A quiesced snapshot and a live clone freeze the guest's file systems
through libvirt, which asks the agent without a timeout: with the agent
hung, the job waited, holding the VM, with no end.
The agent must now first answer a guest-ping within 10 s; a frozen or
dead agent fails the job in seconds, before anything is frozen. The
freeze itself stays libvirt's, without a timeout: a timed freeze was
tried and left the guest frozen, since the freeze still ran in the
guest when the agent came back, after its thaw had failed.
The IP search and the VM report read each running VM's addresses and
OS from its guest agent with interfaceAddresses() and guestInfo(),
which wait for a hung agent indefinitely. The search answered by its
deadline, but the thread stayed blocked, counted against the search
thread limit and held up the worker's exit.
They now use the 10 s agent command the VM page uses, shared as
vrtManager.instance.agent_command() and agent_interfaces(). With the
agent frozen, a search answers in 5 s from the leases and the ARP
table, counts the VM unread, and its thread ends about 5 s later.
The VM page read the guest's addresses with interfaceAddresses() from
the agent, and the OS details with agent commands without a timeout.
A connected agent that hangs (a frozen guest) kept the page from
loading at all and held a worker: seen still loading after 120 s.
Both now go through the 10 s guest agent command the root password and
SSH key use. The addresses come from guest-network-get-interfaces, in
the form the page's address code already reads; an agent that does not
answer leaves the ARP address. With the agent frozen the page loads in
about 10 s and the OS details answer with the agent's error (400) in
5 s, instead of a server error.
The Access tab's Root Password and SSH Keys worked only on a shut-off
VM, through gstfsd on the compute. A running VM whose QEMU guest agent
is connected now gets them through the agent: guest-set-user-password
with the SHA-512 crypt hash, and guest-ssh-add-authorized-keys adding
the key to root's keys. A shut-off VM still goes through gstfsd; any
other VM is refused with "Shut the VM down, or start its QEMU guest
agent, and try again", and the buttons are disabled.
Each is one agent command with a 10 s timeout. libvirt's own calls for
them (setUserPassword, authorizedSSHKeysSet) wait for a hung agent
indefinitely, holding a worker and the VM's lock, and the domain's
agent timeout would change every other agent call of the VM too.
The agent's or libvirt's error is shown as is through error_text, so
users who do not see the hosts get no connection details, and the
attempt is logged as failed. The README names the SELinux boolean a
RHEL-family guest needs before its agent may add SSH keys.
libvirt's logical pool cannot change a volume's capacity, so every
resize of such a disk ended in a 500. Disks are resized one at a time;
VIR_ERR_NO_SUPPORT becomes a message naming the disk, and the other
disks still grow.
A live test creates an LVM pool from the storage page on a spare disk
(TEST_LIBVIRT_LVM_DEVICE; the Vagrant computes get /dev/vdb), attaches
new and existing volumes to stopped and running VMs as block disks, and
checks the refused resize in every state.
Settings > Console gets a Console type field. Serial removes the VM's
graphics device (adding a pty console when there is none); VNC adds the
graphics device new VMs get, on localhost, and then applies the form's
VNC fields, so an administrator can open it to all interfaces for a TCP
compute. Serial is offered only while SERIAL_CONSOLE_ENABLED is on.
The move copies the VM's NVRAM file to the destination with libvirt
streams (unless both hosts share the NVRAM directory) and removes the
copy if the move fails; the source's file goes with the existing
cleanup. The undefine keeps the file, as libvirt refuses to undefine a
VM with NVRAM otherwise. A VM with TPM state is still refused.
libvirt's backup copies the disks of a running or paused VM from one
instant into new volumes while it keeps running; the clone is defined
shut off as before. The VM is held for the copy, the copy can be
cancelled, and recovery reads the job's records and libvirt.
The clone's definition is one method for both paths; wvmInstance can check,
start, follow, read and abort a push backup of a VM's disks into new
volumes, and freeze/thaw the guest.
A disk-only snapshot was taken without any space check, though its
overlays grow as the guest writes and a later delete merges them down;
a full pool pauses the guest and can leave the snapshot undeletable.
Only a memory snapshot was checked, for the memory alone.
Before a snapshot, each storage pool with a disk of the VM must have
free a share of those disks' virtual size (10%, at least 1 GiB), and
the pool of the memory file the VM's memory on top. Both values are on
the Settings page (Snapshot Free Percent, Snapshot Free Min GiB; 0
turns a rule off, an invalid value keeps the default). The pools are
refreshed first, and a failed refresh or an unreadable volume refuses
the snapshot instead of skipping the check; a disk in no pool is not
checked. Nothing is reserved: later writes can still fill a pool.
A disk-only snapshot of a running VM is crash-consistent: data the
guest has not written yet is missing from it. With a connected QEMU
guest agent, the snapshot form now offers to freeze the guest's file
systems while the disks are snapshotted (libvirt's QUIESCE flag); the
box is checked by default and can be cleared. It is offered only for a
running VM whose agent is connected, and not together with memory:
libvirt accepts quiesce only for disk-only snapshots, and a memory
snapshot keeps the whole guest state anyway. A request that does not
fit is refused before libvirt is called.
libvirt keeps snapshot metadata on the source host and a migration does
not carry it, even when both hosts share the disks. Only the shut-off
path refused such a VM itself; a running or paused one relied on
libvirt's own refusal. moveto now refuses it for every state, before
any check against the destination, with the same message.
After a snapshot revert or delete, files libvirt could not remove are
named in a warning that says nothing uses them and asks to remove them
by hand. A file another VM on the same host uses (as a disk, a backing
file or a snapshot file) is no longer named: removing it would break
that VM. The warning now says no VM on this host uses the files.
libvirt deletes any volume it is asked to. The storage page and the API
deleted a VM's disk, the base image of another volume or the file a
snapshot was taken on without a check, breaking the VM or its
snapshots.
wvmStorage.del_volume now refuses a volume a domain on the compute uses
as a disk or a backing file of one, a file of a domain's snapshots, or
the backing file of another volume. The check is the one destroy
already used, moved to wvmConnect and extended to snapshot files, so
destroy also keeps the files of another VM's snapshots. The storage
page shows the refusal; the API answers 400 with it, as for other
refusals. Deleting a disk from a VM refuses after the detach when the
VM's own snapshots need the volume, and says the volume was kept.
Network disks count by their source name, pool volume disks by their
volume's path. A volume whose metadata cannot be read stops the check
instead of counting as unused; only one removed meanwhile is skipped.
The check and the delete hold the compute lock, so no VM operation
attaches the volume in between; a busy compute is a message on the page
and 409 in the API.
Deleting a VM with its data deleted only the images its disks pointed
at. With external snapshots that is the newest overlay: the image the
snapshots were taken on, the older overlays and the memory files stayed
on disk, with nothing referring to them once the VM and its snapshot
metadata were gone.
The files the VM's snapshots were taken on, down each disk's backing
chain, the overlays of snapshots kept off the chain after a revert, and
the snapshots' memory files are now the VM's own too. A template the
VM grew from ends the chain and stays, as does an overlay of a disk
detached since. The backing references of the VM's own volumes no
longer count as another domain's use, and a file another VM uses keeps
the files below it.
A backing file that cannot be read (other than one in no pool) stops the
split before anything is deleted, instead of ending the chain there.
The snapshot tab listed the VM's snapshots three times per page: once
per kind and once more for all of them. The list of all is now made
from the two kinds, so the combined flag is no longer used anywhere.
The revert confirm named only the disks the persistent definition
would lose. A disk hot-plugged into the running VM alone is detached
too, so the live definition is compared with what ran when the
snapshot was taken.
A storage pool refresh error after libvirt had already made, reverted
or deleted the snapshot was shown as if the operation had failed. It
is now logged, and the left-behind check is skipped then, since a stale
pool still lists the removed files. A volume removed while that check
reads it is skipped too.
An error reading the block jobs no longer breaks the instance page;
create, revert and delete still query the jobs again and refuse.
When libvirt cannot remove a snapshot file it merged or discarded, it
only logs a warning and the delete or revert still succeeds. On NFS it
removes them as the qemu user, so an image directory the qemu user may
not write to left every merged overlay on disk without a word.
Revert and delete now compare the files in use before and after and
show a warning naming the ones still on disk. Only a snapshot's own
files (overlays, memory) and overlays on a snapshot point count, and a
disk the revert detached is never named, so a kept disk is not offered
for removal.
Create, revert and delete are one libvirt call each. The pflash-to-rom
loader rewrite, the hand-written block commit and the s1./s2. names are
gone, so UEFI VMs can take external snapshots, a revert keeps disks added
later, and deleting an older snapshot keeps the newer one.
Create refuses what libvirt could not revert or delete later: internal
snapshots on the VM, a saved state, a running snapshot job, disks that
are not files, CD-ROM media, a branch after a revert, and memory without
a running VM or room in the first disk's pool. A disk-only revert keeps
the VM running, paused or shut off; one with memory resumes it as saved.
A revert in the same second as the previous one, which libvirt fails
after stopping the VM, is retried once.
A snapshot the old code took of a UEFI VM saved the loader as rom; its
revert is refused (deleting it still works).
The snapshots tab lists internal and external snapshots separately, can
save memory too, shows running snapshot jobs, and the revert confirmation
names the disks it detaches and that NVRAM is not restored.
On an active VM a device was attached in two calls, running VM first.
When the persistent attach was refused, for example because a disk target
or an agent channel was already pending there, the device stayed in the
running VM only and was gone at the next start. A NIC added without a
MAC got a different generated MAC in each definition. One call with
AFFECT_LIVE|AFFECT_CONFIG lets QEMU check the persistent copy before the
hotplug and keeps one MAC. Detach and update still go per definition,
each from its own XML.
On an active VM the boot order editor, the video model select and the
console form showed the running VM: a pending boot order, video model or
console setting looked unset, and the boot order editor offered devices
of the running VM only. Choosing one of those cleared the pending order,
since the setter dropped the old order and skipped devices it could not
find. They now read the persistent definition (get_bootmenu,
get_bootorder, get_media_devices, get_video_model and the console
getters take config=), and set_bootorder refuses a device that is not in
it before it changes anything.
On an active VM the options and CPU forms write the persistent definition
but showed the running VM's title, description and vCPUs, so saving one
posted the live values back over a pending change. They now read the
persistent definition (get_vcpu, get_cur_vcpu, get_title and
get_description take config=).
Buttons in the settings and resize tabs that were disabled only by class
get the disabled attribute: Enter in a field no longer submits a form
the page shows as unavailable while the VM runs.
The QoS table showed the running VM's bandwidth while an edit changes the
definition for the next start. Each row posts average, peak and burst,
so an edit wrote the running values back over a pending change, and a
direction added while the VM ran had no row. The table shows the
persistent definition now. An edit also kept only the posted values and
dropped an inbound floor; it now updates them and keeps the rest.
Enabling hot plug on a stopped VM wrote every vCPU up to the maximum as
enabled, so a VM with 2 of 4 vCPUs booted with 4. The current vCPUs stay
enabled now; libvirt adds the rest as disabled hotpluggable vCPUs.
Only a running VM counted as active, so a paused one (or one blocked,
PM-suspended or shutting down) had its maximum and current memory written
for the next boot and nothing applied live. Every active VM now has its
current memory changed live and in its definition and keeps its maximum;
a paused guest takes the new size once resumed. The quota check follows
the same rule. Checked on a real guest: paused, shrunk, resumed, then
grown back.
The .vv console file is built from the VM found by its UUID, with the
console page's getters, instead of looking the VM up by name five times;
a console without a listen address or password now gives 127.0.0.1 and
an empty password instead of "None". Removed methods nothing calls: the
by-name getters and power methods of wvmInstances, managed save, the vCPU
mode getter (it read the wrong attribute), the old NIC editor and its
bridge lookup. CD-ROM media and NIC hotplug use _definitions() like the
other device edits, the disk editor takes the bus prefix from
util.vol_dev_type, and an unreadable disk is logged, not printed.
Cloning a UEFI VM defined a persistent storage pool on the NVRAM's
directory when no pool held it, and dropped <nvram> from the clone,
which found its copy only because libvirt chose the same file name.
The copy is now made through the NVRAM directory's pool (a transient one
when there is none) or the pool that holds the file, with the original's
format, owner and mode, and the clone's definition names it. A failed
definition removes it with the copied disks. An NVRAM that is not a file
is refused before anything is copied.
A volume of an LVM or iSCSI pool can be attached: by its own path, with
the format its volume reports (raw when the pool knows none). Detaching
a disk takes each definition's own description of it, so a disk only in
the persistent definition goes too. Every NIC type can be a boot device;
bridge and direct NICs were dropped from the boot order. The video model
is a value, never parsed as XML. QoS changes clear the page's read cache.
The NIC model "default" leaves the choice to libvirt instead of writing
model "default", and rt18139 is rtl8139: a VM with either did not start
(checked on a RHEL 10 host). A migration corrects the default NIC type
setting.
A NIC without QoS showed the previous NIC's limits, an empty CD-ROM the
previous drive's pool; a NIC without a source or model broke the page.
The boot order shows the target of block and network disks and of bridge
and direct NICs too. Refreshing the VM's pools before a snapshot or a
delete no longer fails on a disk outside every pool.
The overview read the host's info four times, now once (12 libvirt calls).
A connection object keeps it next to the capabilities, and the VM page and
VM stats take it from there; the VM page's cached_reads block holds only
the VM's own data. Free memory is still read each time.
Measured on the create page: 751 ms and 72 libvirt calls, now 123 ms and
13. Its getters parsed the host's capabilities some 60 times, each read
from libvirt. A connection object (one per request) now reads them once,
and the domain capabilities once per emulator, arch, machine and domain
type. The VM page's cached_reads block no longer needs its own copy.
Each stats answer carries a signed sample of the counters, bound to the VM,
its compute and UUID. The next poll sends it back and the rates are the
averages since then; a sample from another run of the VM (domain ID),
another host CPU count or device list, or older than 30 s falls back to
one second as before. Windows are measured on the monotonic clock.
Now that a poll is cheap, the Stats tab polls every 5 s like the host
graph, the first time as soon as it opens (pollWhileVisible takes the
delay of the first request).
A counter the host reports as unsupported (-1) or that went back gives no
value instead of 0 or a jump; a NIC without a target keeps its chart, a
pool volume disk is measured, and a disk or NIC attached after the page
loaded no longer stops the other charts. Disk rates are MB/s (they were
MiB/s labelled Mb/s), network Mbit/s; the CPU tooltip names the window.
Measured on a VM with two disks, a CD-ROM and a NIC: 70 libvirt calls per
page, now 43. Within the page's cached_reads block the VM's state, the
host's info and domain capabilities, and each disk path's volume are read
once; the running and the persistent disk lists share a path's volume.
The ISO list takes the active pools in one call instead of looking each up
by name; it still refreshes them, so an image copied in by hand shows up.
The opt-in DRBD status waits at most 10 s for the compute's ssh, then
leaves the status unknown instead of holding the page.
The ARP lookup passed 3 as the address source; libvirt's
VIR_DOMAIN_INTERFACE_ADDRESSES_SRC_ARP is 2 and libvirt refused 3 as an
unknown source. The error was suppressed, so a VM without a guest agent
never showed an address.
libvirt keeps a VM's UEFI variables (NVRAM) and emulated TPM state in
files of its host. A live migration carries both: on the test hosts
(libvirt 11.10, QEMU 10.1) the destination got the same NVRAM content
once the VM ran and the same TPM state, and the source TPM state was
removed. Moving a shut-off VM moves only its definition, so neither
went along: on its first start on the destination libvirt made a new
NVRAM from the template and a new TPM. Boot entries and enrolled keys
are lost, and disks sealed to the TPM no longer unlock.
A shut-off VM whose definition has an NVRAM file or an emulated TPM is
now refused before anything changes: "Start the VM and migrate it
live." A passthrough or external TPM keeps no state on the host and
still moves. The check sits in moveto, not in the define/undefine
workaround, as libvirt's own offline migration does not carry the files
either. The keep-NVRAM and keep-TPM undefine flags of that workaround
are gone, as no VM that needs them reaches it.
Checked on the test hosts: a shut-off UEFI VM with a TPM is refused and
stays on the source with nothing defined on the destination; the same
VM migrates live. The new tests fail on the previous code.
A VM that changes its memory faster than the network carries it never
finishes a live migration without auto converge: measured with a guest
rewriting about 1.2 GB/s over an 8.5 Gbit/s link. The migration then
ran until gunicorn ended the request after 600 s. libvirt cancelled it
when the request went away and the VM stayed on the source, but the
user got an error page and nothing was logged.
The app now cancels a migration that has not finished after 540 s
(MIGRATION_TIME_LIMIT, below the 600 s of gunicorn.conf.py; a test
keeps it below). The failure is reported and logged like any other:
the migration was cancelled, the VM stays on the source host, and a
VM like that needs auto converge. A timer that fires as the migration
ends finishes before the compute locks are released, and an error
the cancel did not cause keeps its own message.
Migration needs the VM's disks on storage both hosts reach, and nothing
checked that. A VM with a disk the destination did not have failed
during a live migration; a shut-off VM's definition moved and could not
start there.
moveto now refuses the migration before anything changes, naming each
disk (target and source) that is not a volume of the destination's
active storage pools. Checked: the disks, CD-ROM images and block
devices of the live and the persistent definition, pool volumes, and
backing files, as the definition or the destination's volume lists
them (a shut-off VM's definition does not). Pools are refreshed before
a disk counts as missing, so a file another host made on shared
storage is found. Network disks and network backing images are
reached the same way from every host and are not checked; disk types
that cannot be checked (dir, nvme, ...) are refused.
This checks that the destination knows a volume, not that it is the
same file: shared storage should be a storage pool on each host. A
shared file outside any pool, or a block device no pool covers, is
refused.
moveto read the VM state for each flag and dropped options that did
not fit it, so a request could run as something else than asked; the
log then named the mode from the form ("live" or "offline"), not the
one used. The API did not log at all and answered "instance migrate is
started" after the migration had finished.
- moveto reads the state once. A running VM migrates live, or without
live paused for the transfer (non-live); a paused VM stays paused; a
shut-off VM moves offline. Live for a shut-off VM, offline for a
running or paused one, and any other state are refused before
anything changes. unsafe, auto converge (live only) and compressed
apply to a running VM.
- migrate_instance logs the outcome for the VM page and the API: the
mode and options used, or the error. The API answers
"instance is migrated (<mode>)" and accepts requests that leave out
an optional option.
- Post copy is gone from the form, the view and the API: only the
flag was set, and without migrateStartPostCopy, which the app never
called, the migration never switched to post copy.
libvirt's offline migration fails on some builds with "operation
failed: domain is no longer running" (RHEL-156800: Rocky/RHEL 9.8
libvirt 11.10.0-12 has the change that broke it, not the fix, which is
upstream in 12.2.0). A shut-off VM could not be migrated at all there.
A shut-off VM is only its definition, as the disks are on storage both
hosts share, so moveto now moves that itself:
- refused first when the VM has snapshots or a saved state, or the
destination already has a VM with its UUID or name;
- defined on the destination from the source's migratable XML, then
undefined on the source; NVRAM and TPM state stay on the source;
- after a failure the VM stays defined on one host: the destination
definition is removed only while the source still has its own, and
a VM another client started on the source meanwhile gets its
definition and autostart back there.
The destination's qemu migration hook does not run for it. Running and
paused VMs still migrate through libvirt.
Migration is client-driven: libvirt sends the VM's memory to the
destination's own host name unless a migration URI is given, and the
app never gave one. Where the hosts cannot resolve each other's names
(or the names point at a management network) every migration failed
with "Unable to resolve address '<host name>' service '49152'".
A compute now has an optional migration address (an IP address or a
host name, no scheme or port), editable in the compute form and the
API. A VM migrating to that compute is sent to tcp://<address>
(bracketed for IPv6). Empty, the default for existing computes, keeps
the previous behavior. The compute's host name is not reused, because
for SSH computes it can be an ssh_config alias.
The migration log entry names the address when one is set.
Migration left the VM defined on the source host unless "Delete
original" was ticked (VIR_MIGRATE_UNDEFINE_SOURCE). With the shared
storage migration needs, that definition points at the disks the
destination VM is running on: the VM showed on both hosts and could be
started (or autostarted) a second time against the same disks; only a
configured lock manager stops that.
moveto now always passes VIR_MIGRATE_UNDEFINE_SOURCE with
VIR_MIGRATE_PERSIST_DEST. The option is gone from the form, the view,
the API (MigrateSerializer.xml_del) and the clone auto-migration. It
removes the definition, not the disk files.
The VM page's getters read the domain XML on their own: one load made
86 libvirt calls for a real VM, 29 of them for the same three XML
documents (live, live with secrets, persistent) and 6 for the host
capabilities. Each call is a round trip to the host.
wvmInstance.cached_reads() is a block in which _XMLDesc (per flags
value) and get_cap_xml are read from libvirt once; it ends with the
block, even on an error, does not nest, and _defineXML drops it.
Outside a block every read goes to libvirt as before. Only the VM page
view, which does not change the VM, uses it, around its rendering.
Views that change the VM keep reading fresh XML: most changes (devices,
QoS, memory) do not go through _defineXML, and several views read,
change and define more than once in one request.
Measured on the test host: 86 -> 55 calls (XMLDesc 29 -> 3,
getCapabilities 6 -> 1) for one VM, 82 -> 54 for another; the page took
0.23-0.25 s before, 0.14-0.15 s after. The HTML of both pages is
identical before and after (CSRF tokens aside). A test renders the page
on libvirt's test driver through the real getters: 25 XML reads before,
one per kind now.
Each open VM page asks for its statistics every 10 seconds. For a
running VM the stats view measured the CPU, then each disk, then each
NIC, sleeping one second for each. With two disks and a NIC a request
held a sync gunicorn worker for over 4 s, so a few dozen open VM pages
could keep every worker busy.
wvmInstance.usage() replaces cpu_usage, disk_usage and net_usage, the
stats view's only callers: it reads all counters, sleeps one second
once and reads them again. Values and units are unchanged (CPU percent
of the host, disk bytes, network bits per second); compared against the
previous methods on the same counters, the results are identical.
Measured on the test host for a running VM with two disks and a NIC:
the stats request took 4.1-4.3 s before, 1.1-1.3 s now. The new test
fails on the previous code.
A VM owner who may manage the console (is_change and is_vnc) could set
its listen address to all interfaces and remove its password: a VNC
console reachable from the network without a password. Checked on a
test host: the owner's form changed the VM's listen address to 0.0.0.0
and then removed the password.
- The listen address field is disabled for users who are not
superusers; a disabled field ignores posted values. They can still
set or change the password and the keymap.
- A console whose persistent definition does not bind it to loopback
(or a Unix socket) only loses its password through an administrator.
A binding the XML does not pin down counts as not local: listen type
network, or no address (the host's default in qemu.conf).
On the test host, the owner's form now leaves the listen address alone
and is refused when it removes the password of a console listening on
0.0.0.0; with the console on localhost the password can be removed.
The new tests fail on the previous code.
attach_disk (new or existing volume, CD-ROM), add_guest_agent and
remove_guest_agent acted only when the status was running (1) or shut
off (5). On a paused VM they changed neither definition and the request
reported success. They now use the same rule as the NIC edits: the
running definition while the VM is active, paused included, and the
persistent one always. remove_guest_agent reads each definition from its
own XML, so a channel present only in the persistent one is removed too.
Unit tests check paused, running and stopped VMs (paused fails on the
previous code). The live effects test adds attach_disk and
add_guest_agent, and now also checks that edits meant for the running VM
reach it, which showed that an ejected CD-ROM keeps an empty <source>
in the live XML; NIC unplug is left to the guest tests, as these VMs
have no OS to acknowledge it.
Comments, docstrings and live test names cited item IDs of a roadmap
that is not part of the repository. They now say what is meant, or
nothing where the sentence stands without them. Test assertions and
behavior are unchanged.
- change_network detached the NIC and attached one built from scratch,
dropping its QoS, link state and other settings. It now edits the NIC
of the persistent definition in place and defines once; a running VM
gets the change at its next start, as the UI already says.
- add_network, delete_network and set_link_state did nothing on a
paused VM, and set_link_state rewrote the persistent NIC from the live
XML. They now update the running definition when the VM is active and
the persistent one always, each from its own XML.
- Boot and console edits used xml.etree, which renamed the namespace
prefixes of other tools' metadata (<app:info> became <ns0:info>); they
use lxml now.
- The NIC table and the clone form are filled from the persistent
definition (Instance.config_networks, config_disks), so a pending NIC
change is shown and can be edited again.
A live test runs 18 domain edits on stopped, running and paused VMs
whose definition has elements WebVirtCloud does not manage, and checks
for each the effect, that nothing else in the persistent definition
changed, and that persistent-only edits leave the running VM alone.
