Australian cloud for business software — built and run in Sydney since 2002

NetLeverage
NetLeverage

Proxmox backup 'dirty-bitmap status: created new': why the whole disk was read

Technical article
Backup & Recovery

About this article: This article is reviewed against the cited sources. This is source research, not lab validation. Check the stated versions and your own requirements before making changes.

When a Proxmox VE backup to Proxmox Backup Server (PBS) logs dirty-bitmap status: created new for a disk, Proxmox VE had no valid record of which blocks of that disk changed since the last backup to that target. It created an empty record and, because it cannot tell what changed, it read the whole disk. The backup is not broken, and the data uploaded is still limited to chunks the server does not already hold. What you lose is time: the read of the full disk, not a small changed portion.

A Proxmox staff member wrote in 2025 that stopping the VM always invalidates the bitmap. Other documented causes are a resized disk, a switch between encrypted and plain, and a previous backup that is no longer on the server.

What the error means

Despite the look of it, this is a status line, not an error. The Proxmox VE backup code (qemu-server, the part that writes the backup log) prints one line per drive in the form <drive>: dirty-bitmap status: <text>. The possible texts in the source are:

Log text Meaning in the source
OK (drive clean) An existing bitmap was used and nothing changed
OK (X of Y dirty) An existing bitmap was used; X is what will be read, Y is the disk size
created new No usable bitmap existed; a new one was started
existing bitmap was invalid and has been cleared A bitmap was found but could not be trusted
expected bitmap was missing and has been recreated A bitmap that should exist was not there
disabled (no support) and disabled (old bitmap cleared) Bitmaps not used for this run

The code prints using fast incremental mode (dirty-bitmap), X dirty of Y total only when the dirty amount is smaller than the total. In the source, the dirty and total amounts are summed across all drives, so on a VM with several disks the line appears if any drive is incremental. If the line is absent, no drive was incremental. Read each drive's own status line to see which disks were read in full.

The Proxmox Backup Server technical overview explains why bitmaps exist. For VMs, disks are split into fixed-size chunks, "typically 4 MiB". Proxmox VE "can make use of 'dirty bitmaps', which can track the changed blocks of an image", and so "only modified chunks of the disk need to be uploaded to a backup". Without a bitmap, every block must be read and hashed again, but the server still recognises chunks it already has: "If it detects a chunk that already exists on the server, it can send only the checksum instead of data and checksum. This way the actual upload of Snapshots is incremental while each Snapshot references all chunks and is thus a full backup."

One drive is special. In the source, when a bitmap that should exist is missing on the TPM state drive, the log shows created new instead of the "missing and has been recreated" text. The code comment reads "Lie about the TPM state, because it is newly attached each time." A created new on tpmstate0 alone is expected and is not a sign of a problem.

Likely causes

The Proxmox documentation pages do not list these. The following come from the source above and from statements by Proxmox staff in the support forum, so treat them as staff guidance, not a specification.

Cause What you see How to confirm
The VM was stopped or restarted created new on every disk after a restart The VM's task history shows a shutdown, stop or reboot since the last backup. A staff member wrote in 2021 that the bitmap "is only available (atm) for running vms (and gets discarded on shutdown)"
The backup job uses stop mode (a staff member wrote in 2025 that "stopping the VM will always result in invalidating the dirty bitmap") created new every run Job mode is stop; the vzdump documentation says stop mode does "an orderly shutdown of the VM"
First backup to a new target, datastore or namespace (reasoning from the next row, not a staff statement) created new once, then OK (...) The previous backup of this VM sits on a different target. A staff member wrote in 2020 that the bitmap is cleared when the "previous backup [is] no longer available on the server", which is the case for a target that holds no earlier backup
The VM is powered off at backup time created new on every run The VM's status at the backup time. A staff member wrote in 2021 that the bitmap is "only available (atm) for running vms", and in 2022 that "the dirty bitmap is not persistent" in reply to a user whose powered-off VMs were backed up in full every time
The disk was resized or moved created new on that disk A staff member listed resize, and switching between encrypted and plain, in 2020 and said moving a disk or migrating the VM also gave a full backup
The last backup the bitmap pointed to is gone created new or existing bitmap was invalid The snapshot was pruned or deleted on the server. A staff member said in 2022 that the bitmap is cleared if "the latest snapshot (that the bitmap corresponds to) no longer exists"
Guest IDs clash with another cluster's backups on the same target existing bitmap was invalid and has been cleared on every run A staff member asked in 2022 whether several clusters with the same guest IDs back up into the same datastore and namespace. Check the backup groups on the datastore for an ID used by more than one cluster
Template No status lines printed The source skips the line for templates

Prove the cause before changing anything

These checks only read information. They are taken from the Proxmox documentation and source.

  1. Open the log of the backup task that printed created new and note each drive's status line. Compare it with the log of the previous run.
  2. Look at the VM's task history between the two backups for a stop, shutdown, reboot, disk move, resize or migration.
  3. List the backups held for the VM on the PBS storage. The Proxmox VE storage manual documents pvesm list <STORAGE_ID> --vmid <VMID> ("List volumes allocated by VMID"). If the previous backup is missing or sits on another storage, the bitmap had nothing to continue from.
  4. Check the job mode and the backup target in the job definition.

Example (illustrative): app01 (VM 105) has one 2 TiB disk and is backed up nightly to pbs01. About 40 GiB of data changes each day. On most nights the log shows scsi0: dirty-bitmap status: OK (X of Y dirty) with X about 40 GiB. After a maintenance window in which the VM was shut down and started again, the next run shows scsi0: dirty-bitmap status: created new. At a sustained read rate of 200 MiB/s, a full read of 2 TiB takes 2 x 1024 x 1024 / 200 = 10,486 s, about 2.9 hours. Reading 40 GiB takes 40 x 1024 / 200 = 205 s, about 3.4 minutes. The slow run is the first after the restart, and the next night returns to the short run. The 200 MiB/s figure is an assumption; your rate will differ.

Fix

For most causes there is nothing to repair. The next backup of a running VM to the same target builds on the new bitmap. A staff member wrote in 2025 that after a stop "there is no way around this at the moment".

  • Stop mode: if the full read each night is too long, use snapshot mode for this VM. The vzdump chapter describes snapshot mode as "the lowest operation downtime, at the cost of a small inconsistency risk", and says that with the guest agent enabled and running, Proxmox VE freezes and thaws the file system to improve consistency. Risk: changing the mode changes the consistency of the backup, so test a restore from the first snapshot-mode backup. To back out, set the mode back to stop.
  • Planned restarts: schedule the restart so that the following full read fits the backup window, and do not start other jobs that compete for the same storage.
  • Alternating targets: keep one VM backing up to one target and namespace. The bitmap belongs to the last backup of that VM on that target, so a target that holds no earlier backup has nothing to continue from. This follows from the staff statement about the previous backup being gone; no staff post states the rule for a different target in so many words.
  • Repeated existing bitmap was invalid without a restart or a change: the Proxmox documentation does not give the cause. A staff member named two in 2022: the latest snapshot the bitmap belongs to no longer exists, or guest IDs clash because several clusters back up the same IDs into the same datastore and namespace. Check both, then collect the task logs and task history and take them to the Proxmox support forum or your support provider.

Confirm the fix

  1. The first run after the event shows created new and takes the full-read time you calculated.
  2. The next run on the same running VM shows OK (X of Y dirty) and the line using fast incremental mode (dirty-bitmap), with X well under Y.
  3. A faster backup does not prove a usable one. Restore the VM to a scratch ID on a test host, boot it, and compare its data with the source. Time the restore as well, so you know the window it needs.

What to watch afterwards

  • Track how many nights per month show created new on a data disk. If a disk shows created new more often than the VM restarts, look at the other rows of the cause table (this is advice, not a documented rule).
  • Alert on a backup that runs far longer than your own recorded full-read time.
  • Keep a note of planned stops, because each one predicts a long night.
  • Remember that chunks are shared. A long read does not mean a large upload, but it does mean the VM's storage is read at full size for hours, which can slow the guest.

Risks and limits

  • These checks have not been run or lab-tested. They are taken from the Proxmox documentation, the qemu-server source and staff forum posts.
  • Proxmox documentation (the VE vzdump chapter and the PBS technical overview) explains that bitmaps exist but does not list what clears them. The cause table relies on staff forum posts from 2020 to 2025 and on the source. Behaviour may differ between versions.
  • A Proxmox staff post of 2025 says that stopping the VM always invalidates the bitmap. A newer Proxmox version could change this.
  • The log wording is from the current qemu-server master branch. Older releases may print other text.
  • The 200 MiB/s read rate and the 40 GiB daily change are invented for the example. Measure your own.

Sources