[CSIT-1979] 3na-spr: Flow related ipsec tests fail on CX7

[vvalderrv]

Description

The affected suites are ethip4ipsec1000tnlsw-fixtnlip-ip4base-policy-flow-dir-aes256gcm and ethip4ipsec1000tnlsw-fixtnlip-ip4base-policy-flow-rss-aes256gcm, both failing [0] due to retval -1 from flow_enable API call, which means VNET_FLOW_ERROR_NOT_SUPPORTED.

I bet only new-ish Intel NICs support such hardware flows, but we should confirm that before deleting those tests from jobspecs.

Note that similarly named suites such as ethip4ipsec10000tnlsw-ip4base-policy-flow-cache-aes256gcm do not actually rely on hardware flows, so they are passing also on Mellanox NICs.

[0] https://s3-logs.fd.io/vex-yul-rot-jenkins-1/csit-vpp-perf-report-coverage-2410-3na-spr/33/log.html.gz#s1-s1-s1-s1-s2-t1-k2-k9

Assignee

Unassigned

Reporter

Vratko Polak

Comments

• vrpolak (Fri, 22 Nov 2024 16:03:34 +0000): Using experiments [2] in VPP I convinced myself the NIC really does not offer those capabilities (in ways DPDK libraries could expose to VPP dpdk plugin).

[2] 41875: experiment: distinguish flow enable errors | https://gerrit.fd.io/r/c/vpp/+/41875

• vrpolak (Tue, 19 Nov 2024 10:00:38 +0000): Those tests were failing [1] also previously in rls2406, but the fails were misidentified, probably as a different ipsec policy issue such as CSIT-1938.

[1] https://s3-logs.fd.io/vex-yul-rot-jenkins-1/csit-vpp-perf-report-coverage-2406-3na-spr/6/log.html.gz#s1-s1-s1-s1-s2-t1-k2-k9

Original issue: https://jira.fd.io/browse/CSIT-1979

[vvalderrv]

Using experiments [2] in VPP I convinced myself the NIC really does not offer those capabilities (in ways DPDK libraries could expose to VPP dpdk plugin).

[2] 41875: experiment: distinguish flow enable errors | https://gerrit.fd.io/r/c/vpp/+/41875

[vvalderrv]

Those tests were failing [1] also previously in rls2406, but the fails were misidentified, probably as a different ipsec policy issue such as CSIT-1938.

[1] https://s3-logs.fd.io/vex-yul-rot-jenkins-1/csit-vpp-perf-report-coverage-2406-3na-spr/6/log.html.gz#s1-s1-s1-s1-s2-t1-k2-k9

[vrpolakatcisco]

Fixed [3] by no longer running those tests on that NIC.

[3] 41920: fix(jobspec): Do not test hardware flows on CX7 | https://gerrit.fd.io/r/c/csit/+/41920