#fdio-csit: FD.io CSIT Project Meeting

Meeting started by mackonstan at 14:00:27 UTC (full logs).

Meeting summary

    1. Vratko Polak. (vrpolak, 14:00:44)

  1. Agenda bashing (mackonstan, 14:01:08)
    1. Tibor Frank (tifrank, 14:01:21)
    2. Dave Wallace (dwallacelf, 14:02:03)

  2. Inputs from projects - VPP (mackonstan, 14:02:17)
    1. Fan Zhang (fan_zhang, 14:02:33)
    2. Andrew - branch made for stable/2106, v2110-rc0 tag applied (mackonstan, 14:03:45)

  3. Inputs from projects - VSAP (mackonstan, 14:04:00)
    1. Maciek, main patch merged, now need to work on the way benchmark results get presented in the report (mackonstan, 14:05:17)

  4. Inputs from projects - TSC (mackonstan, 14:05:24)
    1. Juraj LinkeŇ° (jlinkes, 14:07:22)
    2. Here is the pointer to technical community doc edits reflecting TSC evolution (mackonstan, 14:07:37)
    3. https://github.com/FDio/site/pull/92 (mackonstan, 14:07:40)
    4. https://lists.fd.io/g/tsc/message/1531 <-- Ed's e-mail with link to github PR (above) inside. (vrpolak, 14:08:11)

  5. Releases CSIT-2106 (mackonstan, 14:08:11)
    1. Tibor, pulling the branch later today (mackonstan, 14:10:06)
    2. Reviewing status of 2106 content per table in wiki (mackonstan, 14:11:52)
    3. https://wiki.fd.io/view/CSIT/csit2106_plan (mackonstan, 14:12:00)
    4. https://gerrit.fd.io/r/c/csit/+/32025 (fan_zhang, 14:13:22)
    5. Yulong, have 2 patches for review: i) gtpu sw offload perf test case, ii) dpdk options for graph node variants. (mackonstan, 14:16:42)
    6. https://gerrit.fd.io/r/c/csit/+/32025 (mackonstan, 14:19:24)
    7. 32409: Add dpdk max-simd-bitwidth configuration. | https://gerrit.fd.io/r/c/csit/+/32409 (mackonstan, 14:19:57)
    8. Yulong, would like to proceed with moving the flow offload functional tests to vpp_device execution once the CVL NICs are installed (mackonstan, 14:21:40)
    9. Peter, confirmed that CVL NICs got received, and will be installed in vpp_device testbeds in the matter of week (currently delayed by the IPS DPDK VF racing condition issue) (mackonstan, 14:22:35)

  6. Physical Infrastructure (mackonstan, 14:22:52)
    1. FD.io DC move in Apr/May 2021 - completed (mackonstan, 14:23:19)
    2. Peter, one CLX build/storage machine down due to CLX failure, Ray shipped replacement (mackonstan, 14:24:16)
    3. Maciek, 3n-hsw testbed decommissioning, see point 7.e. from last week minutes (mackonstan, 14:30:43)
    4. https://ircbot.wl.linuxfoundation.org/meetings/fdio-csit/2021/fd_io_csit_project_meeting/fdio-csit-fd_io_csit_project_meeting.2021-05-19-13.59.html (mackonstan, 14:30:52)
    5. Vratko, revised the view about keeping one 3n-hsw per point 7.g. in notes from last week, and okay to delete 3n-hsw from csit (3n-dnv have qat and can run qat tests once they are supported again) (mackonstan, 14:32:17)
    6. Maciek, going ahead with removing 3n-hsw from CSIT (jobs, job_specs, trending, report, tests) (mackonstan, 14:33:33)

  7. Virtual Infrastructure, CI/CD (mackonstan, 14:33:50)
    1. Steps for removing 3n-hsw: 1.PAL, 2.ci-management, 3. job specs, 4. lab docs, 5. hardware. (vrpolak, 14:34:45)
    2. Dave, last failures seen over the weekend (git clone failures) (mackonstan, 14:36:11)
    3. Dave, last monday bigger call with LF IT team, discussion about the root cause, the actual RC not identified. Can't repro the problem either (mackonstan, 14:36:35)
    4. Dave, at the moment things are looking good from CI/CD operations perspective (mackonstan, 14:37:11)

  8. Performance Tests (mackonstan, 14:41:19)
    1. Vratko, nat44ed multi-worker tests are not failing any more. (mackonstan, 14:42:24)
    2. Vratko, re avf multi-worker tests failures, situation has improved, frequency of failures has decreased (mackonstan, 14:44:51)
    3. Juraj, also noticed some failures, container-memif tests 4c 8c failing on arm and xeons (mackonstan, 14:45:50)
    4. Vratko, may be related to avf failures (mackonstan, 14:46:32)
    5. Reduce number of tests from trending, iterative, report (mackonstan, 14:49:43)
    6. Maciek, asking for comments / reviews on listed patches if folks care about those tests (mackonstan, 14:50:08)

  9. Functional Tests (mackonstan, 14:51:05)
    1. Peter, Juraj, finally nailing down the long-term saga issue "Faulty VF initialization during DPDK startup when multiple DPDK instances use different VFs with the same PF" (mackonstan, 14:51:37)
    2. Juraj, after a long long conversation and multitude requests to try and test many different / latest-greatest combinations of DPDK stack, that all were broken (mackonstan, 14:52:15)
    3. Juraj, issue narrowed down to a DPDK bug related to switching from i40e to iavf code being in progress, with some DPDK code still using i40e code that got deprecated/removed (mackonstan, 14:53:49)
    4. Juraj, tested patch provided by Ray that forces dpdk to use iavf (mackonstan, 14:55:41)
    5. Juraj, it works now, waiting for official patch from dpdk (mackonstan, 14:56:02)
    6. https://lists.fd.io/g/csit-dev/message/4337 (fan_zhang, 14:59:53)
    7. Juraj, once dpdk patch is confirmed, happy to add it to dpdk patches for vpp (mackonstan, 15:01:19)
    8. Fan, what is the best way to proceed with flow offload for PF? (mackonstan, 15:01:41)
    9. Peter, these can be supported in perf testbed, perf tests (mackonstan, 15:01:57)
    10. Maciek, Peter, Vratko, agreed that it makes sense to code them as perf tests using TRex (mackonstan, 15:02:24)

Meeting ended at 15:04:31 UTC (full logs).

Action items

  1. (none)

People present (lines said)

  1. mackonstan (47)
  2. collab-meetbot (3)
  3. vrpolak (3)
  4. fan_zhang (3)
  5. tifrank (1)
  6. dwallacelf (1)
  7. jlinkes (1)

Generated by MeetBot 0.1.4.