Extending an attached volume. Refused, error_extending, or the instance still sees the old size¶
Symptom¶
One of these, after openstack volume set --size <new-GB> <volume-id> on a volume attached to an instance:
- the command is refused:
Failed to set volume size: --os-volume-api-version 3.42 or greater is required to extend in-use volumes.followed byOne or more of the set operations failed; or, also from the client before any API call,Volume is in <status> state, it must be available or in-use before size can be extended.orNew size must be greater than <N> GB; or, from the API,Volume <volume-id> status must be '...' to extend, currently <status>.; - the command is accepted, but the volume keeps its old size and status, with a user message
extend volume:Could not find any available weighted backend.; - the volume goes to
error_extending; - Cinder shows the new size, but inside the instance the disk still has the old size.
Impact¶
One volume and the instance that uses it. Data is not at risk from the extend itself; a volume in error_extending
cannot be extended or detached until its status is corrected (a detached one can still be deleted; an attached one
cannot).
Quick checks¶
Read-only. With admin credentials:
# 1. Cinder's view: size, status, attachment
openstack volume show <volume-id> -c size -c status -c attachments -c os-vol-host-attr:host
# 2. Messages Cinder recorded for the volume (user messages, API 3.3+)
openstack --os-volume-api-version 3.3 volume message list | grep <volume-id>
# 3. The instance's actions: Nova receives a volume-extended event
openstack server event list <server-id>
Inside the guest, the disk's size as the guest sees it (lsblk).
Logs: the cinder-volume service that owns the volume, and the nova-compute of the instance's compute:
Kolla-Ansible:
sudo grep -E 'Extend volume|Failed to extend|volume-extended' /var/log/kolla/cinder/cinder-volume.log | tail
sudo grep <volume-id> /var/log/kolla/nova/nova-compute.log | grep -iE 'extend' | tail # on the compute
sudo grep <volume-id> /var/log/kolla/cinder/cinder-scheduler.log | tail # extend not placed
The cinder-volume lines Extend volume completed successfully., Extend volume failed. and Failed to extend
volume. do not carry the volume ID (only the Nova event response line does, as tag): match them by time, or by
the request ID that volume message list shows.
Any: the same grep on the cinder-volume, cinder-scheduler and nova-compute logs.
Diagnosis¶
How an attached volume is extended (Cinder and Nova 2026.1)¶
- Cinder API (
cinder/volume/api.py,_extend): an attached (in-use) volume can be extended only with volume API 3.42 or later (VOLUME_EXTEND_INUSE); python-openstackclient refuses before calling the API without it. The new size must be larger, and within the quota and the volume type's size limits. The status goes toextending. The request then goes throughcinder-scheduler, which checks the backend's capacity; if it finds none (Filtering removed all hosts for the request with volume ID '<volume-id>',Failed to schedule_extend_volumeincinder-scheduler.log), the volume goes back to its status at its old size, with the user messageextend volume:Could not find any available weighted backend.; the client command itself succeeded. cinder-volume(cinder/volume/manager.py,extend_volume) asks the backend driver to grow the volume:- on failure:
Extend volume failed.andFailed to extend volume.in the log, a user messageVolume Driver failed to extend volume., and the statuserror_extending; - on success: the size is updated, the status goes back to
in-use,Extend volume completed successfully.is logged, and Cinder sends Nova avolume-extendedevent for each instance the volume is attached to.
- on failure:
nova-compute(nova/compute/manager.py,extend_volume) logsCinder extended volume <id>; extending it to detect new size, then asks the libvirt driver to refresh the device: for iSCSI and Fibre Channel, os-brick rescans the host's device; for network volumes such as Ceph RBD there is no host device to rescan. Then QEMU is told the new size, and the guest sees a larger disk.- On failure,
nova-computelogsExtend volume failed, volume_id=<id>, reason: <error>. Cinder is not told: it shows the new size andin-use, while the guest still sees the old size. - If the guest is not running, Nova skips the resize (
Skipping block device resize, guest is not running, not logged at the default INFO level); the guest sees the new size at its next start.
- On failure,
The guest¶
Even when the disk is larger, the partition and the file system inside the guest stay as they were: growing them is a step inside the guest, for its owner.
Possible causes¶
Cause 1: refused by the API or the client¶
An in-use volume without API 3.42; a size not larger than the current one; a quota or volume type limit; a volume
not in available or in-use; or a policy that forbids extending attached volumes
(volume:extend_attached_volume).
Cause 2: the backend could not grow the volume (error_extending)¶
The cinder-volume log has the driver's error: a backend limit, a backend unreachable, a backend volume that
cannot be found. When the driver is not initialised, the volume also goes to error_extending, but without a user
message. When the pool has too little space for the new size, the scheduler refuses the extend before the
driver is called: no error_extending, the volume keeps its old size and status, and the user message is
extend volume:Could not find any available weighted backend. (see the diagnosis).
Cause 3: Cinder grew it, Nova did not refresh it¶
The volume-extended event did not reach Nova, or the refresh failed on the compute (Extend volume failed), or the
instance's nova-compute is down. The guest sees the old disk size.
Cause 4: the disk is larger, the file system is not¶
lsblk in the guest shows the new size, but the partition and file system were not grown.
Resolution¶
For cause 1: use the right API version, or the right size¶
⚠️ State-changing. Grows the volume on the backend; a volume cannot be shrunk afterwards. The project's quota is charged for the new size. Undo: none (create a smaller volume and copy the data, if needed).
openstack --os-volume-api-version 3.42 volume set --size <new-GB> <volume-id>
For cause 2: fix the backend, reset the status, extend again¶
Fix what the cinder-volume log names. The volume's size in Cinder is still the old one. Then put the status back to
what it was before the extend:
⚠️ State-changing. Changes Cinder's record only; the backend is not touched. Precondition: the backend volume still has its old size and is healthy (check on the backend), and the volume is attached as Cinder says. Undo:
cinder reset-state --state error_extending <volume-id>(openstack volume set --statedoes not accepterror_extending).
openstack volume set --state in-use <volume-id> # or available, if it is not attached
Then extend again (cause 1).
For cause 3: make Nova refresh the device¶
Bring the instance's nova-compute back if it is down, and read the Extend volume failed reason. If
nova-compute was only down, the event waits in its message queue: once it is back, look for Cinder extended
volume <volume-id> in its log and check lsblk in the guest before anything else. The refresh is triggered by
Cinder's event, which Cinder does not send again: if the event was lost or the refresh failed, the simplest ways to
make the guest see the new size:
⚠️ State-changing. A hard reboot stops the guest abruptly: what it had not yet written to disk is lost. A detach and attach interrupts its access to that volume (unmount it inside the guest first). Agree it with the owner. Undo: none needed.
openstack server reboot --hard <server-id> # the volume is reattached at its new size
or, for a data volume, detach and attach it again (openstack server remove volume then openstack server add
volume).
For cause 4: grow the partition and the file system in the guest¶
The owner does it inside the guest, with the guest's tools (for example growpart and resize2fs or xfs_growfs
on Linux).
Verification¶
openstack volume show <volume-id> -c size -c status # the new size, in-use
Inside the guest: lsblk shows the new size, df -h the grown file system.
Lab observations¶
Kolla-Ansible 22.2.0 (2026.1) all-in-one, Cinder LVM (thin) over iSCSI, python-openstackclient 9.0.0, a CirrOS instance with a 1 GB data volume, on 2026-10-27:
- Normal path:
Extend volume completed successfully.2 s after the command,Nova event response1 s later,Cinder extended volume <volume-id>; extending it to detect new sizeinnova-compute.log1 s after that, then os-brick'sExtend volume for {...}; the volume wasin-useat 2 GB andlsblkin the guest showed 2 GB. - Cause 1:
volume set --size 2without a microversion printedFailed to set volume size: --os-volume-api-version 3.42 or greater is required to extend in-use volumes.andOne or more of the set operations failed, exit code 1. - Extend beyond the thin pool's capacity (500 GB): exit code 0,
CapacityFilterremoved the only backend (Insufficient free virtual space (376.0GB) to accommodate thin provisioned 496GB volume), the volume stayedin-useat its old size, user messageextend volume:Could not find any available weighted backend. - Cause 2, injected by renaming the backend LV:
error_extendingwithin 8 s,Extend volume failed.(Failed to find logical volume) andFailed to extend volume., user messageextend volume:Volume Driver failed to extend volume.. After the LV was renamed back,volume set --state in-useand the extend again worked. - Cause 3,
nova-computestopped (systemctl stop kolla-nova_compute-container.service), then an extend: Cinder showed the new size andin-useand loggedNova event response: {'name': 'volume-extended', ... 'status': 'completed', 'code': 200}; the guest kept the old size. Aftersystemctl start,nova-computeprocessed the queued event 18 s later and the guest saw the new size without a reboot. With the event removed from the compute's queue whilenova-computewas down (to simulate a lost event), the guest kept the old size after the start;server reboot --hard(22 s with--wait) brought the new size to the guest. - Guest shut off during the extend: no
Skipping block device resizeat INFO level; the guest saw the new size at its start.
Validity and limits¶
- Lab-run on Kolla-Ansible 22.2.0 (2026.1), all-in-one, Cinder LVM over iSCSI (see Lab observations): causes 1, 2
and 3, the capacity refusal and the shut-off guest. Not run: the detach and attach alternative of cause 3, cause 4
(no file system was grown in the guest), a refresh failure on the compute (
Extend volume failed), Ceph RBD and Fibre Channel. - Written from Cinder 2026.1 (
cinder/volume/api.py_extend,cinder/volume/manager.pyextend_volumeandextend_volume_completion,cinder/message/message_field.py, the 3.42 microversion), Nova 2026.1 (nova/compute/manager.pyextend_volume,nova/virt/libvirt/driver.pyextend_volume,nova/virt/libvirt/volume/net.py) and python-openstackclient 2026.1. - Not covered: shrinking, encrypted volumes (Nova handles them differently), multi-attached volumes, and extending a root volume of a running instance from the guest's point of view.
Related¶
- Volume stuck in attaching or detaching
- Cinder 2026.1: API microversion history
Provenance: written by agent:ozarland-author, from the Cinder, Nova and python-openstackclient 2026.1 source and documentation. Reviewed in its pull request (see the page history). Checked by a human operator: not yet.
Worked for you, or not? Tell us in one line: your release, your deployment method, and what happened.