From 278c0c03b375ffa144d68e0a07f442d079c34028 Mon Sep 17 00:00:00 2001
From: scott <scott@sdgarren.com>
Date: Thu, 16 Apr 2026 13:17:53 -0400
Subject: [PATCH] Wait for IOM to come back online after restart instead of
 fixed sleep

A fixed 30s sleep is not enough for a full BMC boot cycle. Add
_wait_for_iom_online() which polls GET /redfish/v1/ every 15s until the
IOM responds (up to 5 minutes), then call it after every IOM restart in
both _update_iom_fw and _update_fabric_fw. This ensures the fabric card
update (and post-update validation) don't run while the IOM is still
booting and unreachable.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---
 modules/redfish.py           | 19 +++++++++++++++++++
 modules/workflow_firmware.py | 17 +++++++++++++----
 2 files changed, 32 insertions(+), 4 deletions(-)

diff --git a/modules/redfish.py b/modules/redfish.py
index c4e5fb6..1eada80 100644
--- a/modules/redfish.py
+++ b/modules/redfish.py
@@ -181,6 +181,25 @@ def _redfish_poll_tasks(password: str, host: str, timeout: int = 600) -> tuple:
     return False, f"Timeout after {timeout}s waiting for tasks."
 
 
+def _wait_for_iom_online(password: str, host: str, timeout: int = 300) -> bool:
+    """
+    Poll GET /redfish/v1/ until the IOM responds successfully, indicating it
+    has finished rebooting.  Waits up to `timeout` seconds.
+    Returns True if the IOM came back online, False if timeout was exceeded.
+    """
+    deadline = time.monotonic() + timeout
+    attempt  = 0
+    while time.monotonic() < deadline:
+        attempt += 1
+        ok_flag, _ = _redfish_request(password, "GET", "/redfish/v1/", host=host)
+        if ok_flag:
+            return True
+        elapsed = int(time.monotonic() - (deadline - timeout))
+        info(f"  Waiting for IOM to come back online... [{elapsed}s elapsed]")
+        time.sleep(15)
+    return False
+
+
 def _redfish_restart_iom(password: str, host: str, iom: str) -> tuple:
     return _redfish_request(
         password, "POST",
diff --git a/modules/workflow_firmware.py b/modules/workflow_firmware.py
index dc1ad6a..6218923 100644
--- a/modules/workflow_firmware.py
+++ b/modules/workflow_firmware.py
@@ -12,6 +12,7 @@ from redfish import (
     _redfish_poll_tasks,
     _redfish_restart_iom,
     _redfish_reset_fabric,
+    _wait_for_iom_online,
     _show_fw_versions,
 )
 from ui import (
@@ -161,8 +162,12 @@ def _update_iom_fw(password: str, ip: str, iom: str, fw_path: str) -> bool:
 
     info(f"Restarting {iom}...")
     _redfish_restart_iom(password, ip, iom)   # connection drop on restart is normal
-    ok(f"{iom} restart initiated. Waiting 30s for reboot...")
-    time.sleep(30)
+    ok(f"{iom} restart initiated. Waiting for IOM to come back online...")
+    time.sleep(30)  # allow time for the IOM to begin shutting down before polling
+    if _wait_for_iom_online(password, ip):
+        ok(f"{iom} is back online.")
+    else:
+        warn(f"{iom} did not respond within 5 minutes — proceeding anyway.")
     return True
 
 
@@ -204,8 +209,12 @@ def _update_fabric_fw(password: str, ip: str, iom: str, fw_path: str) -> bool:
 
     info(f"Restarting {iom} after Fabric Card update...")
     _redfish_restart_iom(password, ip, iom)
-    ok(f"{iom} restart initiated. Waiting 30s for reboot...")
-    time.sleep(30)
+    ok(f"{iom} restart initiated. Waiting for IOM to come back online...")
+    time.sleep(30)  # allow time for the IOM to begin shutting down before polling
+    if _wait_for_iom_online(password, ip):
+        ok(f"{iom} is back online.")
+    else:
+        warn(f"{iom} did not respond within 5 minutes — proceeding anyway.")
     return True