On FortiGate 4xF/6xF models running FortiOS 7.6.4, issue 1206506 can stop IPsec traffic while the tunnel remains up. The documented temporary fix is to disable npu-offload on the affected tunnel; fixed releases are listed below. Use this workaround only for that model/build case.
4xF/6xF on 7.6.4: apply the scoped offload workaround
Fortinet documents issue 1206506 on 4xF/6xF running FortiOS 7.6.4: IPsec may stop carrying both user and local-out traffic. The exact issue is listed resolved in 7.4.10 and 7.6.5.
Before applying the workaround, read the affected existing tunnel in its owning VDOM. On a multi-VDOM unit, a global administrator first enters config vdom and edit "<owning-vdom>"; stay in that context for both the read and change below.
show full-configuration vpn ipsec phase1-interface "<affected-existing-tunnel>"Record the effective npu-offload value before changing it. If it is already disable, skip the change: this workaround is already applied, so continue the data-path investigation. If it is enable and the documented issue matches, the temporary workaround below disables NPU offload. It flushes the tunnel, interrupts traffic and increases CPU processing:
config vpn ipsec phase1-interface
edit "<affected-existing-tunnel>"
set npu-offload disable
next
endAfter the VDOM-local change block, a global administrator uses next and end to leave the VDOM context. Restoring the recorded offload value also requires a coordinated interruption. Do not apply the workaround to every tunnel as a generic diagnostic.
Scope: FortiGate route-based IPsec data-path failures. The hardware-offload exception above applies only to its named models/build.
Other models or failures tied to rekey
Run tunnel diagnostics in the VDOM that owns the tunnel. On a multi-VDOM unit, a global administrator enters config vdom and edit "<owning-vdom>" before the command below; use next and end afterward to return to the top-level prompt. A VDOM administrator is already in the assigned context.
diagnose vpn tunnel list name <existing-tunnel>If the model/build does not match, compare the relevant proxyid, sa count and expire through the failure. A missing Child SA or failure at rekey belongs in the proposal/PFS correction; a subnet-only failure belongs in the selector, route and policy fixes. Keep SA output private.
Identify the model and current load
get system status
get system performance statusRecord the model and exact version/build from get system status, plus CPU, memory and session load from get system performance status. Compare the same fields before and after any offload workaround; CPU pressure can replace the original fault with a different bottleneck.
Verify through the original failure interval
Repeat the application workload and any affected BGP/SD-WAN probes. For rekey failures, observe the next relevant rekey; for load-related failures, compare the same load and CPU behavior.