Right now, we never do anything interesting with stale cache entries. There are some cool things we could do:
1. Let Cronet embedders specify that they're okay with using stale entries after a delay, to reduce long-tail latency.
2. When asked for a TCP connection to a hostname with a stale DNS entry in cache, speculatively connect to the stale address(es) while we re-request the DNS entry. If the address we connected to is still listed in the entry, keep it and enjoy lower connection latency. If not, retry with the addresses in the new entry.
Maybe other things, too.
Comment 1 by bugdroid1@chromium.org
, May 12 2016