Support Diagnostics
If the steps in the rest of this section don't resolve the issue, collect the following before reaching out to support. The more of this is included up front, the faster the issue can be localized.
What to include
- Your
hhfab/hhfabctlversion and a summary of your topology (gateway present, external peering, VPCs in use). - A plain description of the symptom and roughly when it started.
- Anything that changed around that time - a config push, upgrade, reboot, or node/server move - even if you're not sure it's related.
- Which specific switches or nodes are affected, if you've already narrowed it down.
- Any diagnostics you've already collected yourself. Include the raw output so we can reach our own conclusions and rule out other causes, plus your own summary or theory if you have one.
Collecting a support bundle
hhfabctl is installed on the control node as a kubectl plugin, so it's
invoked as kubectl hhfab, not as a standalone command:
This produces a single timestamped .hhs file containing cluster resources
(Fabricator, Agent, Connection, VPC, and related objects) and pod logs.
Secrets are redacted, but the bundle still reflects your deployment's real
topology and state.
Switch agent logs
For issues where BGP, BFD, or interface state won't stabilize or keeps
flapping, the support bundle's Agent CR status can look fully converged even
while the problem is ongoing - it only reflects the most recent apply, not
churn happening between applies. /var/log/agent.log on the affected
switches is the artifact that actually shows this. It's a plain file, not a
kubectl or sonic-cli command:
Include the whole file, or at least the span covering when the issue was observed, rather than a filtered excerpt.
Switch-level diagnostics
For issues that look like a dataplane or hardware problem rather than a
control-plane one (for example, traffic not forwarding despite correct BGP/EVPN
state), support may also ask for a show techsupport capture from specific
switches. This is a per-device SONiC command - if asked, run it only on the
switches identified as relevant rather than the whole fabric, both to keep
the capture a manageable size and because a comparison against a switch that
is behaving correctly is often the most useful artifact.
Note
Support bundles and switch dumps reflect your real deployment and should be treated as confidential. Share them only through the channel your support contact provides, not in a public issue or channel.