From 9380c6391360bb058eec543080e71543eea701bf Mon Sep 17 00:00:00 2001
From: Benjamin Renard <brenard@zionetrix.net>
Date: Wed, 26 Oct 2022 15:08:45 +0200
Subject: [PATCH] Add retry options to retreive device status to avoid false
 positive unreachable device alert

---
 check_esphome_devices | 43 ++++++++++++++++++++++++++++++++++++++++---
 1 file changed, 40 insertions(+), 3 deletions(-)

diff --git a/check_esphome_devices b/check_esphome_devices
index 0c9d66f..a80dd02 100755
--- a/check_esphome_devices
+++ b/check_esphome_devices
@@ -21,6 +21,7 @@ Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA  02111-1307, USA.
 
 import argparse
 import sys
+import time
 import logging
 
 import requests
@@ -35,6 +36,8 @@ STATUS = {
 }
 
 DEFAULT_HOST = 'http://127.0.0.1:6052'
+DEFAULT_RETRY_COUNT = 4
+DEFAULT_RETRY_DELAY = 1
 
 parser = argparse.ArgumentParser()
 parser.add_argument(
@@ -53,6 +56,24 @@ parser.add_argument(
     default=DEFAULT_HOST
 )
 
+parser.add_argument(
+    '-r', '--retry',
+    action="store",
+    dest="retry_count",
+    help=f'Number of retry to retreive device status (default: {DEFAULT_RETRY_COUNT})',
+    type=int,
+    default=DEFAULT_RETRY_COUNT
+)
+
+parser.add_argument(
+    '-D', '--delay',
+    action="store",
+    dest="retry_delay",
+    help=f'Delay in second between two retry to retreive device status (default: {DEFAULT_RETRY_DELAY}s)',
+    type=int,
+    default=DEFAULT_RETRY_DELAY
+)
+
 options = parser.parse_args()
 
 logging.basicConfig(
@@ -70,9 +91,25 @@ if not devices_data:
     print('UNKNOWN - Fail to retreive devices using ESPHome Dashboard API')
     sys.exit(STATUS['UNKNOWN'])
 
-r = requests.get(f'{options.host}/ping')
-ping_data = r.json()
-logging.debug('Ping data: %s (%s)', ping_data, type(ping_data))
+count = 0
+while count < options.retry_count:
+    r = requests.get(f'{options.host}/ping')
+    count += 1
+    ping_data = r.json()
+    logging.debug('Ping data: %s (%s)', ping_data, type(ping_data))
+
+    if ping_data:
+        unreachable = False
+        for dev in ping_data:
+            if not ping_data[dev]:
+                unreachable = True
+                break
+
+        if not unreachable:
+            break
+
+    logging.debug('Wait %d seconds before retry...', options.retry_delay)
+    time.sleep(options.retry_delay)
 
 if not ping_data:
     print(