fix(http): contain DNS retry-budget widening to DNS path only

PR #382 introduced an `effective_retries` widening on the first gaierror,
but the widening leaked: every non-DNS error path (HTTPError, non-DNS
URLError, OSError) was gated on `effective_retries - 1` and so inherited
the expanded bound. A caller passing `retries=2` who hit DNS-then-non-DNS
got 3 attempts instead of 2 — contrary to the PR description and the
fail-fast intent of small retry budgets.

Fix:
- Gate every non-DNS sleep/retry decision on the caller's original
  `retries`, not the widened `effective_retries`.
- Add an explicit `break` in each non-DNS branch when the original
  budget is exhausted, so the (possibly widened) outer loop bound
  can't pull us into an extra attempt.

Adds two regression tests covering the DNS-then-non-DNS-URLError and
DNS-then-OSError sequences flagged in Greptile review on PR #382.
This commit is contained in:
Trevin Chow
2026-05-16 21:13:17 -07:00
parent 5a2fe5279b
commit 719cdef2fb
2 changed files with 60 additions and 4 deletions
+22 -4
View File
@@ -132,7 +132,9 @@ def request(
if rate_limit_count >= max_429_retries:
raise last_error
if attempt < effective_retries - 1:
# HTTP errors respect the caller's original `retries`; only DNS
# failures get the widened `effective_retries` budget.
if attempt < retries - 1:
if e.code == 429:
# Respect Retry-After header, fall back to exponential backoff
retry_after = e.headers.get("Retry-After") if hasattr(e, 'headers') else None
@@ -143,10 +145,15 @@ def request(
delay = RETRY_DELAY * (2 ** attempt) + 1
else:
delay = RETRY_DELAY * (2 ** attempt) + 1 # 3s, 5s, 9s...
log(f"Rate limited (429). Waiting {delay:.1f}s before retry {attempt + 2}/{effective_retries}")
log(f"Rate limited (429). Waiting {delay:.1f}s before retry {attempt + 2}/{retries}")
else:
delay = RETRY_DELAY * (2 ** attempt)
time.sleep(delay)
else:
# Caller's original retry budget exhausted; an earlier DNS
# failure may have widened `effective_retries`, but that
# widening is DNS-only — don't grant extra HTTP attempts.
break
except urllib.error.URLError as e:
log(f"URL Error: {e.reason}")
last_error = HTTPError(f"URL Error: {e.reason}")
@@ -170,8 +177,15 @@ def request(
f"retrying in {delay:.1f}s"
)
time.sleep(delay)
elif attempt < effective_retries - 1:
elif attempt < retries - 1:
# Non-DNS URLError (e.g. ConnectionRefused) respects the
# caller's original retry budget, not the DNS-widened bound.
time.sleep(RETRY_DELAY * (attempt + 1))
else:
# Caller's original retry budget exhausted; an earlier DNS
# failure widening `effective_retries` does not carry over
# to non-DNS error paths.
break
except json.JSONDecodeError as e:
log(f"JSON decode error: {e}")
last_error = HTTPError(f"Invalid JSON response: {e}")
@@ -180,8 +194,12 @@ def request(
# Handle socket-level errors (connection reset, timeout, etc.)
log(f"Connection error: {type(e).__name__}: {e}")
last_error = HTTPError(f"Connection error: {type(e).__name__}: {e}")
if attempt < effective_retries - 1:
if attempt < retries - 1:
# Socket errors respect the caller's original retry budget.
time.sleep(RETRY_DELAY * (attempt + 1))
else:
# Original budget exhausted; DNS widening doesn't apply here.
break
attempt += 1
+38
View File
@@ -180,3 +180,41 @@ class TestDNSResolutionRetry(unittest.TestCase):
# Caller passed retries=2, and non-DNS URLError doesn't expand it.
self.assertEqual(mock_urlopen.call_count, 2)
@patch("lib.http.urllib.request.urlopen")
@patch("lib.http.time.sleep")
def test_dns_widening_does_not_leak_into_subsequent_non_dns_urlerror(
self, mock_sleep, mock_urlopen
):
"""Mixed sequence: DNS-then-non-DNS must respect caller's original retries.
Without the fix, the first gaierror widens effective_retries from 2 to
MIN_DNS_RETRIES=3, and a subsequent ConnectionRefused on attempt 1
slips into a third overall attempt — exceeding what the caller asked
for. Each non-DNS error path must gate on the original `retries`.
"""
import socket
dns_err = urllib.error.URLError(socket.gaierror(-2, "Name or service not known"))
conn_err = urllib.error.URLError(ConnectionRefusedError(111, "Connection refused"))
mock_urlopen.side_effect = [dns_err, conn_err, conn_err] # 3rd would only fire if budget leaked
with self.assertRaises(http.HTTPError):
http.request("GET", "http://flaky.example", retries=2)
# Caller asked for at most 2 attempts. DNS widening must not give us a 3rd.
self.assertEqual(mock_urlopen.call_count, 2)
@patch("lib.http.urllib.request.urlopen")
@patch("lib.http.time.sleep")
def test_dns_widening_does_not_leak_into_subsequent_oserror(
self, mock_sleep, mock_urlopen
):
"""Mixed sequence: DNS-then-OSError must respect caller's original retries."""
import socket
dns_err = urllib.error.URLError(socket.gaierror(-2, "Name or service not known"))
mock_urlopen.side_effect = [dns_err, TimeoutError("timed out"), TimeoutError("timed out")]
with self.assertRaises(http.HTTPError):
http.request("GET", "http://flaky.example", retries=2)
self.assertEqual(mock_urlopen.call_count, 2)