From 7817bd8391d68dd74e0778cee0189784e747bc12 Mon Sep 17 00:00:00 2001
From: Matthieu Baerts <matttbe@kernel.org>
Date: Fri, 23 Oct 2026 10:21:27 +0200
Subject: [PATCH 1/2] contest: remote: fetcher: retry on connect errors

Recently, after a reboot, most vmksft services failed to start because
they failed to resolve the hostname.

Allow them to retry with an exponential backoff, similar to what is done
with Patchwork.

While at it, switch to 'urllib3' instead of 'requests', and use a
'PoolManager', as recommended on urllib3 doc, to handle both the timeout
and the retries settings. We just need to change the response to get
'data' instead of 'content'. Also catch HTTP errors and issues to decode
the JSON files.

Signed-off-by: Matthieu Baerts <matttbe@kernel.org>
---
 contest/remote/lib/fetcher.py | 19 +++++++++++--------
 1 file changed, 11 insertions(+), 8 deletions(-)

diff --git a/contest/remote/lib/fetcher.py b/contest/remote/lib/fetcher.py
index d41ab209..e8cf4727 100644
--- a/contest/remote/lib/fetcher.py
+++ b/contest/remote/lib/fetcher.py
@@ -4,9 +4,10 @@
 import json
 import os
 import re
-import requests
 import subprocess
 import time
+import urllib3
+from urllib3.util.retry import Retry
 
 
 class Fetcher:
@@ -26,6 +27,9 @@ def __init__(self, cb, cbarg, name, branches_url, results_path, url_path, tree_p
         self._tree_path = tree_path
         self._patches_path = patches_path
 
+        self._http = urllib3.PoolManager(timeout=30,
+                                         retries=Retry(connect=10, backoff_factor=1))
+
         # Set last date to something old
         self._last_date = datetime.datetime.now(datetime.UTC) - datetime.timedelta(weeks=1)
         if first_run == "force":
@@ -33,8 +37,8 @@ def __init__(self, cb, cbarg, name, branches_url, results_path, url_path, tree_p
             pass
         elif first_run == "continue":
             try:
-                r = requests.get(self._branches_url, timeout=30)
-                branches = json.loads(r.content.decode('utf-8'))
+                r = self._http.request('GET', self._branches_url)
+                branches = json.loads(r.data.decode('utf-8'))
                 branch_date = {}
                 for b in branches:
                     branch_date[b["branch"]] = datetime.datetime.fromisoformat(b["date"])
@@ -49,7 +53,7 @@ def __init__(self, cb, cbarg, name, branches_url, results_path, url_path, tree_p
 
                     self._last_date = max(branch_date[result["branch"]], self._last_date)
                 print("INFO: Last run date:", self._last_date)
-            except FileNotFoundError:
+            except (FileNotFoundError, urllib3.exceptions.HTTPError, ValueError):
                 pass
         elif first_run == "next":
             # unless there's a crazy race or time error this will skip newest branch
@@ -134,13 +138,12 @@ def _find_branch(self, name):
 
     def _run_once(self):
         try:
-            r = requests.get(self._branches_url, timeout=30)
-        except requests.exceptions.RequestException as e:
+            r = self._http.request('GET', self._branches_url)
+            branches = json.loads(r.data.decode('utf-8'))
+        except (urllib3.exceptions.HTTPError, ValueError) as e:
             print(f'WARN: Failed to fetch branches: {e}')
             return
 
-        branches = json.loads(r.content.decode('utf-8'))
-
         to_test = None
         newest = self._last_date
 

From e29890647fb513511c71f32f7169869f94f90a34 Mon Sep 17 00:00:00 2001
From: Matthieu Baerts <matttbe@kernel.org>
Date: Fri, 23 Oct 2026 10:29:32 +0200
Subject: [PATCH 2/2] contest: remote: vm: retry on connect errors

Similar to the parent commit, retry with an exponential backoff in case
of connection issues.

Also switch to 'urllib3' instead of 'requests' if it is to use Retry
from there, and catch errors instead of crashing: that's OK if we don't
have the filters there.

Signed-off-by: Matthieu Baerts <matttbe@kernel.org>
---
 contest/remote/lib/vm.py | 14 +++++++++++---
 1 file changed, 11 insertions(+), 3 deletions(-)

diff --git a/contest/remote/lib/vm.py b/contest/remote/lib/vm.py
index e2c1d3df..d754f32b 100644
--- a/contest/remote/lib/vm.py
+++ b/contest/remote/lib/vm.py
@@ -1,7 +1,6 @@
 # SPDX-License-Identifier: GPL-2.0
 
 import unicodedata
-import requests
 import subprocess
 import fcntl
 import json
@@ -11,6 +10,8 @@
 import shutil
 import signal
 import time
+import urllib3
+from urllib3.util.retry import Retry
 from .crash import has_crash, extract_crash
 
 
@@ -68,6 +69,9 @@ def __init__(self, config, vm_name=""):
 
         self.cfg_boot_to = int(config.get('vm', 'boot_timeout'))
 
+        self._http = urllib3.PoolManager(timeout=30,
+                                         retries=Retry(connect=10, backoff_factor=1))
+
         self.filter_data = None
         self.has_kmemleak = None
         self.has_gcov = self.config.getboolean('vm', 'gcov', fallback=False)
@@ -384,8 +388,12 @@ def _load_filters(self):
             url = self.config.get("remote", "filters", fallback=None)
             if not url:
                 return None
-            r = requests.get(url)
-            self.filter_data = json.loads(r.content.decode('utf-8'))
+            try:
+                r = self._http.request('GET', url)
+                self.filter_data = json.loads(r.data.decode('utf-8'))
+            except (urllib3.exceptions.HTTPError, ValueError) as e:
+                print(f"WARN{self.print_pfx} failed to fetch filters: {e}")
+                return None
         return self.filter_data
 
     def extract_crash(self, out_path):