Discovery: troubleshooting¶
Every status a run or a host can end in, what it means and what fixes it, then the situations people run into most. The messages in italics are the ones the run page shows.
Where to find it
- A run's status, summary line and Messages: the run page (from Recent runs on Monitoring → Discovery → Profiles, or a profile's Runs tab).
- Each host's result and message: the run's Hosts (or Addresses) tab; hover Transport for every login tried.
- Before a run: the profile's Hosts / Addresses tab and the reason under the profile's name.
- Background jobs and the Package inventory timer: Jobs.
Run statuses¶
| Status | Meaning | What to do |
|---|---|---|
| Queued | Waiting for a worker. | If it stays queued, no worker listens on the discovery queue (Settings). |
| Running | Hosts are being read. A run on Outposts stays running until each Outpost reported or its work expired; the run page counts the hosts still out. | Wait. A run that has not finished after 6 hours is given up as Failed. |
| OK | Every host was read, or left alone on purpose (exempt, deny subnet, maintenance). | - |
| Partial | Some hosts were read, some not. | Look at the host rows that are not OK. |
| Nothing read | No host was logged in to, because each was left alone before a login - no credential, no accepted host key, no address, a blocked target. Nothing failed: there was nothing to read with. The summary line says why by reason (61 without a credential) and links to Shared credentials or Review host keys. | Give the hosts a login, accept their keys (below). |
| Failed | Logins or scans were tried and none succeeded, or the run could not start: The background workers are not reachable., The run did not finish; given up., The port list is not valid: …, an nmap refusal. | Read Messages and the host rows. |
| Skipped | The run did not go ahead, because something changed after it was queued: Discovery is off., The profile is disabled., Record who authorised this profile before it runs., Outside the profile's window., The person who started it no longer has an account., The person who started it may no longer run this profile. A run on Outposts is also Skipped when no host could be handed to an Outpost. | Fix the reason and run again. |
Run now itself is refused, with the reason under the profile's name and in a toast, when the profile is disabled, not authorised, outside its window, already running, set to An Outpost without one, or Discovery is off.
Host statuses¶
| Result | Meaning | What to do |
|---|---|---|
| Ready | (Hosts tab only) Would be read on the next run. | - |
| OK | Read. The row says how many packages, and how many were added, removed or changed; a service scan, how many ports were open. | - |
| Login refused | The host refused the credential (The host refused the credential., The host refused the client certificate.), or Danbyte could not use it (The stored SSH key could not be loaded., The credential holds no password or key.). | Check the user name, the key on the host, the passphrase. A refused password stops there unless the shared credential has Try the next if refused. |
| Host key not accepted | No accepted SSH host key. - nobody accepted a key for this host yet; or The host key does not match an accepted key. - it presents another key than the accepted one. The presented key was recorded. | Review and accept it under Host keys. A key that differs from an accepted one: find out why (reinstall, rotation, interception) before accepting it on the host's SSH host keys panel. |
| Certificate not trusted | WinRM: The host's TLS certificate is not trusted for its name. | Below. |
| Unreachable | The connection failed (the row names the error), or WinRM over HTTPS is not reachable. | Check routing, the firewall, the port on the credential, and whether an Outpost should read it. |
| Timed out | No answer within 60 s. - the host did not finish in the per-host timeout. On an Outpost: The Outpost did not report back in time. | Raise Per-host timeout; lower Hosts at once on a busy network. |
| Unsupported | No package reader for … (a distribution without dpkg, rpm or apk), The host has no readable os-release. (and not Windows), The host's os-release names no ID., or WinRM cannot run here. | See supported distributions. |
| Failed | The read started and could not finish: The dpkg package list could not be read., The host's answer is larger than 16 MB., PowerShell could not read the inventory., The host's answer is not JSON., Its packages could not be stored., WinRM refused the request., The host answered with a redirect.; on an Outpost The Outpost's answer was refused: … or Not run: …; in a scan Not scanned. or Not in the nmap output. | Read the message. For PowerShell, run the script as the same account. |
| No credential | No login for this host on a transport that is on., The credential has no user name. (an SSH credential saved without one: add it), or The credential's secret cannot be read. | Below. |
| No address | No primary address. (and no out-of-band address on a device), or the address is not valid. | Give the host a primary address. |
| Exempt | The host, its address or a prefix around it is marked Do not scan. | Intended. Clear Do not scan to include it. |
| Deny subnet | The address is in a discovery or monitoring deny subnet. | Intended. Edit the deny subnets. |
| In maintenance | The host is in an open maintenance event. | Intended. It is read on the next run after the event. |
| Blocked target | Link-local (cloud metadata) addresses are never discovered., or The server's target policy refuses this address. - internal while DANBYTE_CHECK_BLOCK_INTERNAL is on. |
A link-local address is never read: give the host another address. For an internal one, add the range to DANBYTE_SSRF_ALLOWLIST, or run the profile on an Outpost. |
| No Outpost | The host is bound to no Outpost, or the named one was deleted. | On an Outpost |
| Outpost offline | The Outpost is disabled, not seen recently, or did not pick the work up within 30 minutes. | Bring it back; run again. |
| Outpost cannot run it | The agent is too old, the Outpost is driven over SSH, or it has no nmap it may run. | Which agent version |
Common situations¶
Nothing read¶
Nobody was logged in to, and nothing failed. On a first run this is normal: the hosts have no accepted host keys yet, and the run read the keys they present. Open Host keys, check the fingerprints, Accept N keys, run again (Getting started). If the summary says without a credential, see the next section.
No credential¶
The host has no login Danbyte may use:
- no device credential of kind SSH key, SSH password or WinRM certificate on it, and no shared credential whose scope matches it (an empty scope matches nothing unless All hosts is on; a scope naming device types never matches a VM);
- the matching credential's transport is switched off (hover Transport: Switched off), or it is a WinRM certificate on an Outpost;
- Run now by someone who may not view that credential: a scheduled run would use it;
- the secret store cannot read the secret (The credential's secret cannot be read.).
A host left as No credential still has its SSH key read, so it shows under Host keys for later.
Host key not accepted¶
See the host status above. When every host of a run says it, the run is Nothing read. Accepting on first use is possible (Accept new host keys on Run now) and not advised.
Certificate not trusted¶
The WinRM listener's certificate did not verify:
- It must name the address's DNS name in IPAM, or the IP address when the
address has no DNS name. A certificate for
srv01.example.netfails when IPAM has no DNS name on the address. - Its issuing CA must be in the Danbyte server's trust store, or in the file
DANBYTE_DISCOVERY_WINRM_CA_BUNDLEnames. Restart the web and worker units after changing it. - It must be valid now and chain completely.
Packages but no vulnerabilities¶
Work down the list:
| Check | Fix |
|---|---|
| Vulnerability data is off | A deployment admin turns it on (Known vulnerabilities). |
| The Packages tab says Advisory fetch queued (server mode) | Wait for the background fetch, or the daily refresh. Vulnerability data is updating means one is running. |
| Vulnerability data for these packages has not been fetched (browser mode) | A deployment admin presses Sync now. |
| … has not been imported (upload mode) | Import the distribution's OSV archive (Import OSV file). |
| No advisory source for … | The distribution has no OSV advisories Danbyte uses (Fedora, CentOS, Oracle Linux, SUSE Linux Enterprise, openSUSE Tumbleweed, Amazon Linux, derivatives): its packages are read and never matched (supported distributions). |
| The host is Windows | Windows CVEs come from the OS build, not packages (Windows hosts); OS build not matched says why not. |
| You see packages but not the Vulnerabilities tab's package rows | Package matches need view on installed software. |
| Matches exist but are hidden | An exception covers them; tick Excepted on the Vulnerabilities tab. |
| The packages are genuinely fixed | A Debian or Ubuntu backport is fixed even when the upstream version looks old. |
A fetch failed¶
The vulnerability data's last run (on the Vulnerability data card) ends failed with the products or batches that did not arrive; what did arrive is kept. In server mode the next start within 20 hours resumes it. Check that the server may reach OSV (and Microsoft, for Windows), or a mirror you set (what is contacted).
An Outpost says it needs a newer version¶
Outpost X runs agent 0.9.0 that cannot run package inventory; it needs 0.10.0 or later. The agent's hello does not offer discovery. Upload a newer agent under Outpost versions and roll it out; checks keep working meanwhile (On an Outpost). … has no nmap it may run (the agent runs as root) is about nmap only (nmap on an Outpost).
A platform is not proposed¶
There is no Set platform row when:
- the host is Windows (never proposed);
- the current platform already fits - its name names the OS and no other release (Debian fits Debian 13), or it is mapped to the same end-of-life product and cycle; a device without its own platform is judged by its device type's;
- the read gave no version (
VERSION_IDempty); - someone ignored it, and the OS, release and platform have not changed since (Show ignored);
- you may not view installed software, or the host.
A finding keeps coming back¶
- Ignore is remembered against the evidence. A service-scan finding comes back when the port's service name, product, version or CPE changes; a Set platform row when the OS, its release or the host's platform changes. Add application and Link packages stay hidden until someone shows them again.
- An Accept that was refused leaves the row: read the toast (… refused: application.add required.) and get the permission, or ask someone who has it.
- Add software returns while no install of an application of that exact name exists on the host - an install under another application name does not count. Rename the application or accept the row.
- Set platform returns after you put the host on a per-OS platform that is mapped to another cycle: a platform has one cycle; name platforms per release.
A profile does not run on schedule¶
Check, in order: Enabled; Authorisation recorded (Not authorised
shows on the page); Every above 0; the window; Discovery on for both
switches; the danbyte-inventory timer's last run under Jobs → Scheduled
tasks; a previous run still open (one run per profile at a time).
A switch is greyed out¶
- This deployment: you are not a deployment admin, or
DANBYTE_DISCOVERY_ENABLEDis set (Locked … by the deployment). - WinRM over HTTPS: Forbidden by the deployment or Not available in this build (hover for the reason).
- nmap: Not found - install nmap or set
NMAP_BINARY, then restart the web and worker units; Forbidden by the deployment. - The transports, nmap and deny subnets: only tenant admins change them.