The panel is amber but the business is still running. Should we shut it down right away?

In most cases, no. Log in to iDRAC or iLO remotely and read the log to find the component. When a redundant part fails, a power supply, a RAID disk, an ECC memory module or a fan, the business can keep running and the part can be replaced on a schedule. Only a critical temperature, repeated uncorrectable ECC errors or two failed RAID disks need immediate action or a shutdown. A blind shutdown turns planned maintenance into a business incident.

The disk light is blinking amber slowly. Is the data in danger?

A slow flash usually means predictive failure: the disk is still working normally inside the array, but the firmware has decided it is near the end. The data is safe for now; the risk is the next failure. The right move is to hot-swap a replacement in as soon as you can, after confirming a usable backup. Waiting for the disk to die is exactly what turns a routine swap into a recovery job.

I forgot the iDRAC or iLO password. What now?

Dell prints the iDRAC password on the pull-out tag on the front of the chassis, and newer units ship a random default there; you can also press F2 at boot and reset it from the BIOS. On HPE the chassis tag carries the default iLO credentials, or reset it from Smart Setup (F9). Lenovo XCC works the same way: tag first, BIOS reset second. If none of that applies, we deal with it on site in a few minutes.

The server is old and was never set up for remote management. Can we still get fast diagnosis?

Yes. The management hardware is already there on most machines, since iDRAC Express and iLO are standard, it simply has no IP address. Set an address and an account in the BIOS and you are done in ten minutes, and it is worth adding email alerts while you are in there. Servers built after about 2015 support it. Anything older, such as a Dell before 11th generation, has limited function, so run an external SNMP monitoring tool alongside it.

Does a replacement disk have to match the failed one?

Same model and capacity is the ideal, and the replacement must never be smaller than the disk it replaces, or the array cannot use the space. A slightly larger disk usually works, but the extra capacity goes unused unless every disk in the set is at least that size. Confirm a usable backup before you pull anything, and replace one disk at a time so the array can rebuild in between.

Server Warning Lights: Dell, HPE and Lenovo in 5 Minutes

1. Start with the color and the position

The alarm logic is essentially identical across Dell PowerEdge, HPE ProLiant and Lenovo ThinkSystem servers:

Green: normal. Green flashing means a rebuild or another activity is in progress.
Amber: warning or fault. There are two kinds, component-level (one disk, one power supply) and system-level (the overall alarm on the panel, usually with a beep).
Red, or fast-flashing amber: serious fault. Some models flash this way when temperature or power delivery reaches a critical level.

The key point is that an amber light does not mean shut the server down. Most parts of an enterprise server are redundant: dual power supplies, RAID, ECC memory, hot-swap fans. A lot of amber means "you have lost redundancy, fix it soon," not "stop the business." Yanking the power in a panic is how planned maintenance turns into an outage.

2. The five-minute check: your management port is the dashboard

Do not stand in the server room guessing at a light. Log in to the out-of-band management port and read the log.

Brand	Management port	Default access / IP	Where to look
Dell	iDRAC	IP printed on the pull-out tag on the chassis; default user root	Logs → Lifecycle Log
HPE	iLO	Tag on the chassis; default user Administrator	AHS log, and the IML (Integrated Management Log)
Lenovo	XCC / IMM	Tag on the chassis; default user USERID	Event log

If you do not know the management IP: on a Dell, press F2 at boot and look in the iDRAC settings, or connect a monitor and read the error on the POST screen. If the port has never been configured, set it up when this job is done so you do not have to drive to the server room next time.

3. Six common alarms, ranked from "can wait" to "stop"

1. Predictive drive failure (PFA): can wait, but not for long

What you see: a single disk's amber light blinking slowly, and "Predictive failure" or a SMART warning in the log. The disk is still working normally inside the array; the firmware has simply decided it is dying. The right move is to hot-swap a new disk in during a quiet period and let the array rebuild. Do not wait for it to actually die. One real failure drops RAID5 to degraded, and a second one is a disaster; we wrote up a double-disk case in a separate article. When you swap: same model and capacity if possible, never smaller than the original, and confirm you have a usable backup first.

2. Memory ECC errors: treat them by frequency

What you see: "Correctable ECC" or "Uncorrectable" in the log. For an occasional correctable error, once or twice, note the DIMM slot, watch it, and replace the module in the next maintenance window. For a correctable error at high frequency, dozens to hundreds a day, or any uncorrectable error at all, schedule the work as soon as possible: an uncorrectable ECC fault takes the machine down and restarts it. Server memory is ECC, so the log names the slot, DIMM_A3 for example. Replace that module, and mind the channel pairing rules while you are in there.

3. Power supply failure: redundancy means keep running

What you see: on a dual-supply chassis, one supply goes amber and the log reports a PSU fault or "Power supply AC lost." Check the simple things first: a loose power cable, a dead outlet on the PDU, a tripped breaker on that circuit. If the supply really has failed, a dual-supply machine keeps running on the remaining one until the spare arrives, provided you do not let the second circuit fail as well. The swap is hot-pluggable. A single-supply machine means a shutdown, so do it promptly.

4. Fan and temperature alarms: the one you cannot defer

What you see: fans at full speed and noticeably louder, the panel temperature light, a threshold-exceeded entry in the log. Check the room first, because air conditioning failure is our most common callout in a Guangzhou summer. Then check the intake for a dust-packed filter, then the fan module itself. Near the limit the server throttles to protect itself, and past it, it shuts down hard, which is where the data risk lives. Deal with these immediately, even when the fix is only a vacuum cleaner.

5. RAID controller battery or supercapacitor aging: schedule it

What you see: "RAID battery/capacitor degraded" in the log, and the write policy may switch itself from Write Back to Write Through. Business slows down but stays safe. Nothing urgent, so book the spare part. A machine with an aged battery that then loses power can lose whatever was sitting in the write cache, so do not leave it for months.

6. The system-level alarm light and beep: go by the log

The overall panel alarm comes with a beep that summarizes any of the faults above. Some models let you silence it remotely from iDRAC or iLO. When the system light comes on, go straight to the log and identify the component. Do not let the noise set the pace.

4. Case: an HPE DL380 Gen10 beeping at midnight

At 1:00 a.m. on July 29, 2026, our on-call phone rang. A cross-border e-commerce company in Panyu had an HPE DL380 Gen10 in its server room, carrying ERP and the public website, beeping with an amber panel light while the business kept working.

We told them not to touch the machine and to send us the iLO address, set up during an earlier maintenance visit. The IML showed one uncorrectable ECC error on DIMM_B4, 37 correctable errors on the same module, and a SMART predictive failure on the disk in slot 3. Two faults had stacked into one system alarm.

An uncorrectable ECC error means a real risk of downtime, so this could not wait. We were on site at 2:00 a.m. with the spares held under the maintenance contract, a matching 16GB RDIMM and a 2TB enterprise disk. The business failed ERP over to its standby and took 20 minutes of downtime while we swapped the memory module and the disk and let the array start rebuilding. The rebuild ran in the background and finished the next day; nobody noticed. That machine is on an annual contract, so the visit and the parts cost nothing extra. That is what a maintenance contract buys: midnight firefighting becomes a planned job with the parts already on the shelf.

5. Three things to set up before the next incident

Configure the out-of-band port and put it under monitoring. iDRAC, iLO and XCC all speak SNMP, send email alerts and expose a Redfish API. With email alerts, a fault at 2 a.m. is in your inbox by morning instead of waiting for a user complaint. If nobody has the time to watch it, our IT outsourcing and operations service covers 24/7 monitoring and alarm response.
Keep spares. PFA disks, memory modules and power supplies account for most server failures. On a business-critical machine, keep one of each in the rack. That turns three days waiting for a courier into a swap in half an hour. Annual contracts that include a parts pool are good value.
Write an alarm-response tier list. Copy ours: temperature and power faults immediately; uncorrectable ECC or a double-disk risk within 24 hours; PFA and battery aging scheduled within a week. Pin one page next to the on-call schedule so the person on duty does not have to call and ask.

If a server lights up and you are not sure, photograph the panel and send the model plus a log screenshot to 020-39029800 (also on WeChat), and we will grade it by phone first. Panyu and Nansha, on site within two hours.

Server Warning Lights: What Amber and Red Actually Mean