CI: BPF Checks --TestBPF/bpf_nat_tests.o/nat4_port_allocation_tcp #42599
Description
CI failure
In the past week, ~2.5% of BPF Checks workflow runs have been failing with this error.
--- FAIL: TestBPF (39.98s)
--- FAIL: TestBPF/bpf_nat_tests.o (1.43s)
--- FAIL: TestBPF/bpf_nat_tests.o/nat4_port_allocation_tcp (0.58s)
bpf_test.go:480: bpf_nat_tests.c:1412: assert failed at bpf_nat_tests.c:1412
┌──────────────────────────────────────────────────────────────────────────────────────────────┐
│ STATUS │ ELAPSED │ PACKAGE │ COVER │ PASS │ FAIL │ SKIP │
│─────────┼─────────┼────────────────────────────────────────────┼───────┼──────┼──────┼───────│
│ FAIL │ 40.01s │ github.com/cilium/cilium/bpf/tests/bpftest │ -- │ 536 │ 3 │ 0 │
└──────────────────────────────────────────────────────────────────────────────────────────────┘
Looking at the code, this is the failing assertion:
/* Only occasional failures at 50% of the test. */
assert(retries_50percent[SNAT_COLLISION_RETRIES] < 15);This tests the port allocation algorithm in SNAT, which has a level of stochasticism incorporated into it. Seems that failures should occur to some level, but more investigation is required.
Here are the two runs that have hit this:
- https://github.com/cilium/cilium/actions/runs/18934561156/job/54058216154
- https://github.com/cilium/cilium/actions/runs/19071506098/job/54475532408
The only artifacts generated during the test were the junits zip:
cilium-junits.zip