From ac8227eead60d05213059680402255f5c8b8cde9 Mon Sep 17 00:00:00 2001
From: tegwick <bernd.worsch@gmail.com>
Date: Tue, 6 Oct 2026 00:26:09 +0200
Subject: [PATCH] Retry one dropped OpenBao request and report the last intake
 in health.

The bridge forward was down 2026-10-05 20:30 to about 00:20 and dropped
the first request after reconnecting; 13 timer runs reported
credential_unavailable without health showing it.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Assistant: claude-code
Assistant-Model: opus
Assistant-Process: 7780@bnt-lap001
Assistant-Session: 622cd705-cb31-4322-ba08-dd42e44fa583
---
 inbound_library/__main__.py |  3 +++
 inbound_library/notify.py   |  5 ++++-
 inbound_library/vault.py    | 19 +++++++++++++++++--
 tests/test_vault.py         | 19 +++++++++++++++++++
 4 files changed, 43 insertions(+), 3 deletions(-)

diff --git a/inbound_library/__main__.py b/inbound_library/__main__.py
index a093a56..77c4fd9 100644
--- a/inbound_library/__main__.py
+++ b/inbound_library/__main__.py
@@ -2,6 +2,7 @@
 import argparse
 import fcntl
 import json
+from datetime import datetime, timezone
 import sys
 from pathlib import Path
 
@@ -46,6 +47,8 @@ def run_due(root, config, mailbox=None, post=notify.post_message):
         except (OSError, ValueError) as exc:
             report['intake'] = {'ok': False, 'status': 'error', 'error': str(exc)}
         report['ok'] = report['intake']['ok']
+        store.atomic(state / 'last-intake.json', json.dumps(
+            {**report['intake'], 'at': datetime.now(timezone.utc).isoformat()}, indent=2).encode() + b'\n')
         report['requeued'] = len(intake.requeue_quarantine(root, registry))
         report['notices'] = len(notify.match(root, registry))
         hub = config.get('hub', {})
diff --git a/inbound_library/notify.py b/inbound_library/notify.py
index 02fe42f..ac0ffdb 100644
--- a/inbound_library/notify.py
+++ b/inbound_library/notify.py
@@ -172,7 +172,10 @@ def health(root, now=None, registry=None):
     cursor_path = Path(root) / 'data' / 'state' / 'imap-cursor.json'
     cursor = json.loads(cursor_path.read_text()) if cursor_path.exists() else {}
     pending, _ = pending_hub(root, registry)
-    return {'last_scan': cursor.get('scanned_at'), 'items_per_source': counts,
+    last_path = Path(root) / 'data' / 'state' / 'last-intake.json'
+    last_intake = json.loads(last_path.read_text()) if last_path.exists() else None
+    return {'last_scan': cursor.get('scanned_at'), 'last_intake': last_intake,
+            'items_per_source': counts,
             'silent_sources': silent, 'quarantined_items': len(quarantine),
             'pending_hub_notices': sum(len(v) for v in pending.values()),
             'integrity_problems': store.verify(root)}
diff --git a/inbound_library/vault.py b/inbound_library/vault.py
index 7b89ffb..bee567f 100644
--- a/inbound_library/vault.py
+++ b/inbound_library/vault.py
@@ -15,7 +15,22 @@ class CredentialUnavailable(Exception):
     """The credential could not be obtained. The message never contains a secret."""
 
 
-def _call(addr, method, path, token=None, body=None, timeout=15):
+def _call(addr, method, path, token=None, body=None, timeout=15, attempts=2):
+    """One OpenBao HTTP call. A connection error is retried once, because the
+    SSH bridge forward often drops the first request after it reconnects."""
+    for attempt in range(attempts):
+        try:
+            return _once(addr, method, path, token, body, timeout)
+        except _Unreachable as exc:
+            if attempt + 1 == attempts:
+                raise CredentialUnavailable(str(exc)) from None
+
+
+class _Unreachable(Exception):
+    pass
+
+
+def _once(addr, method, path, token, body, timeout):
     request = urllib.request.Request(
         addr.rstrip('/') + '/v1/' + path.lstrip('/'), method=method,
         data=None if body is None else json.dumps(body).encode(),
@@ -28,7 +43,7 @@ def _call(addr, method, path, token=None, body=None, timeout=15):
         # Status only. The response body could echo request data.
         raise CredentialUnavailable(f'OpenBao {method} {path.split("/")[0]}/… returned {exc.code}') from None
     except (urllib.error.URLError, OSError) as exc:
-        raise CredentialUnavailable(f'OpenBao unreachable at {addr}: {type(exc).__name__}') from None
+        raise _Unreachable(f'OpenBao unreachable at {addr}: {type(exc).__name__}') from None
 
 
 def kv_v2_data_path(kv_path):
diff --git a/tests/test_vault.py b/tests/test_vault.py
index c757ac5..3dc5868 100644
--- a/tests/test_vault.py
+++ b/tests/test_vault.py
@@ -52,6 +52,25 @@ class Vault(unittest.TestCase):
             'auth/approle/login', 'platform/data/workloads/inbound-library/imap',
             'auth/token/revoke-self'])
 
+    def test_dropped_first_request_is_retried_once(self):
+        attempts = []
+
+        def flaky(addr, method, path, token, body, timeout):
+            attempts.append(path)
+            if len(attempts) == 1:
+                raise vault._Unreachable('OpenBao unreachable at x: URLError')
+            return {'ok': True}
+        original = vault._once
+        vault._once = flaky
+        try:
+            self.assertEqual(vault._call('http://bao', 'GET', 'sys/health'), {'ok': True})
+            attempts.clear()
+            vault._once = lambda *a: (_ for _ in ()).throw(vault._Unreachable('down'))
+            with self.assertRaises(vault.CredentialUnavailable):
+                vault._call('http://bao', 'GET', 'sys/health')
+        finally:
+            vault._once = original
+
     def test_denied_read_still_revokes_and_hides_secret(self):
         bao = FakeBao(deny_read=True)
         with self.assertRaises(vault.CredentialUnavailable) as caught: