mirror of
https://github.com/Telecominfraproject/wlan-lanforge-scripts.git
synced 2025-10-30 18:27:53 +00:00
378 lines
14 KiB
Python
378 lines
14 KiB
Python
#!/usr/bin/env python3
|
|
|
|
# - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
|
|
# Class holds default settings for json requests to Ghost -
|
|
# - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
|
|
import ast
|
|
import os
|
|
import sys
|
|
|
|
if sys.version_info[0] != 3:
|
|
print("This script requires Python 3")
|
|
exit()
|
|
|
|
import requests
|
|
|
|
import jwt
|
|
from datetime import datetime as date
|
|
import json
|
|
import subprocess
|
|
from scp import SCPClient
|
|
import paramiko
|
|
from GrafanaRequest import GrafanaRequest
|
|
import time
|
|
from collections import Counter
|
|
import shutil
|
|
|
|
|
|
class CSVReader:
|
|
def read_csv(self,
|
|
file,
|
|
sep=','):
|
|
df = open(file).read().split('\n')
|
|
rows = list()
|
|
for x in df:
|
|
if len(x) > 0:
|
|
rows.append(x.split(sep))
|
|
length = list(range(0, len(df[0])))
|
|
columns = dict(zip(df[0], length))
|
|
return rows
|
|
|
|
def get_column(self,
|
|
df,
|
|
value):
|
|
index = df[0].index(value)
|
|
values = []
|
|
for row in df[1:]:
|
|
values.append(row[index])
|
|
return values
|
|
|
|
def get_columns(self, df, targets):
|
|
target_index = []
|
|
for item in targets:
|
|
target_index.append(df[0].index(item))
|
|
results = []
|
|
for row in df:
|
|
row_data = []
|
|
for x in target_index:
|
|
row_data.append(row[x])
|
|
results.append(row_data)
|
|
return results
|
|
|
|
def to_html(self, df):
|
|
html = ''
|
|
html = html + ('<table style="border:1px solid #ddd"><tr>')
|
|
for row in df[1:]:
|
|
for item in row:
|
|
html = html + ('<td style="border:1px solid #ddd">%s</td>' % item)
|
|
html = html + ('</tr>\n<tr>')
|
|
html = html + ('</table>')
|
|
return html
|
|
|
|
def filter_df(self, df, column, expression, target):
|
|
target_index = df[0].index(column)
|
|
counter = 0
|
|
targets = [0]
|
|
for row in df:
|
|
try:
|
|
if expression == 'less than':
|
|
if float(row[target_index]) < target:
|
|
targets.append(counter)
|
|
counter += 1
|
|
else:
|
|
counter += 1
|
|
if expression == 'greater than':
|
|
if float(row[target_index]) > target:
|
|
targets.append(counter)
|
|
counter += 1
|
|
else:
|
|
counter += 1
|
|
if expression == 'greater than or equal to':
|
|
if float(row[target_index]) >= target:
|
|
targets.append(counter)
|
|
counter += 1
|
|
else:
|
|
counter += 1
|
|
except:
|
|
counter += 1
|
|
return list(map(df.__getitem__, targets))
|
|
|
|
def concat(self, dfs):
|
|
final_df = dfs[0]
|
|
for df in dfs[1:]:
|
|
final_df = final_df + df[1:]
|
|
return final_df
|
|
|
|
|
|
class GhostRequest:
|
|
def __init__(self,
|
|
_ghost_json_host,
|
|
_ghost_json_port,
|
|
_api_token=None,
|
|
_overwrite='false',
|
|
debug_=False,
|
|
die_on_error_=False):
|
|
self.debug = debug_
|
|
self.die_on_error = die_on_error_
|
|
self.ghost_json_host = _ghost_json_host
|
|
self.ghost_json_port = _ghost_json_port
|
|
self.ghost_json_url = "http://%s:%s/ghost/api/v3" % (_ghost_json_host, _ghost_json_port)
|
|
self.data = dict()
|
|
self.data['overwrite'] = _overwrite
|
|
self.ghost_json_login = self.ghost_json_url + '/admin/session/'
|
|
self.api_token = _api_token
|
|
self.images = list()
|
|
self.pdfs = list()
|
|
|
|
def encode_token(self):
|
|
|
|
# Split the key into ID and SECRET
|
|
key_id, secret = self.api_token.split(':')
|
|
|
|
# Prepare header and payload
|
|
iat = int(date.now().timestamp())
|
|
|
|
header = {'alg': 'HS256', 'typ': 'JWT', 'kid': key_id}
|
|
payload = {
|
|
'iat': iat,
|
|
'exp': iat + 5 * 60,
|
|
'aud': '/v3/admin/'
|
|
}
|
|
token = jwt.encode(payload, bytes.fromhex(secret), algorithm='HS256', headers=header)
|
|
return token
|
|
|
|
def create_post(self,
|
|
title=None,
|
|
text=None,
|
|
tags=None,
|
|
authors=None,
|
|
status="published"):
|
|
ghost_json_url = self.ghost_json_url + '/admin/posts/?source=html'
|
|
post = dict()
|
|
posts = list()
|
|
datastore = dict()
|
|
datastore['html'] = text
|
|
datastore['title'] = title
|
|
datastore['status'] = status
|
|
posts.append(datastore)
|
|
post['posts'] = posts
|
|
|
|
headers = dict()
|
|
|
|
token = self.encode_token()
|
|
headers['Authorization'] = 'Ghost {}'.format(token)
|
|
response = requests.post(ghost_json_url, json=post, headers=headers)
|
|
if self.debug:
|
|
print(datastore)
|
|
print(ghost_json_url)
|
|
print('\n')
|
|
print(post)
|
|
print('\n')
|
|
print(headers)
|
|
print(response.headers)
|
|
|
|
def upload_image(self,
|
|
image):
|
|
print(image)
|
|
ghost_json_url = self.ghost_json_url + '/admin/images/upload/'
|
|
|
|
token = self.encode_token()
|
|
bashCommand = "curl -X POST -F 'file=@%s' -H \"Authorization: Ghost %s\" %s" % (image, token, ghost_json_url)
|
|
|
|
proc = subprocess.Popen(bashCommand, shell=True, stdout=subprocess.PIPE)
|
|
output = proc.stdout.read().decode('utf-8')
|
|
print(output)
|
|
self.images.append(json.loads(output)['images'][0]['url'])
|
|
|
|
def upload_images(self,
|
|
folder):
|
|
for image in os.listdir(folder):
|
|
if 'kpi' in image:
|
|
if 'png' in image:
|
|
self.upload_image(folder + '/' + image)
|
|
print('images %s' % self.images)
|
|
|
|
def custom_post(self,
|
|
folder,
|
|
authors,
|
|
title='custom'):
|
|
self.upload_images(folder)
|
|
head = '''<p>This is a custom post created via a script</p>'''
|
|
for picture in self.images:
|
|
head = head + '<img src="%s"></img>' % picture
|
|
head = head + '''<p>This is the end of the example</p>'''
|
|
self.create_post(title=title,
|
|
text=head,
|
|
tags='custom',
|
|
authors=authors)
|
|
|
|
def list_append(self, list_1, value):
|
|
list_1.append(value)
|
|
|
|
def kpi_to_ghost(self,
|
|
authors,
|
|
folders,
|
|
parent_folder=None,
|
|
title=None,
|
|
server_pull=None,
|
|
ghost_host=None,
|
|
port=22,
|
|
user_push=None,
|
|
password_push=None,
|
|
customer=None,
|
|
testbed='Unknown Testbed',
|
|
test_run=None,
|
|
target_folders=list(),
|
|
grafana_token=None,
|
|
grafana_host=None,
|
|
grafana_port=3000,
|
|
grafana_datasource='InfluxDB',
|
|
grafana_bucket=None):
|
|
text = ''
|
|
csvreader = CSVReader()
|
|
if grafana_token is not None:
|
|
grafana = GrafanaRequest(grafana_token,
|
|
grafana_host,
|
|
grafanajson_port=grafana_port
|
|
)
|
|
print('Folders: %s' % folders)
|
|
|
|
ssh_push = paramiko.SSHClient()
|
|
ssh_push.set_missing_host_key_policy(paramiko.client.AutoAddPolicy)
|
|
ssh_push.connect(ghost_host,
|
|
port,
|
|
username=user_push,
|
|
password=password_push,
|
|
allow_agent=False,
|
|
look_for_keys=False)
|
|
scp_push = SCPClient(ssh_push.get_transport())
|
|
|
|
if parent_folder is not None:
|
|
print("parent_folder %s" % parent_folder)
|
|
files = os.listdir(parent_folder)
|
|
print(files)
|
|
for file in files:
|
|
if os.path.isdir(parent_folder + '/' + file) is True:
|
|
if os.path.exists(file):
|
|
shutil.rmtree(file)
|
|
shutil.copytree(parent_folder + '/' + file, file)
|
|
target_folders.append(file)
|
|
print('Target folders: %s' % target_folders)
|
|
else:
|
|
for folder in folders:
|
|
print(folder)
|
|
target_folders.append(folder)
|
|
|
|
testbeds = list()
|
|
pdfs = list()
|
|
high_priority_list = list()
|
|
low_priority_list = list()
|
|
images = list()
|
|
|
|
for target_folder in target_folders:
|
|
try:
|
|
target_file = '%s/kpi.csv' % target_folder
|
|
df = csvreader.read_csv(file=target_file, sep='\t')
|
|
csv_testbed = csvreader.get_column(df, 'test-rig')[0]
|
|
pass_fail = Counter(csvreader.get_column(df, 'pass/fail'))
|
|
dut_hw = csvreader.get_column(df, 'dut-hw-version')[0]
|
|
dut_sw = csvreader.get_column(df, 'dut-sw-version')[0]
|
|
dut_model = csvreader.get_column(df, 'dut-model-num')[0]
|
|
dut_serial = csvreader.get_column(df, 'dut-serial-num')[0]
|
|
if pass_fail['PASS'] + pass_fail['FAIL'] > 0:
|
|
text = text + 'Tests passed: %s<br />' % pass_fail['PASS']
|
|
text = text + 'Tests failed: %s<br />' % pass_fail['FAIL']
|
|
text = text + 'Percentage of tests passed: %s<br />' % (
|
|
pass_fail['PASS'] / (pass_fail['PASS'] + pass_fail['FAIL']))
|
|
|
|
except:
|
|
print("Failure")
|
|
target_folders.remove(target_folder)
|
|
break
|
|
testbeds.append(csv_testbed)
|
|
if testbed == 'Unknown Testbed':
|
|
raise UserWarning('Please define your testbed')
|
|
|
|
local_path = '/home/%s/%s/%s' % (user_push, customer, testbed)
|
|
|
|
transport = paramiko.Transport(ghost_host, port)
|
|
transport.connect(None, user_push, password_push)
|
|
sftp = paramiko.sftp_client.SFTPClient.from_transport(transport)
|
|
|
|
print(local_path)
|
|
print(target_folder)
|
|
scp_push.put(target_folder, local_path, recursive=True)
|
|
files = sftp.listdir(local_path + '/' + target_folder)
|
|
for file in files:
|
|
if 'pdf' in file:
|
|
url = 'http://%s/%s/%s/%s/%s' % (
|
|
ghost_host, customer.strip('/'), testbed, target_folder, file)
|
|
pdfs.append('PDF of results: <a href="%s">%s</a><br />' % (url, file))
|
|
scp_push.close()
|
|
self.upload_images(target_folder)
|
|
for image in self.images:
|
|
if 'kpi-' in image:
|
|
if '-print' not in image:
|
|
images.append('<img src="%s"></img>' % image)
|
|
self.images = []
|
|
|
|
results = csvreader.get_columns(df, ['short-description', 'numeric-score', 'test details', 'test-priority'])
|
|
|
|
low_priority = csvreader.filter_df(results, 'test-priority', 'less than', 94)
|
|
high_priority = csvreader.filter_df(results, 'test-priority', 'greater than or equal to', 95)
|
|
high_priority_list.append(high_priority)
|
|
|
|
low_priority_list.append(low_priority)
|
|
|
|
now = date.now()
|
|
|
|
high_priority = csvreader.concat(high_priority_list)
|
|
low_priority = csvreader.concat(low_priority_list)
|
|
|
|
high_priority = csvreader.get_columns(high_priority, ['short-description', 'numeric-score', 'test details'])
|
|
low_priority = csvreader.get_columns(low_priority, ['short-description', 'numeric-score', 'test details'])
|
|
|
|
if title is None:
|
|
title = "%s %s %s %s:%s report" % (now.day, now.month, now.year, now.hour, now.minute)
|
|
|
|
# create Grafana Dashboard
|
|
target_files = []
|
|
for folder in target_folders:
|
|
target_files.append(folder.split('/')[-1] + '/kpi.csv')
|
|
print('Target files: %s' % target_files)
|
|
grafana.create_custom_dashboard(target_csvs=target_files,
|
|
title=title,
|
|
datasource=grafana_datasource,
|
|
bucket=grafana_bucket)
|
|
|
|
text = 'Testbed: %s<br />' % testbeds[0]
|
|
dut_table = '<table><tr><td>DUT_HW</td><td>DUT_SW</td><td>DUT model</td><td>DUT Serial</td></tr>' \
|
|
'<tr><td>%s</td><td>%s</td><td>%s</td><td>%s</td></tr></table>' % (dut_hw, dut_sw, dut_model, dut_serial)
|
|
text = text + dut_table
|
|
|
|
for pdf in pdfs:
|
|
text = text + pdf
|
|
|
|
for image in images:
|
|
text = text + image
|
|
|
|
text = text + 'High priority results: %s' % csvreader.to_html(high_priority)
|
|
|
|
if grafana_token is not None:
|
|
# get the details of the dashboard through the API, and set the end date to the youngest KPI
|
|
grafana.list_dashboards()
|
|
|
|
grafana.create_snapshot(title='Testbed: ' + title)
|
|
time.sleep(3)
|
|
snapshot = grafana.list_snapshots()[-1]
|
|
text = text + '<iframe src="http://%s:3000/dashboard/snapshot/%s" width="100%s" height=1500></iframe><br />' % (
|
|
grafana_host, snapshot['key'], '%')
|
|
|
|
text = text + 'Low priority results: %s' % csvreader.to_html(low_priority)
|
|
|
|
self.create_post(title=title,
|
|
text=text,
|
|
tags='custom',
|
|
authors=authors)
|
|
|