detect vendor (hpe), cache token per host, reuse token

python/redfish-api/redfish_exporter.py

@@ -25,6 +25,11 @@ class HostConfig:
     failures: int = 0
     next_retry_time: float = field(default=0.0, init=False)
 
+    # New attributes for Redfish stuff
+    vendor: str | None = None
+    session_token: str | None = None
+    token_expiry = float | None = None
+
     def should_skip(self) -> bool:
         """Check if host is still in cool-down window"""
         return time.monotonic() < self.next_retry_time
@@ -86,20 +91,70 @@ async def fetch_with_retry(session, host: HostConfig, url: str) -> dict | None:
         up_gauge.labels(host=host.fqdn).set(0)
         return None
 
+    if not host.vendor:
+        try:
+            async with session.get(f"https://{host.fqdn}/redfish/v1/", ssl=False, timeout=10) as resp:
+                if resp.status == 200:
+                    data = await resp.json()
+                    host.vendor = data.get("Vendor", "")
+                    logging.debug("Detected vendor for %s: %s", host.fqdn, host.vendor)
+                else:
+                    logging.warning("Vendor probe failed on %s: HTTP %s", host.fqdn, resp.status)
+        except Exception as e:
+            logging.warning("Vendor probe failed for %s: %s", host.fqdn, e)
+    
+    is_hpe = host.vendor and host.vendor.strip().upper().startswith("HPE")
+
     for attempt in range(1, host.max_retries + 1):
         try:
-            async with session.get(
-                url,
-                auth=aiohttp.BasicAuth(host.username, host.password),
-                ssl=False,
-                timeout=10,
-            ) as resp:
-                if resp.status == 200:
-                    host.mark_success()
-                    return await resp.json()
-                logging.warning(
-                    "HTTP %s from %s (attempt %d)", resp.status, host.fqdn, attempt
-                )
+            headers = {}
+
+            if is_hpe:
+                # Try to reuse existing session token
+                if host.session_token:
+                    headers["X-Auth-Token"] = host.session_token
+                    logging.debug("Reusing cached session token for %s", host.fqdn)
+                else:
+                    # Need to login and store new session token
+                    # HPE Redfish login
+                    login_url = f"https://{host.fqdn}/redfish/v1/SessionService/Sessions"
+                    payload = {"UserName": host.username, "Password": host.password}
+                    async with session.post(login_url, json=payload, ssl=False, timeout=10) as login_resp:
+                        if login_resp.status == 201:
+                            host.session_token = login_resp.headers.get("X-Auth-Token")
+                            if not host.session_token:
+                                raise RuntimeError("No X-Auth-Token in login response")
+                            headers["X-Auth-Token"] = host.session_token
+                            logging.info("New session token obtained for %s", host.fqdn)
+                        else:
+                            logging.warning("Login failed for %s: HTTP %s", host.fqdn, login_resp.status)
+                            continue  # retry login next attempt
+
+                async with session.get(url, headers=headers, ssl=False, timeout=10) as resp:
+                    if resp.status == 200:
+                        host.mark_success()
+                        return await resp.json()
+                    elif resp.status in (401, 403):
+                        # Token expired or invalid, clear it and retry
+                        logging.warning("Invalid token for %s, reauthenticating...", host.fqdn)
+                        host.session_token = None
+                        continue
+                    logging.warning("HTTP %s from %s (attempt %d)", resp.status, host.fqdn, attempt)
+
+            else:
+                # Default: BasicAuth, like Supermicro and so
+                async with session.get(
+                    url,
+                    auth=aiohttp.BasicAuth(host.username, host.password),
+                    ssl=False,
+                    timeout=10,
+                ) as resp:
+                    if resp.status == 200:
+                        host.mark_success()
+                        return await resp.json()
+                    logging.warning(
+                        "HTTP %s from %s (attempt %d)", resp.status, host.fqdn, attempt
+                    )
 
         except asyncio.TimeoutError:
             logging.warning("Timeout on %s (attempt %d)", host.fqdn, attempt)