This commit is contained in:
2024-01-03 13:40:42 +01:00
parent 8d4882d7c4
commit d5d964ea8e
3 changed files with 87 additions and 4 deletions

View File

@@ -24,6 +24,7 @@ from flask_sqlalchemy import SQLAlchemy
from sqlalchemy.sql.expression import func
import icingatools
import smarttools
app = flask.Flask("Icinga Report In Gateway")
@@ -72,6 +73,7 @@ class SMARTStatus(db.Model):
model_number = Column(String)
power_cycles = Column(Integer)
power_on_hours = Column(Integer)
wearleveling_count = Column(Integer)
def buildReponseDict(status, service=None):
@@ -311,6 +313,13 @@ def default():
smart = flask.request.json.get("smart")
# check smart json quoting problems #
if smart and type(smart) == str:
try:
smart = json.loads(smart)
except json.decoder.JSONDecodeError as e:
return ("Error in SMART-json {}".format(e), 415)
if not service:
return ("'service' ist empty field in json", 400)
elif not token:
@@ -343,7 +352,10 @@ def default():
def record_and_check_smart(service, timestamp, smart):
health_info = smart["nvme_smart_health_information_log"]
if "nvme_smart_health_information_log" in smart:
health_info = smart["nvme_smart_health_information_log"]
else:
health_info = smarttools.normalize(smart)
if not service.special_type == "SMART":
raise AssertionError("Trying to record SMART-record for non-SMART service")
@@ -356,7 +368,8 @@ def record_and_check_smart(service, timestamp, smart):
power_cycles=health_info["power_cycles"],
power_on_hours=health_info["power_on_hours"],
available_spare=health_info.get("available_spare"),
model_number=smart.get("model_name"))
model_number=smart.get("model_name"),
wearleveling_count=health_info.get("wearleveling_count"))
db.session.add(smart_status)
db.session.commit()
@@ -378,6 +391,10 @@ def record_and_check_smart(service, timestamp, smart):
if smart_last.critical_warning != 0:
return ("SMART reports disk critical => oO better do something about this", "CRITICAL")
# wearleveling < 20% (SAMSUNG only) #
if smart_last.wearleveling_count and smart_last.wearleveling_count <= 20:
return ("SMART report prefail disk (wear_level < 20%)", "CRITICAL")
# temp max > X #
if smart_last.temperature > 50:
return ("Disk Temperatur {}".format(smart_last.temperature), "CRITICAL")
@@ -386,7 +403,8 @@ def record_and_check_smart(service, timestamp, smart):
spare_change = smart_old.available_spare - smart_last.available_spare
if smart_last.available_spare <= 25:
return ("SSD spare <25 ({}) YOUR DISK WILL DIE SOON".format(spare_change), "CRITICAL")
return ("SSD spare <25 ({}) YOUR DISK WILL DIE SOON".format(spare_change),
"CRITICAL")
elif smart_last.available_spare <= 50:
return ("SSD spare <50 ({})".format(spare_change), "WARNING")
elif spare_change >= 10:
@@ -395,7 +413,8 @@ def record_and_check_smart(service, timestamp, smart):
# unsafe_shutdowns +1 #
if smart_second_last.unsafe_shutdowns - smart_last.unsafe_shutdowns >= 1:
return ("Disk had {} unsafe shutdowns".format(smart_last.unsafe_shutdowns), "WARNING")
return ("Disk had {} unsafe shutdowns".format(smart_last.unsafe_shutdowns),
"WARNING")
return ("", "OK")

49
smarttools.py Normal file
View File

@@ -0,0 +1,49 @@
def normalize(smart):
'''Load different types of SMART outputs'''
ret = dict()
ret.update({ "temperature" : 0 })
ret.update({ "critical_warning" : 0 })
ret.update({ "unsafe_shutdowns" : 0 })
ret.update({ "power_cycles" : 0 })
ret.update({ "power_on_hours" : 0 })
ret.update({ "available_spare" : 100 })
ret.update({ "wearleveling_count" : 100 })
if "ata_smart_attributes" in smart:
# get main table #
table = smart["ata_smart_attributes"]["table"]
# temperatur #
ret["temperature"] = smart["temperature"]["current"]
for el in table:
# look for relevant metrics #
name = el["name"].lower()
target_name = el["name"].lower() # name in return map
# handle value mapping #
use_raw = False
if name == "used_rsvd_blk_cnt_tot":
target_name = "available_spare"
elif name == "power_cylce_count":
target_name = "power_cycles"
use_raw = True
elif name == "power_on_hours":
target_name = "power_on_hours"
use_raw = True
# check if metric should be recorded #
if target_name in ret:
# set return dict #
if use_raw:
value = el["raw"]["value"]
else:
value = el["value"]
ret[target_name] = value
return ret

View File

@@ -115,6 +115,20 @@
{% endif %}
</div>
{% if smart %}
<h5 class="my-4">Windows</h5>
<div class="ml-3 example">
$SMART = @{ <br>
<div class="example-indent">
service = "{{ service.service }}"<br>
token = "{{ service.token }}"<br>
status = "N/A"<br>
smart = "$(smartctl -a C: --json | Out-String)"<br>
</div>
} | ConvertTo-Json<br><br>
Invoke-RestMethod -TimeoutSec 2 -Uri "{{ flask.request.url_root.replace("http://", "https://" )}}report" -Method Post -Headers @{"Content-Type"="application/json"} -Body $SMART
</div>
{% else %}
<h5 class="my-4">Python</h5>
<div class="ml-3 example">
import requests<br>
@@ -129,6 +143,7 @@
</div>
</div>
</div>
{% endif %}
<table class="mb-4 mt-5 status-table">
<thead>