[CI - FAILURE]: / Handshake_WithConnectionPoolCreateArgs.ConnectionPoolCreate_11 (msquictest.exe)

[github-actions]

https://github.com/microsoft/msquic/actions/runs/29368776697

[github-actions]

https://github.com/microsoft/msquic/actions/runs/29582048663

[guhetier]

Gaurav Singh (@gaurav2699) Anthony Rossi (@anrossi)
This issue seems started to appear (more frequently) since the fix to move the start of the connection to a worker queue.
It seems to be hitting a port reservation more often since #6003 and/or #6114.
It might be the sign of a regression or that some better handling of failure in the connection pool is needed (to handle the case when the port initially select is right before a reserved plage that will cause the following attempts to all fail).

[anrossi]

I wonder if the XSK map changes in #6003 have altered how the RSS port to CPU mapping is done in XDP and if the previous way connection pools calculated their connection spreading is broken.

[guhetier]

Interesting theory. We also have a pending investigation that shows poor repartition of connections on workers with connection pools (although without XDP maps).