feat(client): re-resolve the server hostnames at the DNS TTL

hbc, hbc_mini.py and hbc_mini.c resolved the server once at startup, so an address change needed a client restart. They now re-resolve every TTL seconds (smallest A/AAAA TTL, floor 30s, 300s when unknown) and re-point, add or drop connections to follow the new addresses; a failed lookup keeps the current ones. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
2026-10-28 04:23:55 -04:00
parent e8d9c4a958
commit 885d5dc937
9 changed files with 887 additions and 1 deletions
@@ -0,0 +1,119 @@
+"""TTL-driven re-resolution of the server hostnames in the full client."""
+import socket
+import struct
+
+from hbd.client import dnsttl
+
+
+def _rr(rtype, ttl, rdata):
+    return b"\xc0\x0c" + struct.pack("!HHIH", rtype, 1, ttl, len(rdata)) + rdata
+
+
+def _response(qid, qtype, answers, rcode=0):
+    q = dnsttl.build_query("hbd.example.com", qtype, qid)[12:]
+    return struct.pack("!HHHHHH", qid, 0x8180 | rcode, 1, len(answers), 0, 0) + q + b"".join(answers)
+
+
+def test_min_ttl_covers_cname_and_address():
+    resp = _response(7, dnsttl.TYPE_A, [_rr(dnsttl.TYPE_CNAME, 60, b"\x01x\x00"),
+                                        _rr(dnsttl.TYPE_A, 300, b"\x01\x02\x03\x04")])
+    assert dnsttl.min_answer_ttl(resp, 7, dnsttl.TYPE_A) == 60
+
+
+def test_no_answer_of_requested_type_is_unknown():
+    assert dnsttl.min_answer_ttl(_response(7, dnsttl.TYPE_AAAA, []), 7, dnsttl.TYPE_AAAA) is None
+
+
+def test_error_rcode_and_foreign_id_are_unknown():
+    ans = [_rr(dnsttl.TYPE_A, 300, b"\x01\x02\x03\x04")]
+    assert dnsttl.min_answer_ttl(_response(7, 1, ans, rcode=3), 7, 1) is None
+    assert dnsttl.min_answer_ttl(_response(7, 1, ans), 8, 1) is None
+
+
+def test_truncated_response_is_unknown():
+    assert dnsttl.min_answer_ttl(b"\x00\x07\x81", 7, 1) is None
+
+
+def test_refresh_delay():
+    assert dnsttl.refresh_delay([300, 120, None]) == 120
+    assert dnsttl.refresh_delay([0]) == dnsttl.MIN_TTL
+    assert dnsttl.refresh_delay([None]) == dnsttl.DEFAULT_TTL
+
+
+def test_ip_literal_has_no_ttl():
+    assert dnsttl.lookup_ttl("192.0.2.1") is None
+    assert dnsttl.lookup_ttl("2001:db8::1") is None
+
+
+class _C:
+    def __init__(self, addr, af=socket.AF_INET):
+        self.addr, self.af = addr, af
+
+    def set_target(self, addr, af):
+        self.addr, self.af = addr, af
+
+
+V4, V6 = socket.AF_INET, socket.AF_INET6
+
+
+def test_retarget_follows_address_change_in_place():
+    first, second = _C("1.1.1.1"), _C("2001:db8::1", V6)
+    new, gone = dnsttl.retarget([first, second], [(V4, "9.9.9.9"), (V6, "2001:db8::9")])
+    assert (new, gone) == ([], [])
+    assert (first.addr, second.addr) == ("9.9.9.9", "2001:db8::9")
+
+
+def test_retarget_unchanged_is_a_noop():
+    conns = [_C("1.1.1.1")]
+    assert dnsttl.retarget(conns, [(V4, "1.1.1.1")]) == ([], [])
+
+
+def test_retarget_reports_surplus_new_and_gone():
+    a, b = _C("1.1.1.1"), _C("2.2.2.2")
+    new, gone = dnsttl.retarget([a, b], [(V4, "3.3.3.3")])
+    assert new == [] and gone == [b] and a.addr == "3.3.3.3"
+    new, gone = dnsttl.retarget([a], [(V4, "3.3.3.3"), (V4, "4.4.4.4")])
+    assert new == [(V4, "4.4.4.4")] and gone == []
+
+
+def test_retarget_never_drops_the_first_connection():
+    a, b = _C("1.1.1.1"), _C("2.2.2.2")
+    new, gone = dnsttl.retarget([a, b], [(V4, "2.2.2.2")])
+    assert new == [] and gone == [b] and a.addr == "2.2.2.2"
+
+
+def test_resolve_failure_returns_none(monkeypatch):
+    def boom(*a):
+        raise socket.gaierror("nope")
+    monkeypatch.setattr(socket, "getaddrinfo", boom)
+    assert dnsttl.resolve(["x.invalid"], 1, 0) is None
+
+
+def test_refresher_follows_a_change_and_adds_a_family(monkeypatch):
+    import asyncio
+    from hbd.client import main as hbc
+
+    answers = [[(V4, "9.9.9.9"), (V6, "2001:db8::9")], None, [(V4, "9.9.9.9")]]
+    monkeypatch.setattr(dnsttl, "resolve", lambda *a: answers.pop(0) if len(answers) > 1 else answers[0])
+    monkeypatch.setattr(dnsttl, "lookup_ttl", lambda *a: None)
+    monkeypatch.setattr(dnsttl, "DEFAULT_TTL", 0.01)
+    monkeypatch.setattr(dnsttl, "FAILED_RETRY", 0.01)
+
+    async def idle(*a):
+        await asyncio.sleep(10)
+    monkeypatch.setattr(hbc, "heartbeat_sender", idle)
+
+    async def go():
+        hbc.shutdown_event = asyncio.Event()
+        hbc.active_tasks = []
+        conns = [hbc.AsyncConnection(1, "1.1.1.1", 5, V4, "me")]
+        task = asyncio.create_task(hbc.dns_refresher(conns, ["h"], 5, 0, "me", 10, None, 2))
+        await asyncio.sleep(0.2)
+        hbc.shutdown_event.set()
+        await task
+        for t in hbc.active_tasks:
+            t.cancel()
+        return conns
+
+    conns = asyncio.run(go())
+    assert [(c.conn_id, c.addr) for c in conns] == [(1, "9.9.9.9")]
@@ -0,0 +1,72 @@
+"""TTL-driven re-resolution in the standalone scripts/hbc_mini.py."""
+import asyncio
+import importlib.util
+import pathlib
+import socket
+import struct
+import sys
+
+_PATH = pathlib.Path(__file__).resolve().parents[1] / "scripts" / "hbc_mini.py"
+_SPEC = importlib.util.spec_from_file_location("hbc_mini", _PATH)
+mini = importlib.util.module_from_spec(_SPEC)
+sys.modules["hbc_mini"] = mini
+_SPEC.loader.exec_module(mini)
+
+V4, V6 = socket.AF_INET, socket.AF_INET6
+
+
+def _rr(rtype, ttl, rdata):
+    return b"\xc0\x0c" + struct.pack("!HHIH", rtype, 1, ttl, len(rdata)) + rdata
+
+
+def _response(qid, qtype, answers, rcode=0):
+    q = mini._dns_build_query("hbd.example.com", qtype, qid)[12:]
+    return struct.pack("!HHHHHH", qid, 0x8180 | rcode, 1, len(answers), 0, 0) + q + b"".join(answers)
+
+
+def test_min_ttl_covers_cname_and_address():
+    resp = _response(7, 1, [_rr(5, 60, b"\x01x\x00"), _rr(1, 300, b"\x01\x02\x03\x04")])
+    assert mini._dns_min_answer_ttl(resp, 7, 1) == 60
+
+
+def test_unknown_ttl_cases():
+    assert mini._dns_min_answer_ttl(_response(7, 28, []), 7, 28) is None
+    assert mini._dns_min_answer_ttl(_response(7, 1, [_rr(1, 9, b"1234")], rcode=3), 7, 1) is None
+    assert mini._dns_min_answer_ttl(b"\x00\x07\x81", 7, 1) is None
+    assert mini._dns_lookup_ttl("192.0.2.1") is None
+
+
+def test_refresh_delay():
+    assert mini._dns_refresh_delay([300, 120, None]) == 120
+    assert mini._dns_refresh_delay([0]) == mini._DNS_MIN_TTL
+    assert mini._dns_refresh_delay([None]) == mini._DNS_DEFAULT_TTL
+
+
+def test_refresher_follows_a_change_and_adds_a_family(monkeypatch):
+    answers = [[(V4, "9.9.9.9"), (V6, "2001:db8::9")], None, [(V4, "9.9.9.9")]]
+    monkeypatch.setattr(mini, "_dns_resolve", lambda *a: answers.pop(0) if len(answers) > 1 else answers[0])
+    monkeypatch.setattr(mini, "_dns_lookup_ttl", lambda *a: None)
+    monkeypatch.setattr(mini, "_DNS_DEFAULT_TTL", 0.01)
+    monkeypatch.setattr(mini, "_DNS_FAILED_RETRY", 0.01)
+
+    async def idle(*a):
+        await asyncio.sleep(10)
+    monkeypatch.setattr(mini, "_heartbeat_sender", idle)
+
+    async def go():
+        mini._running = True
+        mini._shutdown_event = asyncio.Event()
+        mini._active_tasks = []
+        conns = [mini.AsyncConnection(1, "1.1.1.1", 5, V4, "me")]
+        task = asyncio.create_task(mini._dns_refresher(conns, ["h"], 5, 0, "me", 10, None, 2))
+        await asyncio.sleep(0.2)
+        mini._running = False
+        mini._shutdown_event.set()
+        await task
+        for t in mini._active_tasks:
+            t.cancel()
+        return conns
+
+    conns = asyncio.run(go())
+    assert [(c.conn_id, c.addr) for c in conns] == [(1, "9.9.9.9")]
+    assert conns[0].af == V4