I powered down an HP DL380 G7 in early 2025 and did not touch it for fifteen months. It sat in a closet through a job change. On 27 May 2026 I went back to it.
Would the PSUs hold. Would the disks spin clean. Would Proxmox boot. Would I remember the passwords.
First boot
Front panel button. Fans spun up. POST stopped on a memory training failure: one DIMM in CPU socket 1, "uncorrectable error during training."
Almost opened the chassis to reseat. Waited ten minutes first. A forum post had warned me cold-stored hardware needs a few minutes to settle. Power cycled. Clean on the second attempt. No memory complaints in any boot since.
Wait through the first symptom once before you touch anything.
Proxmox came up easily
F11 at POST. Picked the boot SSD on the P410i. Proxmox came up on kernel 6.5.13-3-pve. Root password I still remembered, typed at the USB keyboard plugged into the back of the chassis.
First job at that console: move the host onto the current network. The VLAN and address from 2024 were gone. Edited /etc/network/interfaces over vi, ran ifreload -a, watched the new bridge come up, walked back to the laptop. SSH worked first try.
From there everything ran over SSH. qm list returned twenty-two VMs and one LXC. None running. Past-me had set onboot=0 on every guest before the shutdown, which is a discipline I have not always had.
Started the NAS (qm start 189), let OpenMediaVault bring up its NFS exports, checked cat /proc/mdstat:
md0 : active raid5 sdd[2] sdc[1] sdb[0]
838860288 blocks super 1.2 level 5, 512k chunk, algorithm 2 [3/3] [UUU]
md1 : active raid5 sdh[3] sdg[2] sdf[1] sde[0]
1431655936 blocks super 1.2 level 5, 512k chunk, algorithm 2 [4/4] [UUUU]
Both arrays clean. The btrfs filesystems on top were readable. Fifteen months of cold did nothing to the storage layer.
iLO was the hard half.
The iLO problem
I wanted remote console access. The DL380 G7 has iLO 3, HP's out-of-band management processor from 2010. The final firmware shipped in December 2020, version 1.94 (HPE iLO firmware list).
iLO 3 v1.94 only negotiates TLS 1.0. Chrome returned ERR_SSL_VERSION_OR_CIPHER_MISMATCH with no override. Edge same answer. Firefox connected after I set security.tls.version.min = 1 in about:config. Fresh Firefox profile, dedicated to this BMC and nothing else.
The TLS workaround I had. The bigger problem was the iLO IP I had configured in 2024 was on a VLAN that no longer existed in 2026. Admin password I still remembered. I just had no route to the login page. The Proxmox host had a current address, but iLO sat on a dedicated NIC behind one that nothing on the current network could speak to.
In-band IPMI saved me a chassis open.
In-band IPMI
iLO speaks IPMI 2.0 over a KCS (Keyboard Controller Style) channel that the host OS sees as /dev/ipmi0. Root on the host gives you admin access to the BMC. No BMC password needed; you already proved you have the keys to Linux.
Debian 12 needed ipmitool. The host had no internet (network was still half-rebuilt), so I sideloaded the .deb stack from the laptop: ipmitool 1.8.19-4+deb12u2 plus freeipmi-common 1.6.10-1 and libfreeipmi17 1.6.10-1+b1, all SHA256-verified. Then:
$ ipmitool user list 1
ID Name Callin Link Auth IPMI Msg Channel Priv Limit
1 true false false Unknown (0x00)
2 kay true true true ADMINISTRATOR
Admin in slot 2 (slot 1 is the IPMI-spec empty placeholder). Always list before you write.
Rotated the admin password. The recorded one still worked. Cheap insurance.
$ ipmitool user set password 2 'a-new-strong-password'
$ ipmitool channel setaccess 1 2 callin=on ipmi=on link=on privilege=4
The actual rescue was the IP. LAN on iLO 3 is channel 2, not 1. Channel 1 errors out, and anything you read online about "channel 1 LAN" is iLO 4 or non-HP.
$ ipmitool lan set 2 ipsrc static
$ ipmitool lan set 2 ipaddr 10.99.99.3
$ ipmitool lan set 2 netmask 255.255.255.0
$ ipmitool lan set 2 defgw ipaddr 10.99.99.1
$ ipmitool mc reset cold
Sent cold reset command to MC
The cold reset is where I burned forty minutes the first time. iLO 3 has gratuitous ARP disabled. Changing the IP does not make the BMC announce its new binding, and the upstream switch keeps the stale ARP entry for the old IP. ipmitool mc reset cold power-cycles the BMC's network stack, which is what forces the announce. Without it you watch pings drop, certain one of your lan set commands was rejected, when in fact all four were accepted just fine.
One more iLO 3 gotcha. The BMC does not log IPMI admin operations to the SEL that ipmitool sel list returns. The audit trail lives in HP's Integrated Management Log, reachable only through the web UI, RIBCL, or hponcfg, and RIBCL needs the same TLS handshake modern OpenSSL refuses. So you do recovery work and there is no on-host record. Keep your own log.
What it taught me
iLO was harder than Proxmox. That ratio will hold for any pre-2014 enterprise server. Hypervisor and storage are still maintained. The BMC is a frozen artefact of whatever firmware shipped before the vendor stopped caring.
Treat it like that. BMC on its own VLAN, no internet path. Credential on paper outside the BMC; recovery is in-band IPMI, which needs the host up. Cold-start drill once a year. Starting from scratch today I would buy a newer server.
The box came back. Data intact. Three sessions over four days. Fifteen months of cold did nothing irreversible.
Resources
- HPE iLO firmware history (Wikipedia)
ipmitoolupstream- Where this fits: the DL380 G7 stays as temporary quorum until a 3-node EliteDesk G5 plus WRX90 Threadripper Pro cluster takes over. More on that build in a future post.
