In our CI, we are seeing long delays (~8 seconds in #16237 and #14959) between an endpoint being created and the corresponding IPCache update on the remote node. A potential cause for the increased delay might be due to the changes in #15230.

As a quick workaround, we are increasing the connect timeout in CI (#16301), but we need to discuss if we deem such long propagation delays acceptable and what we can do about them.