Why does an added disk stay outside any existing array?

The new disk is its own device

After delivery or after the drive is added, compare names with lsblk before and after. The original two NVMe drives of about 1TB are still there, and a disk of about 2TB has appeared. TYPE on all three is disk. No new controller has merged them into one logical disk, so the partition, the filesystem, and the mount are done separately.

If the two 1TB disks were already a software array, the new 2TB disk does not become a member on its own. The member list lives in the system configuration. Adding hardware does not edit that list. Joining the new disk later is a separate decision, and when sizes differ a mirror usually keeps only the space of the smallest member.

There is no hardware array to join

The default disks are 2 × 1TB NVMe, with no hardware RAID, each disk direct-attached. The added disk is a 2TB NVMe, direct-attached, and outside any hardware array. Both sentences describe the same chassis: no array card manages these NVMe drives. A hardware volume that could be “joined” was never created.

A button labeled join-array in a panel or an install document has no matching hardware action. What can be done is a partition on the new disk in the operating system, or a software array with a tool such as mdadm and another empty disk. The second option rewrites data on the disks that join, and it fits only when those disks still hold nothing that must be kept.

Checks after the disk is added

1. Count the block devices

   nvme list should show the new 2TB device. Record the serial or the device name so a partition command is not aimed at the existing system disk.

2. Confirm it has no mount

   Before first use, MOUNTPOINT is empty. If it is already mounted, find out whether the installer or an earlier script wrote that, then decide whether to keep it.

3. Partition it alone and write fstab

   Write the UUID into /etc/fstab. Do not pin /dev/nvmeXn1. Numbers can change after reboot. The UUID does not rename itself when bay positions swap.

4. Point virtual-machine disks or a data directory at the new mount

   Proxmox can point a directory storage at a folder on the new disk. A database data directory can move there as well. Stop writes first, then start the service and check that files land on the new disk.

The charge follows the machine

Each added disk is $84 a month. Zero, one, or two disks are valid. That amount joins the monthly rent, then the month count applies: monthly × 1, quarterly × 3, semiannual × 6, annual × 12. No further discount. The 128GB machine is $455 a month and the 256GB machine is $612, before any added disk. With one added disk, the 128GB monthly rent becomes $539, and the quarterly amount is $1,617.

The memory tier and the added disk are separate. 256GB means the 256GB tier, not extra modules in the 128GB machine. An added disk does not change the CPU. Both tiers are EPYC 7543P, 32 cores and 64 threads.

Layouts that do not fit

The system disk and the new disk are not placed in RAID 0 in order to look like one volume. Any one failure then takes the system and the added data together. The new disk is also not formatted into a hardware volume name that does not exist. The command will not find that device.

When redundancy is required, the tenant still keeps a copy off the machine. The added disk has no hardware mirror. The facility is in Tokyo, Japan. Access does not change because a disk was added: Linux uses SSH, Windows uses RDP, and the address is in the console.

Moving data on the day the disk appears

The first time the new disk is online, format it as its own filesystem and mount it on a new directory, for example a path reserved for data. Copy a large directory that can pause writes, check capacity and file count, then change the service’s data path and restart that service. Do not delete the original on the old disk while the copy is still running. Keep the original until the new path has completed one full read and write.

If both added disks are present, they are still two disks, not one 4TB hardware volume. Each can have its own directory. While both are still empty they can become a software mirror or a software stripe. Before a stripe, note the failure mode: one disk failing makes files on both unreadable. A database and the only copy of site files do not belong on that stripe.

The default 1TB disks and the added 2TB disk differ in size. Putting them in one mirror usually leaves usable space near the smaller disk. The extra part of the 2TB disk does not become free capacity on its own. To use the 2TB, mount it alone. The default two 1TB disks continue to hold the system and the existing data.

The added-disk charge follows the machine’s term. Monthly billing takes that month’s $84 or $168. Quarterly prepay is three times that, semiannual six times, annual twelve times. The order offers 0, 1, or 2, and that count is written into the monthly rent of that order.

After fstab records the mount, reboot once. The check is finished when the new directory is still there and the root partition on the original system disk is still there. If the new disk does not appear, compare the added count on the order with the disk count from lsblk, and put that count in a ticket. Do not partition a device that is still uncertain.

When copying a large directory, stop the service that writes it, compare file count and total bytes, then change the configuration and start again. Database files are not copied while the service still has them open, or the copy can miss pages. When a logical export exists, export and import into the directory on the new disk. Virtual-machine disks are shut down first, then the disk file is moved, then the storage path is changed.

A software mirror of two added disks starts with both disks empty and both near 2TB. The default 1TB system disk, already holding data, stays out of that mirror. Sync lowers performance until it finishes, and workload data goes on after that. Until sync finishes, a missing member leaves the array degraded, and that state is not treated as redundancy. The only copy of data that has no backup does not sit solely on that mirror while sync is still running.