-
Notifications
You must be signed in to change notification settings - Fork 237
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Pass enable-udp-aggregation=true to ovn-kubernetes #1533
Pass enable-udp-aggregation=true to ovn-kubernetes #1533
Conversation
09eb31b
to
fe58c86
Compare
/retest-required |
1 similar comment
/retest-required |
@danwinship the request/response test you mentioned in your 2nd point...It looks like the latency has increased for handling single UDP packets. I'm looking at https://bugzilla.redhat.com/show_bug.cgi?id=2085089#c37 However that is the comment I think where the units are wrong, and now we are focusing on the number of request/response in a minute. What I don't understand is if the latency is faster now that we know the real unit is microseconds, how is it possible that the number of request/responses are so much worse? @mffiedler do you agree with re-enabling this now? |
Because "the number of request/responses" is just 1 second divided by the latency of a single request/response. The units don't matter; if you make a single request/response take twice as long, then the total number of serialized requests/responses you can do in a fixed amount of time will be halved. (Eg, for 64-byte packets, the test reports 76us latency before and 137us latency after, with 14,447 round trips before and 7,574 after, which is within 10% of what you get if you just divide 1s/76us and 1s/137us.) But the relevant questions are:
|
/hold Ben suggests adding a "chicken flag" to disable this in case of emergency |
fe58c86
to
f1dbb51
Compare
/retest |
/hold cancel |
/lgtm |
Always return a nil OVNConfigBootstrapResult on error (since the caller ignores it anyway). If there is no dpu-mode-config override configmap, don't log a message; that's the normal expected case. If dpu-mode-config parsing fails, don't bail out of bootstrapOVNConfig completely. Just skip overriding the NodeMode.
f1dbb51
to
b485074
Compare
rebased and updated the new unit test to include |
/lgtm |
[APPROVALNOTIFIER] This PR is APPROVED This pull-request has been approved by: danwinship, trozet The full list of commands accepted by this bot can be found here. The pull request process is described here
Needs approval from an approver in each of these files:
Approvers can indicate their approval by writing |
/retest-required |
@danwinship: The following tests failed, say
Full PR test history. Your PR dashboard. Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository. I understand the commands that are listed here. |
/retest-required |
Re-push of #1489; that got reverted (#1510) because QE had reported that it caused performance regressions. However
No one has been able to suggest a real-world use case that would see the sort of drastic slowdowns seen in the "Request/Response operations" test:
etc
Additionally, testing a protocol like this inside a kubernetes clusters is an implausibly-"best" best case scenario; in most normal environments, the client and server would be farther apart from each other (in terms of network topology), and thus a protocol like the "Request/Response operations" test that was extremely latency-sensitive would have much worse behavior, making it even more likely that the developers would change the way it worked.
So, this PR re-enables the feature.