tasks/ipmi-node-power-recovery/solution/solution.md

Reference Solution

Step 1 — Identify which node is down

# Check SSH connectivity to each node
ssh node1-mgmt   # → Connection refused or timeout → node1 is down

# Confirm via IPMI
ipmitool -I lan -H 10.134.9.223 -U elabman -P Test1234 chassis power status
# → Chassis Power is off

Step 2 — Power on node1 via IPMI

ipmitool -I lan -H 10.134.9.223 -U elabman -P Test1234 chassis power on
# → Chassis Power Control: Up/On

Step 3 — Wait for boot (~90–120 seconds)

# Poll until SSH is available
until ssh -o ConnectTimeout=5 -o BatchMode=yes node1-mgmt 'echo up' 2>/dev/null; do
    echo "Waiting for node1 to boot..."
    sleep 15
done
echo "node1 is up"

Step 4 — Verify

ssh node1-mgmt "uptime"
# → up N minutes (short uptime confirms fresh boot)

ipmitool -I lan -H 10.134.9.223 -U elabman -P Test1234 chassis power status
# → Chassis Power is on

Root Cause Summary

A physical power event (simulated via IPMI chassis power off) caused node1 to hard-power off. The OS, SSH daemon, and all running services were terminated immediately — no graceful shutdown. Recovery requires out-of-band management via IPMI because the OS is not running and cannot respond to normal network requests. Once chassis power on is issued, the BMC starts the boot sequence independently of the OS.

Why This Is a True L1 Task

Unlike NIC toggling (ip link set down) which is an OS-kernel operation, IPMI chassis power off cuts the ATX 12V power rail from the BMC hardware directly. The CPU, memory, and all OS state are wiped. No OS-level tool can prevent or detect this — only the BMC chip (which has its own power supply) remains active.