E · Error ​

The response when a request can't be processed. Instead of a key, a reason (reason) text is sent in its place — the key is never echoed back, so during pipelining you can't tell which request failed from the frame alone. Therefore every E is connection-fatal. The server flushes E when possible and closes the connection; official clients also discard that session. Every acquire for which at least one byte was sent and no definitive reply arrived ends as Indeterminate, never as assumed success or failure.

• These are the complete eight reasons, and every one closes the connection. A raw client may use E for diagnostics, but must not try to resynchronize and reuse that stream.
• Official clients validate keys and time ranges before sending, so bad_* is not a normal API result.
• no_leader/not_active are temporary node states, but E carries no owner/key and cannot be correlated safely. Discard the session; only requests not yet sent may be tried on a new one. Never automatically resend an acquire that may have been sent.
• M, L, no_leader, and not_active only appear in cluster mode — a single server never sends them.
• A full wait queue comes back as B (busy), not as an E — because it echoes the key and owner, so you can tell exactly which request was rejected.

Byte notation: op is an ASCII character, and reason is ASCII text (variable-length, shown as N in the structure). The trailing \n is 0A.

Flow ​

E lets a peer observe diagnostic information, but never permits safe stream reuse. Recoverable capacity rejection uses B (busy), which echoes owner and key.