Repository navigation
Known vulnerabilities per device and VM, built-in Discovery (package inventory and service scans), device dependencies with root-cause alerting, a Monitoring section of its own, a prefix tree and free space in the IP lists, one action ribbon on every page, styled floor plans, faster vulnerability lists, stricter API parameters, and a round of hardening.
Important
API clients: unknown list parameters are now refused with a 400 that names them, and nested list errors are keyed by position. Check scripts that call the API before upgrading; see Behaviour changes.
Note
Discovery has only been run against a few real hosts so far. Package inventory over SSH on Linux and the SSH terminal on VMs have been used for real. Windows inventory, the service scan, the nmap engine and Discovery on an Outpost are tested on recorded output and local test servers only. The new Outpost installer and the unit sandboxing change how services are installed; try them on a test host first.
New
- #450 - Device dependencies.
- Recording them: say which device depends on which, with a link medium (Ethernet, Fibre, Radio, VPN, LTE/5G or Satellite; the list can be edited) and optional redundancy groups.
- Edit them under DCIM → Dependencies or on a device's Dependencies tab, or bring them in by import.
- A dependency that would close a loop is refused, and the error names the loop.
- What if this fails? On a device it lists the outage and the reduced redundancy that would follow, and can start a maintenance event from that list.
- Root-cause alerting: when an upstream device goes down, the alerts of the devices behind it are explained by it.
- You get one root-cause message rather than one alert per device.
- The root device's own alert still goes out at once.
- Recovery sends one message, with how many devices came back.
- The new Root causes tab on Alerts lists them.
- Settings → Monitoring has the switch, the hold time and the hold mode.
- It fails open: anything uncertain alerts as before.
- Maps: the Topology Diagram and the Site map have a Dependencies layer, and What if this fails? is on a device card's right-click menu.
- Suggest: proposes dependencies from cables (through patch panels) and LLDP, starting from a core device. You confirm each one.
- Recording them: say which device depends on which, with a link medium (Ethernet, Fibre, Radio, VPN, LTE/5G or Satellite; the list can be edited) and optional redundancy groups.
- #507 - Action ribbon. Page headers order their buttons the same way everywhere:
- The order is page actions, Export, More, then Add, Edit and Delete at the right end.
- Outputs (spec sheets, reports, labels, Import / Export) sit in Export, and rarer verbs in More.
- Search moved from the header into the table's toolbar row.
- #517 - Group clusters on the Topology Diagram. It shows a cluster's hosts as one card, with cable counts, and opens it into a frame.
- #514 - VM power state is coloured: running green, suspended amber, stopped plain.
- #451 - Vulnerabilities. Known CVEs per device, VM and installed software, with CVSS, EPSS and the CISA known-exploited list. Map a platform, application or device type to its public product, or let Suggest mappings propose them. Data comes from NVD and OSV by server fetch, from the browser, or by upload. Exceptions with an owner and an expiry, a tenant priority policy, notifications, a Known vulnerabilities compliance check, a CycloneDX VEX export, a count on the Vulnerabilities tab and a score badge in the device and VM lists.
- #451 - Discovery. Under Monitoring → Discovery, off until a deployment admin and a tenant admin both turn it on, and every profile records who authorised it.
- Package inventory reads the OS release and installed packages over SSH with a fixed command list, on Debian, Ubuntu, RHEL-likes, Alpine and openSUSE Leap, and matches them against the distribution's own advisories for that release. Windows hosts are read over SSH first, or WinRM over HTTPS with a client certificate, and matched by OS build against Microsoft's advisory data.
- Service scan is a TCP connect scan of common ports that reads banners, TLS certificates and SSH keys. What it finds goes to Findings, where each proposal is accepted or ignored; a banner never becomes a CVE match by itself. An nmap the admin installed can be used as the engine.
- Shared credentials scoped by role, device type, platform, site and tag, SSH host keys reviewed and accepted in bulk, Do not scan on devices, VMs, IP addresses and prefixes, deny subnets, and profiles that run on an Outpost (Outpost 0.10.0).
- A read sets the host's running version, proposes its platform, and links packages to applications as installed software.
- Monitoring has its own sidebar section: Status, Checks, Alerts, SLAs, Performance, Discovery, and Policies, Settings and Engines under Setup. The Overview leads with what is down. Old
/monitoring?view=…links redirect. - Prefix tree. The prefix and IP address lists, a prefix's IPs and child prefixes tabs, and a VRF's prefixes and IPs tabs get a Filters / Tree rail. The tree loads a branch at a time, searches inside a prefix, and on a prefix's page shows its branch with its siblings; a click opens that prefix.
GET /api/prefixes/tree/, andwithin=<prefix id>on the prefix and IP lists. - Hardware section in the device report, a tick that adds the spec sheet's boxes, faceplates and part tables.
- Free space in the prefix list, Next available,
POST /api/prefixes/<id>/allocate/and a subnet planner. - #492 - Self-service password change, and an optional reset by email.
- #493 - Key rotation for
MONITORING_SECRET_KEYwith a key list andmanage.py rotate_monitoring_key. - #465 - Template versions with diff and restore, a live syntax check, and a warning for unknown CLI lines.
- #467 - Vector basemap for the site map with a choice of label language.
- Connect menu and SSH terminal on VM pages, with a separate
connectpermission for virtual machines. - Tables: pin columns from the Columns menu, share layout presets with everyone, groups or users, and the sideways scrollbar stays at the bottom of the view.
- Floor plans:
- Area names take a position (a 3×3 grid), a size, a colour and a backplate, as a plan default in Display and per area in its inspector. Names keep inside the walls along the area's edges.
- Tiles on a free placement plan turn by degrees: the turn handle, a Rotation field, and R for 90°.
- The tile popover fits the screen, and its faceplate opens large.
- A room's page has Show on floor plan.
- The dashboard's Floor plan widget draws the plan as its page does: walls, areas, labels, cable routes and the plan's Display settings, fitted to the widget.
- In 3D, power, cooling and other tiles can be selected, and rows of racks are named on the floor.
- Tiles show an icon of what they are, the 2D plan draws rooms, then rows, then racks by zoom, click a room to zoom to it, filters and saved views. The cable schedule draws bundles with counts, has trace pages for chosen cables and tighter tables.
- VM drift is shown on any VM with open drift, with Use hypervisor's and Keep Danbyte's.
- ACME issuers take a CA certificate for a private CA and trust the system store.
Fixed
- #520 - Fixes the Vulnerabilities pages slowing down with the number of open matches.
- On a tenant with 628,800 open matches, lists and filters drop from 1-4 s to well under 0.1 s, and under 0.25 s for users limited to some sites.
- The page size no longer changes the cost.
- #521 - Fixes a vulnerability recompute rewriting every open match. A recompute that finds no change now writes almost nothing.
- #509 - Fixes a deleted credential, PSK, keychain, CSR or ACME key leaving its secret in the secret store. That covers single and bulk deletes, and deleting the device that owns it.
- #510 - Fixes every Azure Key Vault call fetching a new Entra ID token.
- #511 - Fixes switching the secret store making every existing secret unreadable. Each secret is now read from the store it was written to, and the settings card shows how many secrets each store holds and asks before switching.
- #512 - Fixes the Azure Key Vault URL and sign-in endpoint accepting
http://. HashiCorp Vault addresses are unchanged. - #513 - Fixes adding a device bay, module bay or inventory item template answering 500.
- #515 - Fixes a cluster's Hosts tab listing every device.
- #516 - Fixes the VM groups tab on a cluster having no count.
- #522 - Fixes the two unlabeled badges on a vulnerability record. They now read Priority and CVSS.
- Fixes the space map showing prefixes from sites you can't see as free space.
- Fixes users limited to some sites seeing Create under parents they can't add to.
- Fixes a manual prefix create, a CIDR or VRF change, or an import racing an allocation into overlapping prefixes.
- Fixes the device list ignoring
?cluster=, and the Fibre cables page listing every cable. - Fixes uploaded 3D models compressed with
gltfpack -cbeing refused. - #501 - Fixes the upgrade deleting prefixes with a legacy CIDR and the IP addresses in them. Also released as 0.19.2 and 0.19.3.
- #504 - Fixes the prefix renumber preview showing ARP entries, BMC addresses and other references on sites the user may not view. Hidden references are now one count.
- #503 - Fixes a long unknown list parameter turning an API call into the "Danbyte is updating" page.
- #505 - Fixes the compliance condition Field showing "Na" for "Name" on 320 px phones.
- #496 - Fixes host names and addresses committed in the development unit.
- Fixes ACME account registration answering 500 when the directory's certificate is not trusted, and a failed order showing an empty error. HTTP-01 challenges are answered for any host name.
- Fixes a VM or device page crashing when opened from its list.
- Fixes the reverse DNS nameservers field refusing a comma.
- Fixes a check template of a kind an integration or plugin registers, such as
zabbix, being refused by the API. - Fixes a restore running out of shared memory on a stock PostgreSQL. The database is restored beside the live objects and checked before they are dropped; a failed restore puts them back.
- Fixes a failed restore, or turning a unit off in
danbyte-admin, deleting the unit's link, so the next upgrade enabled it again. - Fixes the health check showing a migrate or restore by another process up to a minute late.
- Fixes the site map, floor plans and device and VM pages failing in the server render.
- Fixes the spec sheets listing interfaces to readers without interface view.
- Fixes the IP tabs of devices, interfaces and prefixes running one more query per address.
- Fixes a text label's backplate ending short of long capitalised text.
- Fixes OIDC sign-in answering 500 on the callback of a bundle install: the ID token check needed a library the bundle didn't ship. It now uses PyJWT, which every install already has.
- Fixes 3D models compressed with meshopt failing to load under the Content Security Policy.
Hardening
- dnspython 2.9.2 for GHSA-vvf4-9x63-8xvv (excessive memory and CPU use).
- #498 - Content Security Policy: every page gets a per-request nonce, and scripts without it are refused.
DANBYTE_CSP_MODE=reportreports instead,offturns it off. The maintenance page gets a no-script policy. - #508 - OIDC sign-in sends PKCE (S256), so an identity provider client that requires PKCE works. ID tokens must be signed with an asymmetric algorithm (RS, PS, ES or EdDSA).
- #497 - TLS: the nginx template pins TLS 1.2 and 1.3 with forward-secret AEAD suites.
- #499 - Rate limits at nginx for sign-in, MFA, password reset and the CSP report sink.
- #502 - Redis can take a password, an ACL user, TLS or a unix socket, for the cache, the job queue and websockets. The installer has an opt-in
--redis-password. - #500 - Unit sandboxing: the app units get no-new-privileges, system-call and address-family filters, and filesystem confinement where the host supports it.
DANBYTE_UNIT_HARDENING=offopts out. - Python dependencies install from a hashed lock, and CI gates on known-vulnerable dependencies.
- New Outpost installs run as the
danbyte-outpostaccount, not root. - Inlet and rear port template references are checked against the active tenant.
Behaviour changes
- Discovery skips an SSH credential with no user name; it no longer logs in as
root. - An install whose
REDIS_URLandREDIS_HOST/REDIS_PORTname different servers refuses to start. - After upgrade lists the nginx lines to add on a hand-edited site (TLS, rate limits) and, where it applies, a check for prefixes the 0.19 upgrade deleted.
- Unknown list parameters are refused (
API_STRICT_QUERY_PARAMSnow defaults to on).- The 400 names each unknown parameter, suggests the likely intended one and lists the allowed ones.
?strict=0ignores unknown parameters for one request;API_STRICT_QUERY_PARAMS=0in.envturns the check off for the whole install.
- Nested list errors are keyed by position:
{"steps": {"1": {...}}}instead of a list (LIST_SERIALIZER_ERRORS_AS_DICT). - A vulnerability match's
last_seen_at: for an open match it is when its host was last evaluated, not a write on every recompute. - Root-cause alerting is on by default. It does nothing until dependencies are recorded.
Upgrading
- A hand-edited nginx site needs the maintenance page's policy line; After upgrade shows it.
- From a Git checkout, fetch tags with
git fetch --tags --force. - A custom worker unit must also listen on the
discoveryqueue. - Outposts need agent 0.10.0 to run discovery work; older agents keep running checks.
- The vulnerabilities migration backfills each match's rank in batches (about 30 s per 700,000 matches).
- Old secrets left behind before #509: run
manage.py cleanup_secret_storeto list them, then add--deleteto remove them.- For Azure Key Vault this needs the List permission, and for HashiCorp Vault list on
metadata/.
- For Azure Key Vault this needs the List permission, and for HashiCorp Vault list on
- Moving secrets between stores:
manage.py migrate_secret_store --from X --to Y. It is a dry run by default, and--applymoves them.