1+ from flask import Flask , jsonify
2+ import logging
3+ import os
4+ import subprocess
5+
6+ app = Flask (__name__ )
7+
8+ def str_to_bool (value ):
9+ return str (value ).strip ().lower () in {
10+ 'true' ,
11+ '1' ,
12+ 't' ,
13+ 'y' ,
14+ 'yes'
15+ }
16+
17+ class Config :
18+ SC4S_DEST_SPLUNK_HEC_DEFAULT_URL = os .getenv ('SC4S_DEST_SPLUNK_HEC_DEFAULT_URL' )
19+ HEALTHCHECK_PORT = int (os .getenv ('SC4S_LISTEN_STATUS_PORT' , '8080' ))
20+ CHECK_QUEUE_SIZE = str_to_bool (os .getenv ('HEALTHCHECK_CHECK_QUEUE_SIZE' , "false" ))
21+ MAX_QUEUE_SIZE = int (os .getenv ('HEALTHCHECK_MAX_QUEUE_SIZE' , '10000' ))
22+
23+ logging .basicConfig (
24+ format = f"%(asctime)s - healthcheck.py - %(levelname)s - %(message)s" ,
25+ datefmt = "%Y-%m-%d %H:%M:%S"
26+ )
27+ logger = logging .getLogger (__name__ )
28+
29+ def check_syslog_ng_health () -> bool :
30+ """Check the health of the syslog-ng process."""
31+ try :
32+ result = subprocess .run (
33+ ['syslog-ng-ctl' , 'healthcheck' , '-t' , '1' ],
34+ capture_output = True ,
35+ text = True ,
36+ timeout = 5
37+ )
38+ if result .returncode == 0 :
39+ return True
40+
41+ logger .error (f"syslog-ng healthcheck failed: { result .stderr .strip ()} )
42+ return False
43+ except subprocess .TimeoutExpired :
44+ logger .error ("syslog-ng healthcheck timed out." )
45+ return False
46+ except Exception as e :
47+ logger .exception (f"Unexpected error during syslog-ng healthcheck: { e } )
48+ return False
49+
50+ def check_queue_size (
51+ sc4s_dest_splunk_hec_default = Config .SC4S_DEST_SPLUNK_HEC_DEFAULT_URL ,
52+ max_queue_size = Config .MAX_QUEUE_SIZE
53+ ) -> bool :
54+ """Check syslog-ng queue size and compare it against the configured maximum limit."""
55+ if not sc4s_dest_splunk_hec_default :
56+ logger .error (
57+ "SC4S_DEST_SPLUNK_HEC_DEFAULT_URL not configured. "
58+ "Ensure the default HEC destination is set, or disable HEALTHCHECK_CHECK_QUEUE_SIZE."
59+ )
60+ return False
61+
62+ try :
63+ result = subprocess .run (
64+ ['syslog-ng-ctl' , 'stats' ],
65+ capture_output = True ,
66+ text = True ,
67+ timeout = 5
68+ )
69+ if result .returncode != 0 :
70+ logger .error (f"syslog-ng stats command failed: { result .stderr .strip ()} )
71+ return False
72+
73+ stats = result .stdout .splitlines ()
74+ destination_stat = next (
75+ (s for s in stats if ";queued;" in s and sc4s_dest_splunk_hec_default in s ),
76+ None
77+ )
78+ if not destination_stat :
79+ logger .error ("No matching queue stats found for the destination URL." )
80+ return False
81+
82+ queue_size = int (destination_stat .split (";" )[- 1 ])
83+ if queue_size > max_queue_size :
84+ logger .warning (
85+ f"Queue size { queue_size } { max_queue_size }
86+ )
87+ return False
88+
89+ return True
90+ except subprocess .TimeoutExpired :
91+ logger .error ("syslog-ng stats command timed out." )
92+ return False
93+ except Exception as e :
94+ logger .exception (f"Unexpected error checking queue size: { e } )
95+ return False
96+
97+ @app .route ('/health' , methods = ['GET' ])
98+ def healthcheck ():
99+ if Config .CHECK_QUEUE_SIZE :
100+ if not check_syslog_ng_health ():
101+ return jsonify ({'status' : 'unhealthy: syslog-ng healthcheck failed' }), 503
102+ if not check_queue_size ():
103+ return jsonify ({'status' : 'unhealthy: queue size exceeded limit' }), 503
104+ else :
105+ if not check_syslog_ng_health ():
106+ return jsonify ({'status' : 'unhealthy: syslog-ng healthcheck failed' }), 503
107+
108+ logger .info ("Service is healthy." )
109+ return jsonify ({'status' : 'healthy' }), 200
110+
111+
112+ if __name__ == '__main__' :
113+ app .run (host = '0.0.0.0' , port = Config .HEALTHCHECK_PORT )
0 commit comments