def should_display(self):
data_dog = DataDog()
try:
data_dog.get_api_key()
data_dog.get_app_key()
except Exception:
return False
return True
def admin_page(request):
data_dog = None
hosts = []
try:
data_dog = DataDog()
api_key = data_dog.get_api_key()
app_key = data_dog.get_app_key()
url = data_dog.generate_url("hosts")
response = requests.get(url,
params={
'api_key': api_key,
'application_key': app_key
})
res = response.json()
for host in res['host_list']:
res_dict = {}
res_dict['name'] = host['name']
res_dict['os'] = host['meta']['platform']
res_dict['sources'] = host['sources']
res_dict['up'] = host['up']
res_dict['ipaddress'] = json.loads(
host['meta']['gohai'])['network']['ipaddress']
res_dict['agent_version'] = host['meta']['agent_version']
server = Server.objects.filter(hostname__icontains=host['name'],
ip=res_dict['ipaddress']).first()
if server is not None:
res_dict['server_url'] = server.get_absolute_url()
hosts.append(res_dict)
except Exception:
pass
data_dog_context = {
'tabs':
TabGroup(
template_dir='data_dog/templates',
request=request,
tabs=[
(_("Dashboard"), 'dashboard',
dict(
context={"data_dog_ci": data_dog.get_connection_info()})),
(_("Servers"), 'servers', dict(context={"res": hosts})),
],
)
}
return render(request, 'data_dog/templates/admin_page.html',
data_dog_context)
def server_tab_data_dog(request, obj_id=None):
data_dog = DataDog()
context = {}
CustomField.objects.get_or_create(name='datadog_installed',
type='BOOL',
defaults={
'label':
'Is Datadog Agent Installed',
'show_as_attribute': False
})
api_key = data_dog.get_api_key()
app_key = data_dog.get_app_key()
server = get_object_or_404(Server, pk=obj_id)
now = int(time.time())
time_from = now - 3600
# Queries
# Datadog automatically truncates a server name to 15 characters.
# For the truncation, it only happens for windows machines
if server.is_windows():
hostname = server.hostname[:15]
else:
hostname = server.hostname
cpu_idle = f"system.cpu.idle{'{host:'}{hostname}{'}'}"
cpu_io_wait = f"system.cpu.iowait{'{host:'}{hostname}{'}'}"
free_ram_query = f"system.mem.free{'{host:'}{hostname}{'}'}"
total_ram_query = f"system.mem.total{'{host:'}{hostname}{'}'}"
paged_memory = f"system.mem.paged{'{host:'}{hostname}{'}'}"
total_disk_space_query = f"system.disk.total{'{host:'}{hostname}{'}'}"
used_disk_space_query = f"system.disk.used{'{host:'}{hostname}{'}'}"
free_disk_query = f"system.disk.free{'{host:'}{hostname}{'}'}"
disk_in_use_query = f"system.disk.in_use{'{host:'}{hostname}{'}'}"
disk_write_time = f"system.disk.write_time_pct{'{host:'}{hostname}{'}'}"
url = data_dog.generate_url("query")
queries = [
cpu_idle, cpu_io_wait, free_ram_query, total_ram_query,
total_disk_space_query, free_disk_query, disk_in_use_query,
used_disk_space_query, disk_write_time, paged_memory
]
for query in queries:
response = requests.get(url,
params={
'api_key': api_key,
'application_key': app_key,
'from': time_from,
'to': now,
'query': query
})
if response.ok:
info = response.json()
if len(info.get('series')) > 0:
series = info.get('series')[0]
pointlist = series.get('pointlist')[-1]
context.update({series.get('metric'): info})
# cpu operations
if series.get('metric') == 'system.cpu.idle':
points = series.get('pointlist')
_pointlist = [[i[0], round(i[1], 2)] for i in points]
context.update({
'cpu_idle': [{
'data': _pointlist,
'name': 'CPU Idle ',
'color': '#153c55',
'type': 'area'
}]
})
elif series.get('metric') == 'system.cpu.iowait':
points = series.get('pointlist')
_pointlist = [[i[0], round(i[1], 2)] for i in points]
context.update({
'cpu_io_wait': [{
'data': _pointlist,
'name': 'CPU IO Wait ',
'color': '#0086AD',
'type': 'area'
}]
})
# memory operations
elif series.get('metric') == 'system.mem.free':
free_ram = pointlist[-1] / 1e9
context.update({'free_ram': round(free_ram, 1)})
points = series.get('pointlist')
_pointlist = [[i[0], round(i[1] / 1e9, 2)] for i in points]
context.update({
'free_memory_series': [{
'data': _pointlist,
'name': 'Free Memory ',
'color': '#666666',
'type': 'area'
}]
})
elif series.get('metric') == 'system.mem.paged':
points = series.get('pointlist')
_pointlist = [[i[0], round(i[1] / 1e9, 2)] for i in points]
context.update({
'paged_memory': [{
'data': _pointlist,
'name': 'Paged Memory',
'color': '#FF901F',
'type': 'area'
}]
})
# disk operations
if series.get('metric') == 'system.disk.used':
disk_used = pointlist[-1] / 1e9
context.update({'disk_used': round(disk_used)})
points = series.get('pointlist')
_pointlist = [[i[0], round(i[1] / 1e9, 2)] for i in points]
context.update({
'disk_used_series': [{
'data': _pointlist,
'name': 'Disk Used',
'color': '#0069cf',
'type': 'area'
}]
})
elif series.get('metric') == 'system.disk.free':
free_disk = pointlist[-1] / 1e9
context.update({'free_disk': round(free_disk)})
elif series.get('metric') == 'system.disk.write_time_pct':
points = series.get('pointlist')
context.update({
'disk_write_time': [{
'data': points,
'name': 'Disk Write Time',
'color': '#B6A959',
'type': 'area'
}]
})
if not context:
# Check if server is powered on.
if server.power_status == 'POWERON':
# This server has no agent installed
context.update({'agent_installed': False})
context.update({'power_status': True})
else:
context.update({'power_status': False})
status = 'warning'
msg = "Datadog might be installed in this server but the server is powered off"
server_not_powered_on = helper_tags.alert(status, msg)
context.update({'server_not_powered_on': server_not_powered_on})
else:
# Since some servers might have agents not installed from coudbolt,
# their datadog_installed value will be none. We can update this to True.
# If the agent has been uninstalled, the datadog_installed value will be False.
if server.datadog_installed == None:
server.datadog_installed = True
server.save()
context.update({'agent_installed': True})
context.update({'server_id': obj_id})
server_settings_ok = check_server_settings_status(server)
if not server_settings_ok:
status = 'warning'
msg = "Datadog agent is not installed and the server username and password are not correctlty setup. This might make it imposible to install the agent on this server from cloudbolt. You can configure them on the Configuration page on the server details tab. "
server_credentials_not_set = helper_tags.alert(status, msg)
context.update(
{'server_credentials_not_set': server_credentials_not_set})
install_job = server.jobs.filter(
job_parameters__hookparameters__hook__name='Install DataDog Agent'
).last()
uninstall_job = server.jobs.filter(
job_parameters__hookparameters__hook__name='UnInstall Datadog Agent'
).last()
job_running = False
uninstall_job_running = False
if install_job:
if install_job.is_active():
job_running = True
context.update({'job_url': install_job.get_absolute_url()})
if uninstall_job and uninstall_job.is_active():
uninstall_job_running = True
context.update({'uninstall_job_url': uninstall_job.get_absolute_url()})
if not context.get('agent_installed') and server.datadog_installed:
context.update({'waiting_for_metrics': True})
context.update({
'job_running': job_running,
'uninstall_job_running': uninstall_job_running,
'datadog_actions': get_datadog_actions(),
'datadog_installed': server.datadog_installed
})
return render(request, 'data_dog/templates/data_dog_server_tab.html',
context)
def run(server, *args, **kwargs):
set_progress("Starting Install DataDog agent process...")
data_dog = DataDog()
api_key = data_dog.get_api_key()
if server.is_windows():
download_agent_result = data_dog.download_agent(
DOWNLOAD_DATA_DOG_AGENT_WINDOWS, {}, server=server)
if download_agent_result.strip(
'\r\n'
) == "Error occurred while trying to download the agent.":
set_progress("Error occurred while trying to download the agent.")
else:
try:
data_dog.install_agent(INSTALL_DATADOG_AGENT_WINDOWS,
{'api_key': api_key},
server=server,
run_with_sudo=False)
except Exception as error:
set_progress(
f"The following Error occurred while trying to install the agent {error}"
)
return "FAILURE", "", f"{error}"
try:
agent_status = data_dog.check_agent_status(
CHECK_AGENT_STATUS_WINDOWS, {}, server)
if int([
pid.strip() for pid in agent_status.split('\r\n')
if pid.strip().split(' ')[0] == 'Pid:'
][0].split(' ')[-1]):
set_progress("Agent is running")
server.datadog_installed = True
server.save()
return "SUCCESS", "", "Agent installed"
except Exception as error:
set_progress(
f"The following Error occurred while trying to get the status of the agent. {error}"
)
return "FAILURE", "", f"{error}"
elif server.os_family.get_base_name() == 'Linux':
try:
data_dog.install_agent(INSTALL_DATADOG_AGENT_LINUX,
{'api_key': api_key},
server=server,
run_with_sudo=True)
except Exception as error:
set_progress(
f"The following Error occurred while trying to install the agent. {error}"
)
return "FAILURE", "", f"{error}"
try:
agent_status = data_dog.check_agent_status(
CHECK_AGENT_STATUS_LINUX, {'api_key': api_key}, server=server)
status = [
status.strip() for status in agent_status.split('\n')
if status.strip().split(':')[0] == 'Active'
]
if status:
set_progress(
f"{status[0][:25]}. Initial metrics might take few minutes before showing up."
)
server.datadog_installed = True
server.save()
return "SUCCESS", "Agent Installed", ""
except Exception as error:
set_progress(
f"The following Error occurred while trying to get the status of the agent. {error}"
)
return "FAILURE", "", f"{error}"
else:
set_progress("Data Dog Agent can't be installed on this server.")
return "FAILURE", "Agent Not Installed", "The OS type is not currently supported."