PVE publishes a guest's new status while the operation still holds
/var/lock/qemu-server/lock-<vmid>.conf, so WaitForStatusTransition could return
while the guest was still locked and the caller's next request would fail with
"can't lock file ... got timeout".
Integration run 183 failed four tests from this one cause: Restart-PveVm -Wait
returned after 4.1s having seen "running", the following Stop-PveVm spent exactly
10.0s failing to take the lock, and that cascaded into the template convert,
clone, and remove tests. Run 184 - same commit, re-run - passed because its status
poll happened to take 10.1s, by which point the lock had cleared. The same
settling happens either way; the only variable is whether the wait absorbs it or
the next caller does.
The check goes in WaitForStatusTransition because all nine lifecycle call sites
(Start/Stop/Restart/Reset/Resume across VMs and containers) route through it.
`lock` comes from the status/current response the poll already fetches - present
on both qemu and lxc since PVE 5.4, below the module's 7.0 floor - so it costs no
extra request.
If the status is reached but the lock outlasts -Timeout the cmdlet still returns
success, so a call that succeeded before this change cannot become an exception
after it.
Recorded as D015, the guest-lock sibling of D014.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
fix: New-PveCluster -Wait blocks until quorate; make cluster failures diagnosable
Adds ClusterConfigService.WaitForQuorum with a -Timeout parameter (D014), so
Add-PveClusterMember no longer hits "cluster not ready - no quorum?" after a
completed create task. Adds diagnose-cluster.sh, which captures corosync and
pmxcfs state plus PVE task logs from both nodes before cleanup destroys them.
`apt-get upgrade` holds back any package whose upgrade needs new dependencies,
so it neither pins nor updates — it produces whatever partial set the
dependency graph allows that day. On these nodes it upgraded pve-cluster to
9.1.6 while leaving libpve-cluster-api-perl at 9.1.0.
Those two ship the halves of the join: cfs_backup_database() in
PVE/Cluster.pm (pve-cluster) and finish_join() in PVE/Cluster/Setup.pm
(libpve-cluster-api-perl). Upstream removed `return $dbfile` from the former
and stopped relying on it in the latter, both at 9.1.1 — 9.1.6's finish_join
calls cfs_unlink_db_unsafe() instead. The 9.1.0 caller against the 9.1.6
callee unlinks an empty string, so the standalone config.db survives the join,
pmxcfs restarts in local mode, and the node reports online=0 forever while
corosync forms a healthy 2-node membership. That is the "2 nodes online"
failure, and it is not reachable on any coherent install.
The ISO is the pin, so drop the upgrade and install only what the harness
needs. Upgrades belong in a separate currency lane that records the package
set it tested.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Run 173 left node B with healthy corosync (2-member primary component, both
links connected) but no /etc/pve/corosync.conf, no dcdb/status journal lines,
and pvecm status reporting it is not part of a cluster. That file is
database-backed: pmxcfs creates it only when it starts with no config.db and
imports /etc/corosync/corosync.conf, so a surviving standalone config.db would
mean silent local mode.
Capture the package versions, pmxcfs command line, /etc/pve mount, .members,
the config.db and its backup dir, whether the database holds a corosync.conf
row, and the CPG group membership. Read-only; the sqlite3 CLI is not guaranteed
on a PVE node, so fall back to strings.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Also documents the two distinct timeout conventions in the module so the
-Wait vs HTTP-client distinction does not have to be re-derived.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
New-PveCluster -Wait now guarantees quorum, so the Start-Sleep -Seconds 5
that lost the race is replaced by an assertion on the new contract.
The link0 pin is dropped: the join-abort and the never-a-member modes both
occurred with and without it, so it was never implicated, and -Links is
already covered at the service level.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Returns once quorate, throws TimeoutException when quorum never arrives,
and keeps polling through a PveApiException from the restarting API.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Polls GET /cluster/status for the cluster entry with quorate = 1, bounded
by a timeout (60 s default) and tolerating PveApiException while pmxcfs
and corosync restart during cluster formation.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
PVE's cluster-create task completes before corosync converges. Until the
node is quorate it rejects a join with "cluster not ready - no quorum?",
so New-PveCluster -Wait followed by Add-PveClusterMember failed for every
caller. Wait for quorum after the task, bounded by -Timeout (default 60 s,
following the -Wait timeout convention used by Stop-PveContainer and
Reset-PveVm). See DECISIONS.md D014.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
"Cluster join aborted!" is PVE's generic wrapper; the reason lives only in
the task log on the joining node. Run 172's log said "An error occurred on
the cluster node: cluster not ready - no quorum?", which is what identified
the race. Capture it so the evidence survives cleanup.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The diagnostic step runs before cleanup so it reaches the nodes while
they still exist. keep_vms skips cleanup entirely for a dispatch run,
leaving the nested nodes up for hands-on inspection.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The API reports a joined-but-offline node as online=0 with no further
detail, and the cleanup job destroys the nodes minutes later, so the
reason corosync membership never forms has never reached a log. Read
corosync.conf, corosync-cfgtool, pvecm status and the corosync journal
off both nodes while they are still alive. Best-effort: never fails the
caller.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The "2 nodes online" check has never passed — run 159 shows the identical
failure while the job reported success (swallowed Pester exit code), and
run 170's 60 s poll expires with node B still online=0. Node B reaches
corosync.conf (Get-PveClusterConfigNode passes) but never becomes a
corosync member, and the suite captures nothing about why.
Pass link0 explicitly so ring0 is the address the harness verified node B
answers on, rather than whatever node B resolves its own DHCP-assigned
.test.local hostname to. Dump quorate and each node's ring0/online/local
unconditionally so the next run distinguishes a wrong ring0 address from
a working address with no corosync transport.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The join task completing does not mean corosync membership has reached
the status endpoint; the instant assertion failed intermittently (runs
159 and 169) while every other cluster check passed.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The repo stops holding opinions about CI VLAN addressing: the VM boots
via DHCP with hostname pvetest-storage (deterministic MAC for an
optional reservation), registers in the operator's CI DNS zone, and
everything addresses it by STORAGE_VM_FQDN. Replaces the static-IP +
explicit-DNS variables.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Run 168 burned 13 minutes to report 'unable to locate package tgt' when
the real failure was dead egress from the CI VLAN (apt lists never
updated). The setup script now probes gateway/internet/DNS from inside
the VM and prints the verdicts, fails apt fast with the real error, and
marks each phase.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
PVE 8 reached EOL 2026-08-31. The full matrix needs 42GiB free on one
parent node; no node has that after the fleet rebuild, so the headroom
guard blocks every run. Dropping the 8a/8b pair brings the requirement
to 26GiB. The module keeps PVE 8 runtime support and the script still
accepts PVE_VERSIONS='9 8' for local runs; the PVE-8-only Pester
contexts stay for SKIP_PROVISION runs against real PVE 8 hosts. Also
counts the storage VM in the headroom error message.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
run-integration.sh: two-phase apply (storage VM configured before the
PVE guests boot), Pester PassThru with explicit failed/zero-test checks,
auto-ISO cache names embed the answer host, force-cleanup covers the
storage VM and imported cloud image, docker usage removed.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The nested PVE guests live on an isolated VLAN with no route to the
runner network (ADR 0032, homelab repo), and the runner-hosted Docker
services died with each ephemeral ARC pod anyway. A small cloud-image VM
in the ci pool now serves NFS, iSCSI, and the auto-install answer files
from inside the sandbox. Includes fixes from Codex and in-house review:
bpg provider >= 0.79.0, serial console for the resized cloud image,
key-based SSH (cloud images refuse password auth), CIDR validation,
storage VM in the headroom check, docker socket mounts removed.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
PVE_TARGET_NODE becomes optional: unset (or 'auto') queries
/cluster/resources for the online node with the most free memory and
refuses to provision when no node has vm_count x 8GiB + 8GiB headroom
free. The chosen node persists on the shared mount so cleanup and
force-cleanup aim at the node provision picked.
2026-08-31: a provision onto the pinned node OOM-killed a 16GiB
services-plane VM on the parent hypervisor; the pin predated a memory
bump that consumed the headroom it silently relied on. Needs PVEAuditor
on /nodes for the token's user (a non-privsep token ignores token ACLs).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
goodolclint-claude[bot]