Still present in 26.04 LTS, fixed in 26.10 — here is the measurement, and a request for an SRU.
**Versions, checked in the archive today** * `resolute` (26.04 LTS): `ceph 20.2.0-0ubuntu2.1`. The shipped `ceph-volume_20.2.0-0ubuntu2.1_all.deb` still contains `ExecStart=/bin/sh -c 'timeout $CEPH_VOLUME_TIMEOUT @CMAKE_INSTALL_PREFIX@/sbin/ceph-volume-systemd %i'`. Nothing newer is published for `resolute`, not even in `-proposed`. * `stonking` (26.10) and `tenacious` (27.04): `ceph 20.2.4+ds1-0ubuntu1`. The same file in that package reads `/usr/sbin/ceph-volume-systemd`. Fixed. **Upstream**: fixed by ceph/ceph#66861 (tentacle backport of #66670), merged 2026-02-12. It adds `ceph-volume@` to the `configure_file` loop in `systemd/CMakeLists.txt` and stops installing the raw file, renaming it to `systemd/[email protected]`. That `.in` file exists from tag `v20.2.1` onward, so `v20.2.0` — exactly what 26.04 ships — is the only Tentacle release carrying the defect. **Impact on a 26.04 LTS cluster**: no OSD comes back after a reboot, on any node, from the day it was installed. `ceph-volume lvm create` is the documented way to create an OSD, and `ceph-volume@lvm-<id>-<uuid>.service` is what brings it back at boot; with the placeholder that unit exits 127 (`timeout: failed to execute process: No such file or directory`) before `ceph-volume-systemd` ever runs. There is no fallback an operator can reach for: `ceph-osd@<id>.service` is enabled `--runtime`, so its link lives under `/run` and is gone after a boot; `ceph-osd.target` wants no instance and reports `active` regardless; and `/var/lib/ceph/osd/ceph-<id>` is the tmpfs the activation mounts, so starting the daemon by hand finds an empty directory and no keyring. On a six-node cluster, every power cycle left the monitors, managers and metadata servers up and **0 of 6 OSDs**, i.e. the whole data path down until a human intervened. Nothing in the OSD logs says why, because the OSD never started. **Test case** (26.04, one OSD created with `ceph-volume lvm create`) 1. `systemctl cat [email protected] | grep ExecStart` → shows `@CMAKE_INSTALL_PREFIX@`. 2. Reboot the node. 3. `ceph osd tree` → the OSD is `down`, and `journalctl -b -u 'ceph-volume@*'` → `Main process exited, code=exited, status=127/n/a`. With the fix, or the workaround below, step 3 reads `Finished ceph-volume@lvm-<id>-<uuid>.service - Ceph Volume activation` and the OSD is `up` and `in`. **Workaround**, verified on six arm64 nodes running 26.04.1 and then verified across a full power cycle of all six: mkdir -p /etc/systemd/system/[email protected] cat > /etc/systemd/system/[email protected]/override.conf <<'EOF' [Service] ExecStart= ExecStart=/bin/sh -c 'timeout $CEPH_VOLUME_TIMEOUT /usr/sbin/ceph-volume-systemd %i' EOF systemctl daemon-reload The empty `ExecStart=` is required: `ExecStart` is a list-typed setting, so without it systemd appends to the broken command instead of replacing it. **Request**: could this be nominated for a 26.04 SRU? The delta is one line in the shipped unit, or a cherry-pick of the upstream commit. As it stands, `ceph-volume lvm create` on an LTS release produces OSDs that silently never survive a reboot, and the failure is invisible until the next power cycle. Bug #2160244, reported 2026-07-09, looks like the same defect and could be marked as a duplicate of this one. -- You received this bug notification because you are a member of Ubuntu Bugs, which is subscribed to Ubuntu. https://bugs.launchpad.net/bugs/2167962 Title: [email protected] ships with unsubstituted @CMAKE_INSTALL_PREFIX@ To manage notifications about this bug go to: https://bugs.launchpad.net/ubuntu/+source/ceph/+bug/2167962/+subscriptions -- ubuntu-bugs mailing list [email protected] https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
