Community CI Meeting Minutes - March 6, 2025

Community CI Meeting Minutes - March 6, 2025

[Patrick Robb]

[-- Attachment #1: Type: text/plain, Size: 6353 bytes --]

#####################################################################
March 6, 2025
Attendees
. Patrick Robb
. Paul Szczepanek
. Luca Vizzarro
. Aaron Conole
. Andre Muezerie

#####################################################################
Minutes

=====================================================================
General Announcements
* DPDK Summit Prague is May 8-9:
   * The CFP has been extended to March 9
   * Aaron will submit a CFP for the patch submission policy in DPDK. This
pertains to the frequency of patchseries submissions, patch size, and the
importance for developers to do some basic testing (like compile testing)
before submitting to the CI. One issue is that patch volume is so high that
the dev mailing list is more of a “write only mailing list” than one which
a human can reasonably read and provide reviews for.
   * Patrick will submit a CFP for a talk about UNH’s experience running
DTS in it’s CI automation

=====================================================================
CI Status

---------------------------------------------------------------------
UNH-IOL Community Lab
* UNH has added a 32 bit x86_32 cross compile test (debian), using a cross
compile target Bruce R added during 24.11:
https://git.dpdk.org/dpdk/tree/config/x86/cross-32bit-debian.ini
* Robin Jarry (redhat) has reached out to ask if the DPDK graph router
(Grout) that he is developing can be tested on UNH lab hardware. We
discussed it with tech board yesterday and unless there are significant
hurdles which will require lots of man hours at UNH, we should do this.
I’ve reached out to Robin with some initial questions and gotten these
clarifications:
   * 1. We could test 1x/week on the nightly build
https://github.com/DPDK/grout/releases/tag/edge
   * 2. The initial testing would be for functional regressions only. Once
they do performance optimizations, they may be interested in perf testing.
   * 3. Test reports can be sent to the grout mailing list:
https://mails.dpdk.org/listinfo/grout
   * 4. They have some tests currently, which may have to be adapted
slightly for bare-metal, but can be used by UNH:
https://github.com/DPDK/grout/tree/main/smoke
   * 5. Robin indicated any recent NIC is okay (X710, E810, CX5, CX6)
   * Patrick will try to stand up Grout on the hardware to investigate the
basics, and send any further questions to Robin. Then we can loop back to
tech board and seek their final approval for using the lab resources for
this.
   * Aaron has highlighted that the grout guys should complete most/all of
the scripting, so that the UNH resources (which are really for DPDK CI
testing) are not exhausted in efforts to maintain grout testing
* Arm Grace had an infra failure yesterday which cut the performance
numbers in half for runs in the 2nd half of the day. I didn’t have time to
do a deep dive in this case (just restarted the machine and triggered
retests) but we’ll keep an eye on it.
* Dpdk-test: per_lcore_autotest had a spurious fail in UNH CI - documented
it on Bugzilla https://bugs.dpdk.org/show_bug.cgi?id=1668
   * David has a suggestion for what the issue is: “it starts a callback
that wait 100ms on a separate thread, and thinks that nothing will disturb
the main thread.”
* lab.dpdk.org got an update to the code coverage report page, which allows
the coverage reports to be opened within the website, instead of the user
downloading the .zip and opening locally:
   * https://lab.dpdk.org/results/dashboard/code-coverage
* Cody is updating our XL710 testing, recreating the test on a newer
testbed (Intel Xeon 3rd gen processor). This will probably run silently for
the rest of 25.03, then we will shift over to the new server with the new
release.
* Had some false failures for new DTS because we rebooted the server
without modprobing vfio-pci before triggering new CI tests. Legacy DTS
loads the DPDK driver, but we elected not to for new DTS in order to allow
space for users to load vfio-pci with custom params. This is fine, I am
just noting here that this is still on the user to do before running
in-repo DTS.
Intel Lab
* Reported some failures yesterday, I think they were accurate but Patrick
should double check to make sure.

---------------------------------------------------------------------
Github Actions
* David added a patch which adds windows compilation for the github robot,
which has been running for a few weeks. It’s a server 2022 VM running the
MSVC compile toolchain.
* No new news for FreeBSD

---------------------------------------------------------------------
AWS
* They are working on their test-report email templating script

---------------------------------------------------------------------
Loongarch Lab
* None

=====================================================================
DTS Improvements & Test Development
* Can we make a “dev” branch on dpdk-next-dts. Indicate to maintainers that
they should pull from for-main?
   * Patrick can email Thomas and David asking for approval
* VF creation:
   * Can be pushed to next release
   * Mellanox VFs seem to work fine, but Patrick is having some issues
forwarding packets with XL710 VFs, even with promisc=on
* Per testsuite configs:
   * Update to run DTS, enforcing a single “testrun.”
* Build-less driver binding:
   * Are the temp dirs being used?
https://patchwork.dpdk.org/project/dpdk/patch/20250224132823.196509-4-luca.vizzarro@arm.com/
      * Yes, we are copying the dpdk-devbind.py into these dirs
      * This also stages some future work, like the artifact manager. We
will have to discuss at the start of the 25.07 development cycle.
* Luca has picked up two old patches:
   * https://patches.dpdk.org/project/dpdk/list/?series=32704
   * https://patches.dpdk.org/project/dpdk/list/?series=32888
* Performance testing:
   * Nick experimented with running TREX from .pcaps which can be written
from Scapy, and saw no benefit as compared to our initial approach (run
from /stl python defined traffic profiles)
* The testpmd patch which adds the L4 port # to testpmd verbose output is
on next-net.

=====================================================================
Any other business
* Next Meeting Mar 20, 2025

[-- Attachment #2: Type: text/html, Size: 7196 bytes --]