mempalace.auto_wake

Source: mempalace/auto_wake.py

Wake-on-demand for a sleeping palace host.

The palace daemon often runs on a host that suspends to save power (Wake-on-LAN-armed), so "connection refused / no route" is a routine state, not a fault. When a CLI request hits a connection-level failure, this module can run a user-configured wake command (a WoL sender, an IPMI call, anything), wait for the daemon's /health endpoint to come back, and retry the original request once. A proxy in the path (HTTP_PROXY, reverse proxy) reports the same sleeping host as its own 502/504 response, so those statuses count as wake-eligible too.

Strictly opt-in: enabled only by an auto_wake entry in ~/.mempalace/config.json (see :meth:MempalaceConfig.auto_wake); PALACE_AUTO_WAKE=0 force-disables without editing config. The wake command runs through the shell with the same trust level as the user's own shell startup files — it comes from their config file, never from palace content.

Scope: interactive CLI calls only. Hooks deliberately stay out — they have a latency budget and their failed mines are already journaled and replayed by :mod:mempalace.pending_queue.

Functions

attempt_wake

def attempt_wake(daemon_url: str, settings: dict) -> bool

Run the wake command, then poll /health until the deadline.

Returns True once the daemon answers. At most one attempt per process regardless of outcome.

urlopen_with_wake

def urlopen_with_wake(req, timeout)

urllib.request.urlopen with an optional wake-and-retry.

Drop-in replacement for the CLI's daemon calls: on a wake-eligible failure with auto_wake configured — a connection-level error, or a proxy's 502/504 standing in for the sleeping host — wake the host, wait for /health, and retry the request once. Everything else — other HTTP errors, disabled config, failed wake — re-raises the original error unchanged so existing DaemonError handling is untouched.