Container Vulnerability Lesson Learned
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
2 hours ago
When a new CVE drops, the first question security gets is not
"How many?"
It's "Which service, and who owns it?"
For containers, that question is unusually hard to answer.
Scanners only know images.
To get from an image to an owner, you have to walk the CMDB:
Vulnerability → image as the scanner saw it → CMDB image CI
→ running container → pod · namespace → cluster
→ application service → owner
While diagnosing a ServiceNow container vulnerability integration, I found four places where this chain breaks — and how to fix each one.
① No image tags
Problem: Without tags, no relationship links the image to its registry repository. In this environment most images arrived untagged, leaving repository CIs isolated.
Fix: Include tags in the scanner query. If images genuinely have none, build reports on the repository reference VR already stores instead of CMDB relationships.
② One image, two CIs
Problem: The scanner integration and ITOM Discovery write to the same CI classes. A slightly different image ID format splits one image into two CIs. Vulnerabilities attach to one, runtime data to the other.
Fix: Align the image ID format (digest length) across both sources and standardize IRE identification on the digest. Keep a recurring report of "CIs with the same digest from more than one source".
③ Silent failures
Problem: There was a code path where the CMDB engine (IRE) failed without writing a single log line. Vulnerabilities arrived, but nobody could explain why they had no CI.
Fix: Treat the per-run "ignored CI" count as an operational KPI and alert when it isn't zero. For root cause, replay the same payload with an IRE dry run (identifyCIEnhanced) — and keep that script in the runbook.
④ No label standard
Problem: Service mapping groups workloads by labels or namespaces. Without a rule, services are empty — and so are owners.
Fix: Make labels like app and env mandatory in your deployment standard and enforce them at deploy time. If that's not realistic yet, start with namespace = service.
And if Discovery isn't in place yet, keep a fallback path open:
registry → cloud account → organization.
At minimum, you can always answer "which org owns this?"
My first-week checks on every project:
• Same image = same CI? (zero duplicates across sources)
• Trace one sample image all the way to its service
• Is there a fallback path to an owner?
Vulnerability management is not a scanner project.
It's a CMDB connectivity project.
How many clicks does it take your team to get from one CVE to its owner?
#CMDB #CyberSecurity #VulnerabilityManagement #ContainerSecurity #ServiceNow
- Labels:
-
Architect