Cyrille BéraudandCopilot App e9208cfbe4 http: bound name resolution and reuse the endpoints it found ...

Request::send waits for Resolver before doing anything else, and the
timeout a caller sets through Request::timeout is only armed later, in
connect. A system resolver that never answers therefore holds the request
for good: async_resolve has no deadline of its own, no timer covers the
wait, and the callbacks queued in Resolver are never invoked. The proxy
client reaches that path on every connectivity change, since
connectivityChanged calls getProxyInfos, so a device coming back from
sleep can sit there with no status, no listener and no error.

getProxyInfos also builds a new Resolver each time and drops the previous
one, discarding endpoints that were still perfectly usable. Resolution is
then redone from scratch at the exact moment the network stack has just
come up and is at its slowest.

Give Resolver a deadline of its own, armed as soon as a query starts, and
report timed_out to whoever is waiting. Let it also start from endpoints
that were already resolved: those answer callbacks immediately so a
connection begins without waiting, while the query still runs and
replaces them as soon as it returns. An address that went stale in the
meantime now costs one connection attempt, which Request retries against
the pending query rather than reporting as a failure.

A failed query no longer clears endpoints that are known to work, so a
transient DNS outage leaves the client with something to connect to.

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

2026-09-03 07:50:51 -04:00

..

opendht

http: bound name resolution and reuse the endpoints it found

2026-09-03 07:50:51 -04:00

opendht.h

Switch project license from GPL-3.0-or-later to MIT

2026-01-17 22:25:43 -05:00