c-ares 1.22.1 (and newer) does not change servers on a request timeout

[b-spencer]

If I understand correctly, with the new dynamic server list behaviour of #594, the first server in the server list is always used by ares__requeue_query(). But, when a query times out in process_timeouts(), there is no call to server_increment_failure(), so the server list always stays in the same order, and thus the retry is send to the same server.

This means that when a DNS server is not responding, c-ares gets stuck querying it forever, at least in all the cases I've seen. Real users often have misconfigurations such as non-functional IPv6 DNS server entries, etc. If a "bad" server is first in the list, no DNS lookups work at all.

All other (similar) calls to ares__requeue_query() are preceded by a call to server_increment_failure(). Adding that call in process_timeouts() immediately before ares__requeue_query() locally fixed the behaviour in my testing so far.

[bradh352]

Hmm, that seems really bad if true. I thought the test cases handled the retry scenarios and would switch servers. Let me investigate.

[bradh352]

ok, confirmed. Created PR #651 which also adds a test case for this. Once all CI/CD tests complete I'll merge into main as well as the 1.23 branch.