Recover a wedged iDRAC

The iDRAC is a computer too, and sometimes it wedges while the host underneath runs fine. The cardinal rule: never reach for host power actions when only the management plane is sick. Every rung below leaves the host OS untouched.

1. Is it actually wedged?

“Ping r740-lab’s iDRAC.”

If idrac_ping succeeds, it’s not wedged — check health triage for whatever prompted the concern.

2. Auth failures: stop and wait

A 401 with credentials you believe are right often means the iDRAC has IP-blocked you: by default, 3 failed logins in 60 seconds block the source IP for 60 seconds, and the window extends while attempts continue. A retry loop with a wrong password locks you out indefinitely. Stop retrying, wait a minute, fix the credential, try once.

mcidrac itself never retries auth in a loop, and uses token sessions partly to keep failed-Basic-auth counts down — but other tools sharing your source IP count against the same limit.

3. HTTPS dead, SSH alive

The SSH CLI is a separate daemon from the web/Redfish stack and often survives when HTTPS hangs:

ssh root@192.0.2.10
racadm racreset soft

racreset soft reboots only the iDRAC — the host keeps running. Expect 2–3 minutes of management-plane downtime, then everything comes back.

4. Nothing answers, host still up

Reset the iDRAC from the host side: racadm racreset via OpenManage on the OS, or ipmitool mc reset cold locally. Physical option: hold the front-panel i button for ~15 seconds.

5. Last resort

Full power drain: both PSU cords out, hold the power button 15 seconds, reconnect. This is the only rung that involves host downtime, and it’s for the case where the iDRAC hardware itself needs a cold start.

Session exhaustion

A subtler wedge: the iDRAC allows roughly 8 concurrent sessions across everything — web UI tabs, scripts, console viewers, Redfish clients. Exhaust the pool and logins fail even though nothing is broken. mcidrac caps itself at 3 concurrent sessions per host and always logs out, but if you suspect exhaustion, close browser tabs and wait for the 30-minute idle timeout to reap orphans. The diagnose_wedged MCP prompt walks this whole ladder interactively.