#fdio-csit: FD.io CSIT project meeting

Meeting started by mackonstan at 13:59:06 UTC (full logs).

Meeting summary

    1. Juraj Linkeš (jlinkes, 13:59:54)

  1. Agenda bashing (mackonstan, 14:00:43)
    1. Vratko Polak. (vrpolak, 14:00:51)
    2. Jan Gelety (jgelety, 14:00:58)

  2. FD.io CSIT physical labs (mackonstan, 14:02:37)
    1. Peter Mikus (pmikus, 14:02:50)
    2. Dave Wallace (dwallacelf, 14:02:57)
    3. Maciek, Peter: Denverton testbeds are finally working, test jobs are running, Ray and Luyong confirmed. (mackonstan, 14:04:03)

  3. FD.io lab extension (mackonstan, 14:08:41)
    1. Maciek: provided update on the spec of lab extension (mackonstan, 14:08:55)
    2. Following new servers are expected to arrive in Jul/Aug: (mackonstan, 14:09:14)
    3. Servers: 7x SuperMicro 7049GP-TRT (4U), 3x SuperMicro 1029P-WTRT (1U) (mackonstan, 14:09:23)
    4. Processors: 14x Cascade Lake Platinum 8280, 6x Cascade Lake Gold 6252N (with priority based frequency, will arrive later) (mackonstan, 14:09:32)
    5. NICs: 7x XXV710-DA2, 7x X710DA4 (mackonstan, 14:09:42)
    6. Proposed setup: i) Performance testbeds: 2x 2n-clx plus 1x 3n-clx, or 3x 2n-clx plus 1x 1n-clx, ii) Backends: 3x SuperMicro 1029P-WTRT (mackonstan, 14:10:09)
    7. Propose to evaluate the pros/cons on the csit-dev list (mackonstan, 14:16:36)

  4. Inputs from VPP (mackonstan, 14:16:57)
    1. Dave: v19.08 continues, no calls for maintenance releases (mackonstan, 14:17:12)

  5. Inputs from HC (mackonstan, 14:17:29)
    1. it's quiet, no changes for CSIT (mackonstan, 14:18:19)
    2. Peter: 1908 plan for HC is unclear (mackonstan, 14:19:03)

  6. Inputs from TSC (mackonstan, 14:19:18)
    1. Vratko: week before last there was discussion about ROI/efficiency of FD.io minisummits co-located with major events. Calls for proposals for better, more efficient ways for FD.io community get togethers (mackonstan, 14:21:38)

  7. Operational status (mackonstan, 14:22:09)
    1. Maciek: Jenkins maintenance last night UTC, impacted perf-mrr-daily jobs (mackonstan, 14:23:27)
    2. Peter: jobs seem to be running (mackonstan, 14:23:40)
    3. Vratko: looks like there is still some memory leak in the system, compile jobs are failing, merge jobs issues, other symptoms, so expecting more maintenance (mackonstan, 14:24:48)
    4. Vratko, Ed: admins are adding more logs to track down the memory leak issue (mackonstan, 14:26:28)
    5. above issues were happening before the maintenance (mackonstan, 14:31:53)
    6. so far no issues observed after the maintenance (mackonstan, 14:32:03)

  8. Centos CI (mackonstan, 14:33:40)
    1. Peter: need a separate job for Centos7, need to enable it in Nomad env on 1n-skx. Should not require any changes to Nomad env, job should just work, will report if issues. (mackonstan, 14:36:30)
    2. Jan: two issues in trending: i) ip6 fib scale tests failing, bug in VPP VAT fixed by Neale; ii) ip route api parameter name change was missed in CSIT code, now fixed, some side fix required in VPP code too; iii) vxlan scale test are failing due to PAPI not good with handling scale, moving back to VAT as for all other scale tests (mackonstan, 14:40:02)

  9. VPP code performance (mackonstan, 14:40:14)
    1. patchy coverage due fib api issue per above, and jenkins issues (mackonstan, 14:41:50)

  10. Framework: VAT to PAPI migration (mackonstan, 14:42:49)
    1. Vratko: see email with implementation steps (mackonstan, 14:43:08)
    2. Vratko: already know that CSIT scale tests do not work with PAPI, ip4 fib 200k takes >30min with synch PAPI and 10min with remote socket, show stopper for usage of PAPI (mackonstan, 14:44:38)
    3. Vratko: async mode may help to bring the time down, but expecting issues so unlikely to get it working in 1908 (mackonstan, 14:47:38)
    4. Vratko: another issue is specific to Centos - remote socket PAPI uses ssh tunnel, and this seems to be not supported in Centos7 (mackonstan, 14:49:29)
    5. VAT-to-PAPI migration on track to be finished this release cycle (jgelety, 14:50:36)

  11. Framework: VPP-API "Flag Day" gate-keeping (mackonstan, 14:51:47)
    1. Vratko: will be able to make progress on it once remote socket PAPI is in production (mackonstan, 14:52:06)
    2. Dave: still need to update the proposed process doc (mackonstan, 14:52:25)

  12. Test environments - VPP_Device environment (mackonstan, 14:52:50)
    1. Juraj: Arm onboarding, having issues with understanding how the pipeline works, as the process is unclear, lacking documentation (mackonstan, 14:53:51)
    2. Juraj: posted a pull request to add Arm to VPP_device environment (mackonstan, 14:57:00)
    3. Peter: need to complete review of Juraj's pull request, run tests, and if okay merge (mackonstan, 14:58:47)
    4. Peter: ongoing issue is lack of documentation (mackonstan, 14:59:07)
    5. let's drive the conclusion over email, if not converging then separate TWS call (mackonstan, 14:59:48)
    6. Vratko: discussion on slack with Denverton team want to remove tests that are running daily on Denverton (mackonstan, 15:01:34)
    7. Vratko: awaiting patch from Denverton team side (mackonstan, 15:01:52)
    8. Vratko: e.g. some scale tests can not run on some platforms (mackonstan, 15:02:51)


Meeting ended at 15:03:41 UTC (full logs).

Action items

  1. (none)


People present (lines said)

  1. mackonstan (50)
  2. collabot` (4)
  3. jgelety (2)
  4. pmikus (1)
  5. vrpolak (1)
  6. jlinkes (1)
  7. dwallacelf (1)


Generated by MeetBot 0.1.4.