bryancall opened a new pull request, #13875: URL: https://github.com/apache/trafficserver/pull/13875
The `cookie_remap` tests are the only microserver tests that bind their origins to 127.0.0.10 and 127.0.0.11. Python's `HTTPServer` calls `socket.getfqdn()` on the bind address while it starts, and those addresses are not in `/etc/hosts`, so every origin start sends a reverse DNS query. When the resolver drops that query, glibc retries after 5 seconds and gives up after 10, which is exactly AuTest's default startup timeout. AuTest then interrupts the origin before it is serving and the test fails on the origin's return code. 127.0.0.1 resolves from `/etc/hosts` without a query, which is why no other test hits this. This binds the origins in all twelve tests to 127.0.0.1 and points their configs there. The two origins in each test already listen on different ports, and no gold file checks the address. Testing, in the fedora:44 CI image under CPU load with a resolver that drops about 30% of queries: master failed 108 of 768 test runs, every one with the same "Checking that any test passes" failure CI has been reporting, and this change failed none of 768. With the resolver not answering at all, master failed 96 of 96 and this change none. With a healthy resolver, master passed all 1152 runs, so CPU load alone does not cause it. I can't see the resolver on the CI agents, so the dropped query there is inferred from the matching signature rather than observed. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
