1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
|
import math
import numpy as np
import pandas as pd
import plotly.graph_objs as go
from plotly.offline import plot
def resource_distribution_plot(df_resources, df_task, type, label, option, columns=20,):
assert type == "psutil_process_time_user" or type == "psutil_process_memory_resident"
assert option == "avg" or option == "max"
min_range = min(df_resources[type].astype('float'))
max_range = max(df_resources[type].astype('float'))
time_step = (max_range - min_range) / columns
if min_range == max_range:
x_axis = [min_range]
else:
x_axis = []
for i in np.arange(min_range, max_range + time_step, time_step):
x_axis.append(i)
tasks_dict = dict()
for i in range(len(df_task)):
row = df_task.iloc[i]
tasks_dict[row['task_id']] = []
def y_axis_setup():
items = [0] * len(x_axis)
for task_id, tasks in tasks_dict.items():
if option == 'avg':
task = df_resources[df_resources['task_id'] ==
task_id][type].astype('float').mean()
elif option == 'max':
task = df_resources[df_resources['task_id'] == task_id][type].astype('float').max()
else:
raise ValueError(f"Cannot plot unknown aggregation option {option}")
for i in range(len(x_axis) - 1):
a = task >= x_axis[i]
b = task < x_axis[i + 1]
if a and b:
items[i] += 1
break
if task >= x_axis[-1]:
items[-1] += 1
return items
if type == "psutil_process_time_user":
xaxis = dict(autorange=True,
title='CPU user time (seconds)')
elif type == "psutil_process_memory_resident":
xaxis = dict(autorange=True,
title='Memory usage (bytes)')
else:
raise ValueError(f"Cannot plot unknown type {type}")
fig = go.Figure(
data=[go.Bar(x=x_axis,
y=y_axis_setup(),
name='tasks')],
layout=go.Layout(xaxis=xaxis,
yaxis=dict(title='Tasks'),
title=label + '(' + option + ')'))
return plot(fig, show_link=False, output_type="div", include_plotlyjs=False)
def resource_time_series(tasks, type='psutil_process_time_user', label='CPU user time'):
tasks['epoch_time'] = (pd.to_datetime(
tasks['timestamp']) - pd.Timestamp("1970-01-01")) // pd.Timedelta('1s')
step = int(tasks['resource_monitoring_interval'][0])
start = tasks['epoch_time'].min()
end = tasks['epoch_time'].max()
tasks['relative_time'] = tasks['epoch_time'] - start
if end != start:
bins = pd.cut(tasks['relative_time'],
range(0, end - start + 1, step),
include_lowest=True)
df = tasks.groupby(bins, as_index=False)[type].mean()
df['time'] = step * df.index
fig = go.Figure(
data=[go.Scatter(x=df['time'],
y=df[type],
)],
layout=go.Layout(xaxis=dict(autorange=True,
title='Time (seconds)'),
yaxis=dict(title=label),
title=label))
else:
fig = go.Figure(
data=[go.Scatter(x=[0],
y=[tasks[type].mean()],
)],
layout=go.Layout(xaxis=dict(autorange=True,
title='Time (seconds)'),
yaxis=dict(title=label),
title=label))
return plot(fig, show_link=False, output_type="div", include_plotlyjs=False)
def worker_efficiency(task, node):
try:
node['epoch_time'] = (pd.to_datetime(
node['timestamp']) - pd.Timestamp("1970-01-01")) / pd.Timedelta('1s')
task['epoch_time_start'] = (pd.to_datetime(
task['task_try_time_launched']) - pd.Timestamp("1970-01-01")) / pd.Timedelta('1s')
task['epoch_time_running'] = (pd.to_datetime(
task['task_try_time_running']) - pd.Timestamp("1970-01-01")) / pd.Timedelta('1s')
task['epoch_time_returned'] = (pd.to_datetime(
task['task_try_time_returned']) - pd.Timestamp("1970-01-01")) / pd.Timedelta('1s')
start = int(min(task['epoch_time_start'].min(), node['epoch_time'].min()))
end = int(task['epoch_time_returned'].max())
# worker_plot will be used to make a histogram of worker usage, one bucket per
# second
worker_plot = [0] * (end - start + 1)
total_workers = node['worker_count'].sum()
for i, row in task.iterrows():
if math.isnan(row['epoch_time_running']):
# skip tasks with no running start time
continue
if math.isnan(row['epoch_time_returned']):
# there is no end time for this, so we should assume the "end" time
etr = end
else:
etr = row['epoch_time_returned']
# there's a two cases here:
# The task starts and ends in same one-second bucket:
# populate one bucket with the fraction of second between
# the start and end
#
# or
#
# The task starts in one bucket and ends in another:
# populate middle buckets with 1, and start/end buckets with fraction
start_bucket = int(row['epoch_time_running']) - start
end_bucket = int(etr) - start
if start_bucket == end_bucket:
fraction = etr - row['epoch_time_running']
worker_plot[start_bucket] += fraction
else:
fraction_before = row['epoch_time_running'] - int(row['epoch_time_running'])
worker_plot[start_bucket] += (1 - fraction_before)
for j in range(start_bucket + 1, end_bucket):
worker_plot[j] += 1
fraction_after = etr - int(etr)
worker_plot[end_bucket] += fraction_after
fig = go.Figure(
data=[go.Scatter(x=list(range(0, end - start + 1)),
y=worker_plot,
name='Total busy workers',
),
go.Scatter(x=list(range(0, end - start + 1)),
y=[total_workers] * (end - start + 1),
name='Total of workers in whole run',
)
],
layout=go.Layout(xaxis=dict(autorange=True,
title='Time (seconds)'),
yaxis=dict(title='Number of workers'),
title="Worker efficiency"))
return plot(fig, show_link=False, output_type="div", include_plotlyjs=False)
except Exception as e:
print(e)
return "The worker efficiency plot cannot be generated due to missing data."
def resource_efficiency(resource, node, label):
try:
resource['epoch_time'] = (pd.to_datetime(
resource['timestamp']) - pd.Timestamp("1970-01-01")) // pd.Timedelta('1s')
node['epoch_time'] = (pd.to_datetime(
node['timestamp']) - pd.Timestamp("1970-01-01")) // pd.Timedelta('1s')
resource = resource.sort_values(by='epoch_time')
start = min(resource['epoch_time'].min(), node['epoch_time'].min())
end = resource['epoch_time'].max()
resource['relative_time'] = resource['epoch_time'] - start
node['relative_time'] = node['epoch_time'] - start
resource['key'] = resource['task_id'].astype(str) + "-" + resource['try_id'].astype(str)
task_plot = [0] * (end - start + 1)
if label == 'CPU':
total = node['cpu_count'].sum()
elif label == 'mem':
total = node['total_memory'].sum() / 1024 / 1024 / 1024
resource['total_cpu_time'] = resource['psutil_process_time_user'] + resource['psutil_process_time_system']
for key in resource['key'].unique():
tmp = resource[resource['key'] == key]
tmp['last_timestamp'] = tmp['relative_time'].shift(1)
if label == 'CPU':
tmp['last_cputime'] = tmp['total_cpu_time'].shift(1)
for index, row in tmp.iterrows():
if np.isnan(row['last_timestamp']):
continue
for i in range(int(row['last_timestamp']), int(row['relative_time'])):
if label == 'CPU':
diff = (row['total_cpu_time'] - row['last_cputime']) / (row['relative_time'] - row['last_timestamp'])
elif label == 'mem':
diff = row['psutil_process_memory_resident'] / 1024 / 1024 / 1024
task_plot[i] += diff
if label == 'CPU':
name1 = 'Used CPU cores'
name2 = 'Total CPU cores'
yaxis = 'Number of CPU cores'
title = 'CPU usage'
elif label == 'mem':
name1 = 'Used memory'
name2 = 'Total memory'
yaxis = 'Memory (GB)'
title = 'Memory usage'
fig = go.Figure(
data=[go.Scatter(x=list(range(0, end - start + 1)),
y=task_plot,
name=name1,
),
go.Scatter(x=list(range(0, end - start + 1)),
y=[total] * (end - start + 1),
name=name2,
)
],
layout=go.Layout(xaxis=dict(autorange=True,
title='Time (seconds)'),
yaxis=dict(title=yaxis),
title=title))
return plot(fig, show_link=False, output_type="div", include_plotlyjs=False)
except Exception as e:
print(e)
return "The resource efficiency plot cannot be generated because of exception {}.".format(e)
|
