When the guest agent failed, the Guest Info section stayed empty with no
word: the request's error answer was not handled. An agent without the
time zone or host name command lost the OS details it had answered too.
The section now shows the agent's error, and a missing command leaves out
only its own details.
Only external snapshots had their name checked. An internal one could be
taken with an empty name: libvirt named it, the job and its log row named
nothing, and a recovery looked for "" and marked the job failed though the
snapshot existed. Both kinds now pass the same check when queued, and the
internal form asks for a name as the external one does.
Snapshot dates showed no year, so last year's snapshot looked like this
year's in the list one reverts from; they now read 2026-10-13 23:04:00
under "Date (time zone)", as on the Jobs page.
The saved console form had no initial values, so every posted field
counted as changed: saving only a password also logged the keymap and the
listen address again, each time. A listen address the form does not list
(set in the XML editor) was not a choice, so the select showed localhost
and a save moved the console there. The form now takes its initial values
from the VM, on the page and on the save, and offers the VM's own address.
Only administrators suspend and resume, and the page tells an owner of a
paused VM "Administrator blocked your instance", but the power views let
the owner force it off or cycle it, on the page and the API, and start it
again: the block ended. The power permission now refuses a paused VM to
everyone but administrators.
Disk sizes were pre-filled rounded to whole GiB, so a 3.5 GiB disk showed
4 and was grown, for good, whenever another disk was resized; an empty
field or a disk outside a pool was a 500. A disk's field is now empty with
its size as the placeholder: an empty field leaves the disk alone, a
larger whole number grows it, anything else is refused with a message.
vCPUs change only on a shut-off VM, also on a direct request: a running
VM got them at its next start, past the quota counted against its live
count. Current vCPUs or memory above the maximum are refused instead of
failing (and dropping vCPU hot-plug) or being capped by libvirt. A running
VM's maximum memory select is disabled, since its maximum is kept, and a
custom memory value is sent only while its box is ticked.
On a superuser's page of a user the VM permissions read True/False; they
are now check marks as on the Users list. A user without VMs gets "No VMs."
instead of an empty table. A superuser can delete one of the user's SSH
keys, as with API tokens: the same view takes the user, the row names
whose key it was.
Closes#218
A tombstone kept the VM's name and UUID but not its compute, so the page
could not say where the VM had been. It now stores the compute's name (as
text: the compute may be the one deleted) and the list shows it; records
made before this have no host. Deleting a record says so, and the admin
tables use scope="col" instead of the invalid span="col".
Closes#217
Job dates showed no year and no time zone, unlike the Logs page; they now
read 2026-10-13 23:04:00 under "Created (time zone)". The host filter
matched only a whole host name and the user filter only an exact username;
both now ignore case and take the start of a host or any part of a name.
NEW_USER_DEFAULT_INSTANCES names VMs, and a name may repeat across hosts.
A name with no VM, or with one on two hosts, made creating any user a 500.
The create form now looks the names up and says which one is not exactly
one VM; the user and its default VMs are saved together.
"Lost OTP?" mails the QR code only when exactly one active user has the
address. Any user could enter another user's address in their profile and
silently switch the recovery off for both. The profile and the admin user
forms now refuse an address another user has, ignoring case; an empty
address stays allowed for everyone.
- A group sent all its fields, and the ones left as they were undid
what another administrator had saved meanwhile. Each field now sends
the value the page showed, and only a field edited since is saved.
- A database error partway left the group half saved. The group is now
saved in one transaction, written as one audit row: "Settings: change
Console Scale to True, Console Clip Viewport to True".
- A failed login writes the name it was tried with, as typed: a control
character in it made every download holding that row fail, as the
workbook cannot hold one. Workbooks now write it as \x01; the VM
report shares the code.
- A download held every matching row in memory: it now takes the newest
100000, and its About sheet says when the search should be narrowed.
- The administrator's password form took "password-based
authentication: Disabled" with "must change it at the next login":
the change form asks for the old password, which the user then never
had, so they could not get past it. That combination is refused, and
an administrator cannot disable their own password.
- A user who must change the password could still use the API with
their tokens. A token is now refused until they change it, as the
pages are.
The list pages' search (filter-table.js) read what was typed as a
regular expression: a "(" or "[" stopped it with an error. It also
needed jQuery, while the instance list had its own script for the same
search. Both are now static/js/table_filter.js: plain text in any case,
the instance list's state list and host groups where a page has them,
remembered for a day as before. A group's header row is marked as one
(data-group-header) instead of being any row without a state.
The search box filtered only the 100 rows on screen, in the browser: an
older row seemed not to exist. The page now searches on the server, in
every row's user, host, VM, VM UUID, message and IP, as plain text, and
can show the failed rows only; its pages keep the search.
Failed rows are marked red, dates carry the year and the time zone, and
a VM's rows link to the VM while it is in the panel. Excel downloads the
rows the search chooses, all pages of them, with the VM report's
workbook (text stays text).
The administrator's password form has "The user must change it at their
next login". With it, the user is sent to the change form until they
choose their own, as the generated admin password already does.
- An administrator could block, delete or demote (unset Active or
Superuser on) their own account, and with it the last one, leaving the
panel without anyone to undo it. These are now refused.
- The user form showed "No password set." for users who have one: its
password field never read the user's password. Its "Set password"
link went to ../password/, a page this panel does not have. The form
now says whether a password is set and links the page that sets it.
- "Staff status" said "can log into this admin site"; here it lets a
user change and delete the template VMs given to them, and is now
called Template editor.
- A quota of 0 counted as no limit, like -1, so a user's quota could not
be taken away. Now -1 has no limit and 0 allows none; the help says
so, and the labels give the units: memory in MiB, disk in GiB.
- A user's page for a user that does not exist answered with a server
error instead of not found.
- The user list said "None" for a user who never logged in; it says
"never", and dates fit their column.
- The password form's labels no longer break mid-word.
The descriptions were tooltips, many only repeating the setting's name
("Cpu modes", "Debug for user quotas"). Now shown under each field, they
say what the setting changes: what a new VM or disk gets, how the
console starts, what a page shows. A migration writes them.
The page listed 37 settings by name, each its own form saved as soon as
its list or field changed, with the description only in a tooltip. It
now shows them in groups (New VM defaults, Disks, Clones, Console, VM
access and owners, Snapshots and jobs, Login, Lists and pages) with each
description under its field and the label tied to it. A group's Save
button saves its changed settings, or none when one is refused, and
says which were saved; number settings are number fields with their
range. Language and theme are under Appearance with their own buttons.
A setting in no group shows under Other.
The Settings page saved whatever was posted, and some lists offered
values the panel cannot use:
- Quota Debug could not be turned off: "False" is a non-empty text and
counted as on.
- VM Clone Name Prefix showed as a True/False list, so no prefix could
be typed, and touching it set the prefix to True.
- Disk SCSI Controller listed " lsilogic" with a space, never shown as
chosen, and virtio-blk, which is no SCSI controller.
- VM Machine Type listed x86_64, an architecture; VM CD-ROM Device
listed virtio, a bus a CD-ROM cannot use.
- Logs per Page and Disk Owner UID/GID are read with int(): a letter
broke the logs page and VM create.
A migration fixes the lists and moves a value they no longer offer to
the default (the clone prefix keeps its text). A listed setting now
takes one of its choices, and every number setting a whole number in
its range, checked before it is saved.
The dialog filled its texts with String.replace, which reads $& and $'
in the replacement as patterns: a host named c$&1 showed as
c%(hosts)s1. The values now go in through a function, which inserts
them as they are.
Which users can reach the API with a token was seen only by opening each
user. The user list now has an API tokens column: the number of the
user's tokens that still work (expired ones are left out; they are
deleted only at the user's next token), linked to the user's token tab.
One query counts them for all users.
Cloud images read their first-boot settings with cloud-init. The panel
served them only over HTTP, which a VM reaches only with the network
and a reverse DNS name for its address. A new VM can now carry them
itself: Cloud-init seed (Advanced) on the create form, or cloud_init in
the API.
The job writes a NoCloud seed, an ISO labelled cidata, into the disk's
directory pool as cidata-<uuid>.iso and puts it in the VM's CD-ROM
drive. Its meta-data holds the VM's UUID as instance-id and its name as
hostname, written as JSON so a VM named 123 or yes keeps a string
hostname; its user-data is what the HTTP datasource serves, with the
creating user's SSH keys. The ISO is built with pycdlib and written
through a libvirt stream, as the NVRAM copy is, now one helper for both.
A seed needs a CD-ROM bus and a directory pool, else the create is
refused. A failed create deletes the seed with the new disk, recovery
names one left behind, and deleting the VM with its disks deletes its
seed, found in the pools by name even when ejected. A clone's drive is
left empty: the seed is the source VM's identity.
When the guest agent does not answer a root password or SSH key change
in time, the command was already sent: the agent runs it when it
answers again. Measured: with the agent frozen the change failed with
"guest agent command timed out", and after it resumed, root's hash was
the one sent. The error now adds "The change may still be made when the
agent answers again." A failed synchronization sends no command and
gets no note.
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.
The user page a superuser opens from the user list has an API Tokens
tab: the user's tokens with their name, first characters, creation,
expiry and last use, and a delete button for one that may have leaked.
The token itself is never shown; only its digest is kept.
The delete reuses the owner's delete view with the user's id; only a
superuser may name another user. The confirmation and the log row say
whose token it was ("API token backup of alice"); the owner's own
confirmation now reads "Delete "API token backup"?" too. The profile
and the user page share one token table, and one script that opens a
tab named in the address (#apitokens).
Under each host's CPU and memory load, the dialog shows one line:
Given to VMs: 24 vCPUs (1.5:1), 48.0 GB memory.
Pool default: 120.0 GB free.
The line is read when the dialog opens, from computes/<id>/capacity/,
as plain text, so the VM list itself reads no VM definitions. The CPU
bars and the lines share one queue, host by host and two reads at a
time, as each read holds a worker. A host that cannot be read says so.
Compute.capacity now also gives the vCPUs per CPU and its pools with
the default one first, so the Computes page and the dialog share them.
The capacity read lived in its own module with raw libvirt calls and a
thread pool with a deadline, beside helpers that already read the same
things. It now uses them:
- wvmHostDetails.get_capacity() sums the vCPUs and memory of all and of
the running VMs from get_host_instances(), against the host's info.
- get_storages_info() also gives each pool's free space.
- Compute.capacity joins the two, like cpu_count and ram_size, and is
None when the host is down or the read fails.
The page reads the hosts one after another, as it already connected to
each of them; a host that cannot be reached is given up after the
connect timeout. computes/capacity.py and its thread handling are gone.
The page looks and sorts as before.
Each host is a row of what it has committed, read live from the host (in
parallel, by a 10 s deadline):
- Host: a status dot, the name, its address and details; when it cannot
be reached, why.
- vCPUs: the overcommit ratio, large (warning above 2:1, red above
4:1), over "N vCPUs on M CPUs" and the running VMs' share.
- Memory: memory given to all VMs against the host's memory, as one bar:
running VMs dark, shut-off VMs light (they take their memory again
when they start), and a line where the host's memory ends; red when
the VMs were given more than the host has.
- Storage: the default pool (or the first) with its used share and free
space; the others, a host may have dozens, open on a click.
A host that cannot be reached, or does not answer in time, says so in
one line instead. vCPUs and Memory sort by ratio and share. A host whose
read hangs is not read again until that read ends, so a stuck host keeps
at most one thread. The table's hidden status texts stay inside it, so
the page does not widen on a phone.
The API Tokens tab gets a "Last used" column, so an unused token can be
told from one a script still needs. The token's last use is written when
it authenticates, at most once a minute (accounts.auth.TokenAuthentication,
knox's class with that one addition; the API schema describes it as
knox's).
A user has at most ten tokens; a new one is refused until one is
deleted. Expired tokens do not count, and are deleted when a new token is
made.
knox's first migration creates its AuthToken table even when
KNOX_TOKEN_MODEL replaces it with accounts.ApiToken. The unused
knox_authtoken table references auth_user, and Django, which does not see
it, does not empty it with the others: on PostgreSQL every test that
empties the database failed ("cannot truncate a table referenced in a
foreign key constraint"). SQLite does not check this, so only CI's
PostgreSQL leg showed it. knox stays installed for its token code, with
its migrations turned off (MIGRATION_MODULES).
The REST API took only the login session, so a script could not use it
without a browser. A user now makes API tokens on the profile page (API
Tokens tab): a name, and an expiry of 30, 90 or 365 days or none. The
token is shown once, on a page that is not cached, and then kept only as
a SHA-512 digest; the list shows its name, first characters, creation
and expiry, and a token is deleted there. Tools send it as
"Authorization: Token wvc_...", and act as its user, with that user's
permissions. Making and deleting a token is logged, without the token.
A password alone (HTTP Basic) still does not reach the API, as it would
skip the OTP second factor: a token is made in a session that passed the
login and its OTP. The token class comes first, so a refused request gets
401 with its challenge instead of 403.
django-rest-knox 5.1.0 provides the tokens (accounts.ApiToken adds the
name); settings.py takes "knox", KNOX_TOKEN_MODEL and REST_KNOX from the
template.
The search's request was sent on to the login page, and the dialog said
"The search did not go through. Try again." The page now reloads, which
goes to the login form and back to the Instances page after signing in.
A paused guest cannot answer its agent, so the address search and the VM
report asked it nothing, but counted the VM neither as without an agent
nor as not read: the coverage line read as if every address had been
seen. It now counts among the agents not read, and the line names the
paused case.
The "Find IP" dialog takes a MAC address (aa:bb:..., aa-bb-... or
aabb...) as well as an IP address or a network. A MAC finds the VMs with
a network card of that MAC, shut-off ones too (from their definitions),
with that card's addresses when the VM runs: for a MAC seen on a switch
port or in an external DHCP server's log. A MAC on more than one VM is
marked as a possible conflict. Only the VMs with that MAC are read, and
the card's link-local address, which comes from the MAC, is left out.
The Administration menu gets "VM report (.xlsx)", a workbook for
superusers. The VMs sheet has one row per VM on every compute, shut-off
ones and those the panel has not registered too: name,
state, host, OS and OS family, vCPUs, memory, IP addresses and where they
came from, MAC addresses, owners, and whether the panel has registered
it. A Coverage sheet says when it was made and what could not be read.
The addresses come from the same live reading as the address search,
now shared by both. The OS is the guest agent's (libvirt's guest info)
or, without one, the libosinfo id a tool such as virt-install wrote into
the definition. Link-local IPv6 addresses are left out of the report.
Text cells stay text: a VM name starting with "=" is not a formula.
openpyxl 3.1.5 writes the workbook.
The Instances page's filters and the address search become one control:
the state list, the search box and a "Find IP" button. "Find IP" opens a
dialog (with the address typed in the search box, if any). An IP address
or a network (CIDR) finds the running VMs whose network cards use it,
read live from libvirt within 5 seconds: libvirt's DHCP leases, the host's
ARP table and the guest agent. Only addresses on the VM's own NICs count,
so interfaces inside the guest such as docker0 never match. Each match
shows the VM, its MAC, where the address came from, and a possible-
conflict mark when two VMs report one address. A coverage line says how
many running VMs were checked, how many had no connected guest agent (a
static address on them cannot be seen), which agents could not be read,
which hosts were not reached or not finished in time, and when it was
read.
The hosts are read in parallel: leases and ARP first, then each guest
agent on its own thread, so an agent that never answers holds back only
its own VM; libvirt waits for an agent without limit, so a search starts
only while its threads fit in 16 search threads in the worker.
Administrators search every compute, VMs the panel has not registered
yet included, and see the hosts; other users find only their own VMs and
see no hosts or host errors.
Found by review (Codex and agy) of the XML editor's dialog:
- A busy or moved VM made the page reload and the edit was lost. The
dialog's requests now get these answers as JSON (X-Requested-With, as the
create page's flavor forms) and show them; any other failure says the
edit is still in the editor.
- libvirt finds the VM to define by the XML's UUID: another VM's UUID
overwrote that VM past its lock, a new or missing one added a VM. The
XML must keep this VM's UUID; a malformed one is refused with the
parser's error, as on the create page.
- Save sends the XML whose diff the dialog shows, also when two previews
answer out of order; a failed save keeps the diff and Save.
- The edited XML is defined with the line breaks libvirt gave, not the
form's CRLF, and the VM's state is read before the definition, so a
failed read cannot turn a saved change into an error.
The XML editor of a running or paused VM only showed its XML. It now
shows the persistent definition, the one the VM starts with next (with
secrets, like the other edit forms), and Change saves it through the same
review dialog; the running VM keeps its live configuration until it
restarts. The tab and the saved message say so. The check against a
definition changed since the editor was opened covers this case too.
Change used to define the edited XML at once. It now opens a dialog with
the lines the edit changes in the VM's definition, and Save there defines
it. A definition libvirt refuses comes back to the dialog with the reason,
and the edit stays in the editor; before, the page reloaded with the
current XML and the edit was lost. A definition that changed since the
editor was opened (another tab, a disk added) is not overwritten: the
dialog asks to reload. An unchanged XML has nothing to save.
change_xml answers the dialog with JSON; the editor's script moves to
static/js/xml_editor.js.
sortable.js 0.8.0 (abandoned since 2014) sorted only a table's first
body: in the grouped VM list the first host's row moved among its VMs
and the other hosts never sorted, and memory sorted as text ("1024 MB"
before "512 MB"). tablesort 5.7.1 sorts each table body on its own and
keeps a row marked data-sort-method="none" in place, so hosts stay in
name order with their row on top and their VMs sort below it. Memory,
vCPUs and volume sizes sort by their numbers (data-sort). Headers are
reachable by keyboard and carry aria-sort.
base.html loads tablesort for every table marked data-sortable; the
per-page sortable includes and sortable-theme-bootstrap.css go, so these
tables take the theme's table style like the others. The grouped list's
row numbers go (meaningless once sorted). The VM owners and VM logs
tables were marked sortable but never loaded the script; they stay
unsorted.
The search box floats right of the short Volumes heading and hung
below it, so the table beside it shrank and its headers met the box.
The heading's wrapper now clears the float.
The list gets a state selector beside the search box. The search now
matches plain text in any case (a '(' no longer breaks it), and in the
grouped list a host's name shows all of its VMs while a host row shows
only while one of its VMs does. The filter moves to
static/js/instance_filter.js with a Node test; the other lists keep
filter-table.js. Collapsing a host now sets a class on its table body
instead of each row's display, so a collapsed host and the filter no
longer undo each other.
From the Codex and agy review of the VM audit log before v3.2.0:
- Resolving an uncertain migration wrote its row without the VM's UUID,
because the migrate job writes no entries of its own; the resolution now
takes the UUID from the job's intent, so the VM's Logs tab shows it.
- A console keymap or listen address the VM could not take (no graphics
device, or a console on a socket) was logged as done although nothing
changed; it now writes no row, as an edit that changes nothing.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
