December 11, 2025
Attendees
1. Patrick Robb
2. Aaron Conole
3. Paul Szczepanek
4. Luca Vizzarro
Minutes
General Announcements
* Upcoming meetings:
* 12/25 CI meeting: Canceled because of Christmas
* 01/01 DTS meeting: Canceled because of New Year's Day
* Patchwork API requests: Ali who maintains the patchwork instance has reached out to various labs asking if they can reduce the amount of requests they are making to patchwork.
* UNH:
* Adding a trailing “/” to the URL for requests, so there is no 301 redirect: Will deploy today
* Properly increment the “since” timestamp arg for the test recheck request script - deploying Friday.
* Will check in with Ali at the end of the week to make sure we are good to go
* Github Robot:
* Ozlabs patchwork instance has also instated some new policies which affect the robot
* The robot was working by monitoring series until that series was moved into a different state (i.e. new, accepted, superseded, etc.)
* Since the robot started running about 7 years ago, there have been about 1000 series that are “incomplete,” meaning only a subsection of the patches in a series were actually submitted. These got marked as archived, which meant that the robot continued to track these, resulting in a significant amount of unnecessary requests being made.
* Going forward, the robot drops a patchseries from consideration once it’s age gets beyond 30 days
* Ozlabs: Now enforcing we are using https instead of http
* The robot has been updated to use https now
* The robot has been running for 7 years with minimal downtime. So, it has various older software package, including sqlite3, etc. Aaron’s version of sqlite3 requires strict SQL syntax, but the version of sqllite on the server does not accept this syntax. So, Aaron made a bunch of changes, but the robot is not functioning properly due to this sql issue
* There are now some failures which Aaron is debugging but which are confusing, i.e. an echo command failing
* There may be some bad records in the database now. Aaron is looking through these and will resolve the remaining bugs before correcting the database / rolling back.
* There may be some downtime or a need to reboot the server hosting the CI system
* Loongson:
* Bibo is the new maintainer, and is looking at the requests that Ali has mentioned.
* Patrick will offer to look at the dpdk-ci scripting with Bibo if he has issues identifying where the inadvertent requests are coming from
* AWS: Seems like there are no issues with requests to patchwork API
* CI Status
AWS Lab
* They had a false failure last week due to a conflict on 2 jobs using the same meson dir concurrently. Waleej from AWS has resolved this and put in a retest.
UNH-IOL Community Lab
* ABI Testing:
* We have built the 26.0 ABI reference from the 25.11 release and begun ABI testing for the 26.03 release
* In addition, we are also running ABI testing on 22.11-staging, 23.11-staging, 24.11-staging, and 25.11-staging, against the ABI references of their respective releases.
* ARM Grace:
* Testing will run fine for some weeks, then at some point the NIC gets into a bad state and performance drops significantly
* Weird behavior: NIC insterface 0 showin as /dev/fb0, rebinding to ICE at the end now results in input/output error, etc.
* We are distro upgrading and/or reprovisioning some of our servers in the CI Lab in the next week, which will cause some small downtime
* If there is any significant downtime Patrick can update on Slack or the mailing list
* DTS:
* We are working on enabling the new single core forwarding test on our testbeds, which will mean these can stop running legacy DTS on our servers for this test.
* Otherwise, the only remaining reason to use legacy DTS is cryptodev testing, which is being worked on for the current DTS framework right now and should have coverage arriving in 26.03
* Dean at UNH needs to reach out to John Mcnamera and ask about what information we can share from runs on the flow offload test
Intel Lab
* None
Github Actions Robot
* Aaron is actively working on an LLM integration into the pipeline that is not coderabbit or sourcery
* Using Sonnet4 (which seems to be working better than sonnet4.5)
* It is providing some fairly high quality reviews
* Aaron did post some related patches to the ovs dev mailing list. This is applicable to DPDK as well
* https://mail.openvswitch.org/pipermail/ovs-dev/2025-November/427871.html
* It should be possible to apply this to the DPDK tree eleven though it was targeting OvS
* Example review: https://mail.openvswitch.org/pipermail/ovs-dev/2025-November/427836.html
* The replies are inline properly and are raising relevant questions
* The downside is the cost (but it should be manageable)
* With properly tailored prompts, Aaron is seeing much higher quality in the reviews, now there is a reasonable ratio of good review comments vs “bogus” comments
* To save money, one option is to run the normal testing first, and then only do an LLM review if the initial testing is passing. That should save on requests to the LLM service.
Loongson Lab
* Bibo is looking at the dpdk-ci scripting to reduce their requests being made to the Patchwork API
DTS Improvements & Test Development
* Roadmap: We will use the roadmap we finalized last Thursday
* Virtio: We sent Maxime an email asking about what additional test coverage is needed for virtio - we will proceed when we have his feedback.
* Otherwise, Dean is working on an addition pvvp testcase currently.
Any other business
* Next meeting is January 8, 2026