Revision as of 12:22, 2 February 2024

CSIT Test Failure Clasification

All known CSIT failures grouped and listed in the following order:

• Always failing followed by sometimes failing.
• Always failing tests:
  • Most common use cases followed by less common.
• Sometimes failing tests:
  • Most frequently failing followed by less frequently failing.
    • High frequency 50%-100%
    • medium frequency 10%-50%
    • low frequency 0%-10%.
  • Within each sub-group: most common use cases followed by less common.

CSIT Test Fixing Priorities

Test fixing work priorities defined as follows:

• (H)igh priority, most common use cases and most common test code.
• (M)edium priority, specific HW and pervasive test code issue.
• (L)ow priority, corner cases and external dependencies.

Current Failures

Deterministic Failures

In Trending

(M) DPDK 23.03 testpmd startup fails on some testbeds

• last update: 2023-11-06
• work-to-fix: medium
• rca: The DUT-DUT link sometimes does not go up. The same consequences as CSIT-1848 but affects more testbed+NIC combinations. Can be fixed by a different startup procedure (better detection before for restart).
• test: Testpmd (no vpp). Around half of tested+NIC combinations are affected.
• frequency: Low, ~1%, as in most testbeds the link does go up.
• testbed: multiple
• example:https://s3-logs.fd.io/vex-yul-rot-jenkins-1/csit-dpdk-perf-report-iterative-2310-2n-zn2/8/log.html.gz#s1-s1-s1-s1-t6-k2-k4
• ticket: CSIT-1904

(M) 2n-spr: zero traffic on cx7 rdma

• last update: 2023-06-22
• work-to-fix: medium
• rca: VPP reports "tx completion errors", more investigation ongoing.
• test: All tests on 2n-spr with 200Ge2P1Cx7Veat NIC and RDMA driver.
• frequency: always (since 2n-spr was set up)
• testbed: 2n-spr
• example: https://s3-logs.fd.io/vex-yul-rot-jenkins-1/csit-vpp-perf-mrr-daily-master-2n-spr/58/log.html.gz#s1-s1-s1-s5-s19-t1-k3-k7-k1-k1-k1-k8-k14-k1-k1-k1-k1
• ticket: CSIT-1906
• note: Also would affect 3n-alt with mlxc6 and rdma. Will probably be made invisible by removing rdma (except mlxc5) from jobspecs.

(L) 3n-icx: negative ipackets on TB38 AVF 4c l2patch

• last update: 2023-04-19
• work-to-fix:
• rca:
• test: TB38 AVF 4c e810cq, earlier l2patch nowadays eth-l2xcbase
• frequency: always
• testbed: 3n-icx (only TB38, never TB37)
• example: https://s3-logs.fd.io/vex-yul-rot-jenkins-1/csit-vpp-perf-mrr-daily-master-3n-icx/214/log.html.gz#s1-s1-s1-s5-s3-t3-k2-k9-k8-k13-k1-k2
• ticket: CSIT-1901

(L) 2n-tx2: af_xdp mrr failures

• last update: 2023-11-08
• work-to-fix:
• rca: Some workers see no traffic, "error af_xdp_device_input_refill_db: rx poll() failed: Bad address" in show hardware. More examination needed.
• test: ip4 and ip6, base and scale
• frequency: more than 80% on a subset of cases, 100% on (multicore) 2n-tx2
• testbed: 2n-tx2; to a lesser extent also 2n-clx and 2n-icx, where just decreased performance (and failure in ndrpdr) is more likely outcome.
• example: https://s3-logs.fd.io/vex-yul-rot-jenkins-1/csit-vpp-perf-mrr-daily-master-2n-spr/120/log.html.gz#s1-s1-s1-s2-s2-t2-k2-k10-k9-k14-k1-k1-k1-k1
• ticket: CSIT-1922

Not In Trending

(M) all testbeds: some 9000B tests

• last update: 2023-02-09
• work-to-fix: hard
• rca: VPP code: 34839: dpdk: cleanup MTU handling. CSIT needs to rework how it sets MTU / max frame rate (CSIT-1797). Some tests will continue failing due to missing support on VPP side, we will open specific Jira tickets for those.
• test: see sub-items
• frequency: always
• testbed: all
• examples: see sub-items
• ticket: CSIT-1809
• gerrit: https://gerrit.fd.io/r/c/csit/+/37824

(M) tests with 9000B payload frames not forwarded over memif interfaces

• last update: 2023-02-09
• work-to-fix: hard
• test: 9000B + memif
• testbed: 2n-skx, 3n-skx, 2n-clx
• examples: 2n-skx Memif
• ticket: CSIT-1808

(L) l3fwd error in 200Ge2P1Cx7Veat-Mlx5 test with 9000B

• last update: 2023-06-28
• work-to-fix: medium
• test: 9000B + Cx7 with DPDK DUT
• testbed: 2n-icx
• examples: [1]
• ticket: CSIT-1924

(L) 2n-clx, 2n-icx: nat44ed cps 16M sessions scale fail

• last update: 2023-07-12
• work-to-fix: hard
• rca: Ramp-up trial takes more than 5 minutes so sessions are timing out.
• test: 64B-avf-ethip4tcp-nat44ed-h262144-p63-s16515072-cps-ndrpdr 1c 2c 4c, 64B-avf-ethip4udp-nat44ed-h262144-p63-s16515072-cps-ndrpdr 1c 2c 4c
• frequency: always
• testbeds: 2n-skx, 2n-clx, 2n-icx
• example: https://s3-logs.fd.io/vex-yul-rot-jenkins-1/csit-vpp-perf-report-coverage-2306-2n-clx/8/log.html.gz#s1-s1-s1-s1-s14-t2-k2-k15-k1
• ticket: CSIT-1799

(L) 2n-clx, 2n-icx: nat44det imix 1M sessions fails to create sessions

• last update: 2023-07-12
• work-to-fix: medium
• rca: One possible cause is CSIT not counting ramp-up rate properly for IMIX (if multiple packets belong to the same session).
• test: IMIX over 1M sessions bidir
• frequency: always
• testbed: 2n-skx, 2n-clx, 2n-icx
• example: https://s3-logs.fd.io/vex-yul-rot-jenkins-1/csit-vpp-perf-report-coverage-2306-2n-clx/7/log.html.gz#s1-s1-s1-s1-s2-t4-k2-k11-k1-k2
• ticket: CSIT-1884

Occasional Failures

In Trending

(H) 2n-icx: NFV density VPP does not start in container

• last update: before 2023-01-31
• work-to-fix: hard
• rca:
• test: all subsequent
• frequency: medium
• testbed: 2n-icx
• example: 2n-icx mrr, 2n-icx ndrpdr
• ticket: CSIT-1881
• note: Once VPP breaks, all subsequent tests fail. Even all subsequent builds will be failing until Peter makes TB working again. Although it's failing with medium frequency when it happens it breaks all subsequent builds on the TB therefore [H] priority.

(M) 2n-clx: e810 mlrsearch tests packets forwarding in one direction

• last update: 2023-07-12
• work-to-fix: hard
• rca: Packets not seen on VPP input interface, deeper causes unknown.
• test: e810Cq ip4base, ip6base
• frequency: high
• testbed: 2n-clx
• example: https://s3-logs.fd.io/vex-yul-rot-jenkins-1/csit-vpp-perf-report-coverage-2306-2n-clx/2/log.html.gz#s1-s1-s1-s1-s1-t1-k3-k7-k1-k1-k1-k8-k14-k1-k1-k1-k1
• ticket: CSIT-1864

(M) 3n-icx, 3n-snr: wireguard 100 and 1000 tunnels mlrsearch tests failing with 2c and 4c

• last update: before 2023-01-31
• work-to-fix: easy
• rca:
• test: wireguard 100 tunnels and more
• frequency: high
• testbed: 3n-icx, 3n-snr
• examples: 3n-icx
• ticket: CSIT-1886

Rare Failures

In Trending

(M) 3n-icx, 3n-snr: 1518B IPsec packets not passing

• last update: before 2023-01-31
• work-to-fix: hard
• rca:
• test: all AVF crypto
• frequency: low
• testbed: 3n-skx, 3n-icx, 3n-snr
• example: 3n-icx daily, 3n-snr, 3n-icx weekly
• ticket: CSIT-1827

(M) all testbeds: mlrsearch fails to find NDR rate

• last update: before 2023-06-22
• work-to-fix: hard
• rca: On 3n-tsh, the symptom is TRex reporting ierrors, only in one direction. Other testbeds may have a different symptom, but failures there are less frequent.
• test: Crypto, Ip4, L2, Srv6, Vm Vhost (all packet sizes, all core configurations affected)
• frequency: low
• testbed: 3n-tsh, 3n-alt, 2n-clx
• example: [2]
• ticket: CSIT-1804

(M) all testbeds: AF_XDP mlrsearch fails to find NDR rate

• last update: before 2023-01-31
• work-to-fix: hard
• rca:
• test: af-xdp multicore tests
• frequency: low
• testbed: 2n-clx, 2n-skx, 2n-tx2, 2n-icx
• example: 2n-skx, 2n-clx
• ticket: CSIT-1802
• note: This is mainly observed in iterative and coverage. It's very low frequency ~ 1 out of 100

(L) all testbeds: vpp create avf interface failure in multi-core configs

• last update: 2023-02-06
• work-to-fix: hard
• rca: issue in Intel FVL driver
• test: multicore AVF
• frequency: low
• testbed: all testbeds
• example: 2n-clx, 3n-icx
• ticket: CSIT-1782
• note: A long standing issue without a final permanent fix.

(L) all testbeds: nat44det 4M and 16M scale 1 session not established

• last update: 2023-02-14
• work-to-fix: hard
• rca: unknown
• test: nat44det udp 4m and 16m (64k is ok, 1m can fail but rarely than bigger scales)
• frequency: low
• testbed: 2n-zn2, 2n-skx, 2n-icx, 2n-clx
• example: 2n-zn2, 2n-clx
• ticket: CSIT-1795

Past Failures