monasca-agent/tests_to_fix/test_monstatsd.py
Chuck Short b8bb2ff4c3 Add python36 support
Use the six library to get monasca-agent to work with
python2.7 and python3.

Story: 2004148
Task: 27621

Change-Id: I0de315967dd5a745741fda0c53ce8cc85cda8cc5
Signed-off-by: Chuck Short <chucks@redhat.com>
2018-10-25 09:09:53 -04:00

465 lines
16 KiB
Python
Raw Blame History

# Licensed under the Apache License, Version 2.0 (the "License"); you may
# not use this file except in compliance with the License. You may obtain
# a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
# WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
# License for the specific language governing permissions and limitations
# under the License.
import random
import time
import unittest
import nose.tools as nt
from six.moves import range
from monasca_agent.common.aggregator import MetricsAggregator
class TestUnitMonascaStatsd(unittest.TestCase):
@staticmethod
def sort_metrics(metrics):
def sort_by(m):
return (m['metric'], ','.join(m['dimensions'] or []))
return sorted(metrics, key=sort_by)
@staticmethod
def sort_events(metrics):
def sort_by(m):
return (m['title'], m['text'], ','.join(m.get('tags', None) or []))
return sorted(metrics, key=sort_by)
def test_counter_normalization(self):
stats = MetricsAggregator('myhost', interval=10)
# Assert counters are normalized.
stats.submit_packets('int:1|c')
stats.submit_packets('int:4|c')
stats.submit_packets('int:15|c')
stats.submit_packets('float:5|c')
metrics = self.sort_metrics(stats.flush())
assert len(metrics) == 2
floatc, intc = metrics
nt.assert_equal(floatc['metric'], 'float')
nt.assert_equal(floatc['points'][0][1], 0.5)
nt.assert_equal(floatc['host'], 'myhost')
nt.assert_equal(intc['metric'], 'int')
nt.assert_equal(intc['points'][0][1], 2)
nt.assert_equal(intc['host'], 'myhost')
def test_histogram_normalization(self):
stats = MetricsAggregator('myhost', interval=10)
for i in range(5):
stats.submit_packets('h1:1|h')
for i in range(20):
stats.submit_packets('h2:1|h')
metrics = self.sort_metrics(stats.flush())
_, _, h1count, _, _, _, _, h2count, _, _ = metrics
nt.assert_equal(h1count['points'][0][1], 0.5)
nt.assert_equal(h2count['points'][0][1], 2)
def test_tags(self):
stats = MetricsAggregator('myhost')
stats.submit_packets('gauge:1|c')
stats.submit_packets('gauge:2|c|@1')
stats.submit_packets('gauge:4|c|#tag1,tag2')
stats.submit_packets('gauge:8|c|#tag2,tag1') # Should be the same as above
stats.submit_packets('gauge:16|c|#tag3,tag4')
metrics = self.sort_metrics(stats.flush())
assert len(metrics) == 3
first, second, third = metrics
nt.assert_equal(first['metric'], 'gauge')
nt.assert_equal(first['dimensions'], None)
nt.assert_equal(first['points'][0][1], 3)
nt.assert_equal(first['host'], 'myhost')
nt.assert_equal(second['metric'], 'gauge')
nt.assert_equal(second['dimensions'], ('tag1', 'tag2'))
nt.assert_equal(second['points'][0][1], 12)
nt.assert_equal(second['host'], 'myhost')
nt.assert_equal(third['metric'], 'gauge')
nt.assert_equal(third['dimensions'], ('tag3', 'tag4'))
nt.assert_equal(third['points'][0][1], 16)
nt.assert_equal(third['host'], 'myhost')
def test_counter(self):
stats = MetricsAggregator('myhost')
# Track some counters.
stats.submit_packets('my.first.counter:1|c')
stats.submit_packets('my.first.counter:5|c')
stats.submit_packets('my.second.counter:1|c')
stats.submit_packets('my.third.counter:3|c')
# Ensure they roll up nicely.
metrics = self.sort_metrics(stats.flush())
assert len(metrics) == 3
first, second, third = metrics
nt.assert_equals(first['metric'], 'my.first.counter')
nt.assert_equals(first['points'][0][1], 6)
nt.assert_equals(first['host'], 'myhost')
nt.assert_equals(second['metric'], 'my.second.counter')
nt.assert_equals(second['points'][0][1], 1)
nt.assert_equals(third['metric'], 'my.third.counter')
nt.assert_equals(third['points'][0][1], 3)
# Ensure that counters reset to zero.
metrics = self.sort_metrics(stats.flush())
first, second, third = metrics
nt.assert_equals(first['metric'], 'my.first.counter')
nt.assert_equals(first['points'][0][1], 0)
nt.assert_equals(second['metric'], 'my.second.counter')
nt.assert_equals(second['points'][0][1], 0)
nt.assert_equals(third['metric'], 'my.third.counter')
nt.assert_equals(third['points'][0][1], 0)
def test_sampled_counter(self):
# Submit a sampled counter.
stats = MetricsAggregator('myhost')
stats.submit_packets('sampled.counter:1|c|@0.5')
metrics = stats.flush()
assert len(metrics) == 1
m = metrics[0]
assert m['metric'] == 'sampled.counter'
nt.assert_equal(m['points'][0][1], 2)
def test_gauge(self):
stats = MetricsAggregator('myhost')
# Track some counters.
stats.submit_packets('my.first.gauge:1|g')
stats.submit_packets('my.first.gauge:5|g')
stats.submit_packets('my.second.gauge:1.5|g')
# Ensure that gauges roll up correctly.
metrics = self.sort_metrics(stats.flush())
assert len(metrics) == 2
first, second = metrics
nt.assert_equals(first['metric'], 'my.first.gauge')
nt.assert_equals(first['points'][0][1], 5)
nt.assert_equals(first['host'], 'myhost')
nt.assert_equals(second['metric'], 'my.second.gauge')
nt.assert_equals(second['points'][0][1], 1.5)
# Ensure that old gauges get dropped due to old timestamps
stats.gauge('my.first.gauge', 5)
stats.gauge('my.first.gauge', 1, timestamp=1000000000)
stats.gauge('my.second.gauge', 20, timestamp=1000000000)
metrics = self.sort_metrics(stats.flush())
assert len(metrics) == 1
first = metrics[0]
nt.assert_equals(first['metric'], 'my.first.gauge')
nt.assert_equals(first['points'][0][1], 5)
nt.assert_equals(first['host'], 'myhost')
def test_sets(self):
stats = MetricsAggregator('myhost')
stats.submit_packets('my.set:10|s')
stats.submit_packets('my.set:20|s')
stats.submit_packets('my.set:20|s')
stats.submit_packets('my.set:30|s')
stats.submit_packets('my.set:30|s')
stats.submit_packets('my.set:30|s')
# Assert that it's treated normally.
metrics = stats.flush()
nt.assert_equal(len(metrics), 1)
m = metrics[0]
nt.assert_equal(m['metric'], 'my.set')
nt.assert_equal(m['points'][0][1], 3)
# Assert there are no more sets
assert not stats.flush()
def test_string_sets(self):
stats = MetricsAggregator('myhost')
stats.submit_packets('my.set:string|s')
stats.submit_packets('my.set:sets|s')
stats.submit_packets('my.set:sets|s')
stats.submit_packets('my.set:test|s')
stats.submit_packets('my.set:test|s')
stats.submit_packets('my.set:test|s')
# Assert that it's treated normally.
metrics = stats.flush()
nt.assert_equal(len(metrics), 1)
m = metrics[0]
nt.assert_equal(m['metric'], 'my.set')
nt.assert_equal(m['points'][0][1], 3)
# Assert there are no more sets
assert not stats.flush()
def test_rate(self):
stats = MetricsAggregator('myhost')
stats.submit_packets('my.rate:10|_dd-r')
# Sleep 1 second so the time interval > 0
time.sleep(1)
stats.submit_packets('my.rate:40|_dd-r')
# Check that the rate is calculated correctly
metrics = stats.flush()
nt.assert_equal(len(metrics), 1)
m = metrics[0]
nt.assert_equals(m['metric'], 'my.rate')
nt.assert_equals(m['points'][0][1], 30)
# Assert that no more rates are given
assert not stats.flush()
def test_gauge_sample_rate(self):
stats = MetricsAggregator('myhost')
# Submit a sampled gauge metric.
stats.submit_packets('sampled.gauge:10|g|@0.1')
# Assert that it's treated normally.
metrics = stats.flush()
nt.assert_equal(len(metrics), 1)
m = metrics[0]
nt.assert_equal(m['metric'], 'sampled.gauge')
nt.assert_equal(m['points'][0][1], 10)
def test_histogram(self):
stats = MetricsAggregator('myhost')
# Sample all numbers between 1-100 many times. This
# means our percentiles should be relatively close to themselves.
percentiles = range(100)
random.shuffle(percentiles) # in place
for i in percentiles:
for j in range(20):
for type_ in ['h', 'ms']:
m = 'my.p:%s|%s' % (i, type_)
stats.submit_packets(m)
metrics = self.sort_metrics(stats.flush())
def assert_almost_equal(i, j, e=1):
# Floating point math?
assert abs(i - j) <= e, "%s %s %s" % (i, j, e)
nt.assert_equal(len(metrics), 5)
p95, pavg, pcount, pmax, pmed = self.sort_metrics(metrics)
nt.assert_equal(p95['metric'], 'my.p.95percentile')
assert_almost_equal(p95['points'][0][1], 95, 10)
assert_almost_equal(pmax['points'][0][1], 99, 1)
assert_almost_equal(pmed['points'][0][1], 50, 2)
assert_almost_equal(pavg['points'][0][1], 50, 2)
assert_almost_equal(pcount['points'][0][1], 4000, 0) # 100 * 20 * 2
nt.assert_equals(p95['host'], 'myhost')
# Ensure that histograms are reset.
metrics = self.sort_metrics(stats.flush())
assert not metrics
def test_sampled_histogram(self):
# Submit a sampled histogram.
stats = MetricsAggregator('myhost')
stats.submit_packets('sampled.hist:5|h|@0.5')
# Assert we scale up properly.
metrics = self.sort_metrics(stats.flush())
p95, pavg, pcount, pmax, pmed = self.sort_metrics(metrics)
nt.assert_equal(pcount['points'][0][1], 2)
for p in [p95, pavg, pmed, pmax]:
nt.assert_equal(p['points'][0][1], 5)
def test_batch_submission(self):
# Submit a sampled histogram.
stats = MetricsAggregator('myhost')
metrics = [
'counter:1|c',
'counter:1|c',
'gauge:1|g'
]
packet = "\n".join(metrics)
stats.submit_packets(packet)
metrics = self.sort_metrics(stats.flush())
nt.assert_equal(2, len(metrics))
counter, gauge = metrics
assert counter['points'][0][1] == 2
assert gauge['points'][0][1] == 1
def test_bad_packets_throw_errors(self):
packets = [
'missing.value.and.type',
'missing.type:2',
'missing.value|c',
'2|c',
'unknown.type:2|z',
'string.value:abc|c',
'string.sample.rate:0|c|@abc',
# Bad event-like packets
'_ev{1,2}:bad_header'
'_e{1,}:invalid|headers',
'_e:missing|size|headers',
'_e:{1,1}:t|t|t:bad_meta|h',
]
stats = MetricsAggregator('myhost')
for packet in packets:
try:
stats.submit_packets(packet)
except Exception:
assert True
else:
assert False, 'invalid : %s' % packet
def test_metrics_expiry(self):
# Ensure metrics eventually expire and stop submitting.
stats = MetricsAggregator('myhost', expiry_seconds=1)
stats.submit_packets('test.counter:123|c')
# Ensure points keep submitting
assert stats.flush()
assert stats.flush()
time.sleep(0.5)
assert stats.flush()
# Now sleep for longer than the expiry window and ensure
# no points are submitted
time.sleep(2)
m = stats.flush()
assert not m, str(m)
# If we submit again, we're all good.
stats.submit_packets('test.counter:123|c')
assert stats.flush()
def test_histogram_counter(self):
# Test whether histogram.count == increment
# same deal with a sample rate
cnt = 100000
for run in [1, 2]:
stats = MetricsAggregator('myhost')
for i in range(cnt):
if run == 2:
stats.submit_packets('test.counter:1|c|@0.5')
stats.submit_packets('test.hist:1|ms|@0.5')
else:
stats.submit_packets('test.counter:1|c')
stats.submit_packets('test.hist:1|ms')
metrics = self.sort_metrics(stats.flush())
assert len(metrics) > 0
nt.assert_equal([m['points'][0][1]
for m in metrics if m['metric'] == 'test.counter'], [cnt * run])
nt.assert_equal([m['points'][0][1]
for m in metrics if m['metric'] == 'test.hist.count'], [cnt * run])
def test_scientific_notation(self):
stats = MetricsAggregator('myhost', interval=10)
stats.submit_packets('test.scinot:9.512901e-05|g')
metrics = self.sort_metrics(stats.flush())
assert len(metrics) == 1
ts, val = metrics[0].get('points')[0]
nt.assert_almost_equal(val, 9.512901e-05)
def test_event_tags(self):
stats = MetricsAggregator('myhost')
stats.submit_packets('_e{6,4}:title1|text')
stats.submit_packets('_e{6,4}:title2|text|#t1')
stats.submit_packets('_e{6,4}:title3|text|#t1,t2:v2,t3,t4')
stats.submit_packets('_e{6,4}:title4|text|k:key|p:normal|#t1,t2')
events = self.sort_events(stats.flush_events())
assert len(events) == 4
first, second, third, fourth = events
try:
first['dimensions']
except Exception:
assert True
else:
assert False, "event['tags'] shouldn't be defined when no tags aren't explicited in the packet"
nt.assert_equal(first['title'], 'title1')
nt.assert_equal(first['text'], 'text')
nt.assert_equal(second['title'], 'title2')
nt.assert_equal(second['text'], 'text')
nt.assert_equal(second['dimensions'], sorted(['t1']))
nt.assert_equal(third['title'], 'title3')
nt.assert_equal(third['text'], 'text')
nt.assert_equal(third['dimensions'], sorted(['t1', 't2:v2', 't3', 't4']))
nt.assert_equal(fourth['title'], 'title4')
nt.assert_equal(fourth['text'], 'text')
nt.assert_equal(fourth['aggregation_key'], 'key')
nt.assert_equal(fourth['priority'], 'normal')
nt.assert_equal(fourth['dimensions'], sorted(['t1', 't2']))
def test_event_title(self):
stats = MetricsAggregator('myhost')
stats.submit_packets('_e{0,4}:|text')
stats.submit_packets(u'_e{9,4}:2intitul<75><6C>|text')
stats.submit_packets('_e{14,4}:3title content|text')
stats.submit_packets('_e{14,4}:4title|content|text')
stats.submit_packets('_e{13,4}:5title\\ntitle|text') # \n stays escaped
events = self.sort_events(stats.flush_events())
assert len(events) == 5
first, second, third, fourth, fifth = events
nt.assert_equal(first['title'], '')
nt.assert_equal(second['title'], u'2intitul<EFBFBD><EFBFBD>')
nt.assert_equal(third['title'], '3title content')
nt.assert_equal(fourth['title'], '4title|content')
nt.assert_equal(fifth['title'], '5title\\ntitle')
def test_event_text(self):
stats = MetricsAggregator('myhost')
stats.submit_packets('_e{2,0}:t1|')
stats.submit_packets('_e{2,12}:t2|text|content')
stats.submit_packets('_e{2,23}:t3|First line\\nSecond line') # \n is a newline
stats.submit_packets(u'_e{2,19}:t4|<7C><><EFBFBD> <20><><EFBFBD><EFBFBD><EFBFBD>U <20><><EFBFBD><EFBFBD><EFBFBD>U <20><><EFBFBD><EFBFBD>u T0<54><30> <20><><EFBFBD>') # utf-8 compliant
events = self.sort_events(stats.flush_events())
assert len(events) == 4
first, second, third, fourth = events
nt.assert_equal(first['text'], '')
nt.assert_equal(second['text'], 'text|content')
nt.assert_equal(third['text'], 'First line\nSecond line')
nt.assert_equal(fourth['text'], u'<EFBFBD><EFBFBD><EFBFBD> <20><><EFBFBD><EFBFBD><EFBFBD>U <20><><EFBFBD><EFBFBD><EFBFBD>U <20><><EFBFBD><EFBFBD>u T0<54><30> <20><><EFBFBD>')
if __name__ == "__main__":
unittest.main()