def __init__(self, name="unnamed", config_options=None):
"""What's that coming over the hill? Is a monitor?"""
if config_options is None:
config_options = {}
self.name = name
self.monitor_logger = logging.getLogger('simplemonitor.monitor-' + self.name)
self.set_dependencies(Monitor.get_config_option(
config_options,
'depend',
required_type='[str]',
default=list()
))
self.set_urgency(Monitor.get_config_option(
config_options,
'urgent',
required_type='bool',
default=True
))
self.set_notify(Monitor.get_config_option(
config_options,
'notify',
required_type='bool',
default=True
))
self.set_group(Monitor.get_config_option(
config_options,
'group',
default='default'
))
self.set_tolerance(Monitor.get_config_option(
config_options,
'tolerance',
required_type='int',
default=0,
minimum=0
))
self.set_remote_alerting(Monitor.get_config_option(
config_options,
'remote_alert',
required_type='bool',
default=False
))
self.set_recover_command(Monitor.get_config_option(
config_options,
'recover_command'
))
self.set_gap(Monitor.get_config_option(
config_options,
'gap',
required_type='int',
minimum=0,
default=0
))
self.running_on = short_hostname()
self.was_skipped = False
def __init__(self, name="unnamed", config_options=None):
"""What's that coming over the hill? Is a monitor?"""
if config_options is None:
config_options = {}
self.name = name
self.monitor_logger = logging.getLogger('simplemonitor.monitor-' +
self.name)
self.set_dependencies(
Monitor.get_config_option(config_options,
'depend',
required_type='[str]',
default=list()))
self.set_urgency(
Monitor.get_config_option(config_options,
'urgent',
required_type='bool',
default=True))
self.set_notify(
Monitor.get_config_option(config_options,
'notify',
required_type='bool',
default=True))
self.set_group(
Monitor.get_config_option(config_options,
'group',
default='default'))
self.set_tolerance(
Monitor.get_config_option(config_options,
'tolerance',
required_type='int',
default=0,
minimum=0))
self.set_remote_alerting(
Monitor.get_config_option(config_options,
'remote_alert',
required_type='bool',
default=False))
self.set_recover_command(
Monitor.get_config_option(config_options, 'recover_command'))
self.set_gap(
Monitor.get_config_option(config_options,
'gap',
required_type='int',
minimum=0,
default=0))
self.running_on = short_hostname()
self.was_skipped = False
def is_remote(self):
"""Check if we're running on this machine, or if we're a remote instance."""
if self.running_on == short_hostname():
return False
return True
def process_batch(self):
"""Save the HTML file."""
ok_count = 0
fail_count = 0
old_count = 0
remote_count = 0
my_host = short_hostname()
try:
temp_file = tempfile.mkstemp()
file_handle = os.fdopen(temp_file[0], "w")
file_name = temp_file[1]
except Exception:
sys.stderr.write("Couldn't create temporary file for HTML output\n")
return
output_ok = StringIO()
output_fail = StringIO()
keys = list(self.batch_data.keys())
keys.sort()
for entry in keys:
if self.batch_data[entry]["age"] > 120:
status = "OLD"
old_count += 1
elif self.batch_data[entry]["status"]:
status = "OK"
ok_count += 1
else:
status = "FAIL"
fail_count += 1
if self.batch_data[entry]["host"] != my_host:
remote_count += 1
try:
monitor_name = entry.split("/")[1]
except Exception:
monitor_name = entry
if status == "FAIL":
output = output_fail
else:
output = output_ok
output.write("<tr class=\"%srow\">" % status.lower())
output.write("""
<td class="monitor_name">%s</td>
<td class="status %s">%s</td>
<td>%s</td>
<td>%s</td>
""" % (
monitor_name,
status.lower(), status,
self.batch_data[entry]["host"],
self.batch_data[entry]["fail_time"],
)
)
if self.batch_data[entry]["fail_count"] == 0:
output.write("<td class=\"vfc\"> </td>")
else:
output.write("<td class=\"vfc\">%s</td>" % self.batch_data[entry]["fail_count"])
try:
output.write("<td>%d+%02d:%02d:%02d</td>" % (self.batch_data[entry]["downtime"][0], self.batch_data[entry]["downtime"][1], self.batch_data[entry]["downtime"][2], self.batch_data[entry]["downtime"][3]))
except Exception:
output.write("<td> </td>")
output.write("<td>%s </td>" % (self.batch_data[entry]["fail_data"]))
if self.batch_data[entry]["failures"] == 0:
output.write("<td></td><td></td>")
else:
output.write("""<td>%s</td>
<td>%s</td>""" % (
self.batch_data[entry]["failures"],
format_datetime(self.batch_data[entry]["last_failure"])
)
)
if self.batch_data[entry]["host"] == my_host:
output.write("<td></td>")
else:
output.write("<td>%d</td>" % self.batch_data[entry]["age"])
output.write("</tr>\n")
count_data = "<div id=\"summary\""
if old_count > 0:
cls = "old"
elif fail_count > 0:
cls = "fail"
else:
cls = "ok"
count_data = count_data + " class=\"%s\">%s" % (cls, cls.upper())
self.count_data = count_data + "<div id=\"details\"><span class=\"ok\">%d OK</span> <span class=\"fail\">%d FAIL</span> <span class=\"old\">%d OLD</span> <span class=\"remote\">%d remote</span></div></div>" % (ok_count, fail_count, old_count, remote_count)
self.status = cls.upper()
with open(os.path.join(self.folder, self.header), "r") as file_input:
file_handle.writelines(self.parse_file(file_input))
file_handle.write(output_fail.getvalue())
file_handle.write(output_ok.getvalue())
with open(os.path.join(self.folder, self.footer), "r") as file_input:
file_handle.writelines(self.parse_file(file_input))
try:
file_handle.flush()
file_handle.close()
os.chmod(file_name, stat.S_IREAD | stat.S_IWRITE | stat.S_IRGRP | stat.S_IROTH)
shutil.move(file_name, os.path.join(self.folder, self.filename))
except Exception:
self.logger_logger.exception("problem closing temporary file for HTML output")
def is_remote(self):
"""Check if we're running on this machine, or if we're a remote instance."""
if self.running_on == short_hostname():
return False
return True