFailed 

pub struct EpCommandFailed(pub Error);

A worker command that was received and then FAILED TO EXECUTE.

🔴 Why this distinction is load-bearing. The EP worker loop used to break on any error, so a per-request fault — a prefill chunk the model legitimately refuses — killed the worker, which then exited with status 0 while the head stayed up. The head’s very next request issued a collective against a peer that no longer existed and spun in NCCL forever at 100 % CPU, with /v1/models, /health and /health/live all still answering 200. Measured 2026-08-30: rank 1 logged this exact refusal and stopped 4 s later; rank 0 accepted a 13-token request 10 minutes on and never produced a single further log line. ANOMALIES A60 (the wedge) and A62 (the refusal that triggered it).

The head raises the SAME error for the SAME command and turns it into an HTTP 500, so the two ranks disagreeing about whether it is fatal is the defect. A receive failure stays fatal: the link is gone, and the next iteration’s receive would fail again anyway.

Tuple Fields

0: Error

Trait Implementations

impl Debug for EpCommandFailed

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Display for EpCommandFailed

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Error for EpCommandFailed

fn source(&self) -> Option<&(dyn Error + 'static)>

Returns the lower-level source of this error, if any.

fn description(&self) -> &str

👎Deprecated since 1.42.0: use the Display impl or to_string()

fn cause(&self) -> Option<&dyn Error>

👎Deprecated since 1.33.0: replaced by Error::source, which can support downcasting

fn provide<'a>(&'a self, request: &mut Request<'a>)

🔬This is a nightly-only experimental API. (error_generic_member_access)

Provides type-based access to context intended for error reports.