moto/tests/test_cloudwatch/test_cloudwatch_boto3.py
2020-03-16 20:14:41 +01:00

294 lines
11 KiB
Python

# from __future__ import unicode_literals
import boto3
from botocore.exceptions import ClientError
from datetime import datetime, timedelta
from nose.tools import assert_raises
from uuid import uuid4
import pytz
import sure # noqa
from moto import mock_cloudwatch
@mock_cloudwatch
def test_put_list_dashboard():
client = boto3.client("cloudwatch", region_name="eu-central-1")
widget = '{"widgets": [{"type": "text", "x": 0, "y": 7, "width": 3, "height": 3, "properties": {"markdown": "Hello world"}}]}'
client.put_dashboard(DashboardName="test1", DashboardBody=widget)
resp = client.list_dashboards()
len(resp["DashboardEntries"]).should.equal(1)
@mock_cloudwatch
def test_put_list_prefix_nomatch_dashboard():
client = boto3.client("cloudwatch", region_name="eu-central-1")
widget = '{"widgets": [{"type": "text", "x": 0, "y": 7, "width": 3, "height": 3, "properties": {"markdown": "Hello world"}}]}'
client.put_dashboard(DashboardName="test1", DashboardBody=widget)
resp = client.list_dashboards(DashboardNamePrefix="nomatch")
len(resp["DashboardEntries"]).should.equal(0)
@mock_cloudwatch
def test_delete_dashboard():
client = boto3.client("cloudwatch", region_name="eu-central-1")
widget = '{"widgets": [{"type": "text", "x": 0, "y": 7, "width": 3, "height": 3, "properties": {"markdown": "Hello world"}}]}'
client.put_dashboard(DashboardName="test1", DashboardBody=widget)
client.put_dashboard(DashboardName="test2", DashboardBody=widget)
client.put_dashboard(DashboardName="test3", DashboardBody=widget)
client.delete_dashboards(DashboardNames=["test2", "test1"])
resp = client.list_dashboards(DashboardNamePrefix="test3")
len(resp["DashboardEntries"]).should.equal(1)
@mock_cloudwatch
def test_delete_dashboard_fail():
client = boto3.client("cloudwatch", region_name="eu-central-1")
widget = '{"widgets": [{"type": "text", "x": 0, "y": 7, "width": 3, "height": 3, "properties": {"markdown": "Hello world"}}]}'
client.put_dashboard(DashboardName="test1", DashboardBody=widget)
client.put_dashboard(DashboardName="test2", DashboardBody=widget)
client.put_dashboard(DashboardName="test3", DashboardBody=widget)
# Doesnt delete anything if all dashboards to be deleted do not exist
try:
client.delete_dashboards(DashboardNames=["test2", "test1", "test_no_match"])
except ClientError as err:
err.response["Error"]["Code"].should.equal("ResourceNotFound")
else:
raise RuntimeError("Should of raised error")
resp = client.list_dashboards()
len(resp["DashboardEntries"]).should.equal(3)
@mock_cloudwatch
def test_get_dashboard():
client = boto3.client("cloudwatch", region_name="eu-central-1")
widget = '{"widgets": [{"type": "text", "x": 0, "y": 7, "width": 3, "height": 3, "properties": {"markdown": "Hello world"}}]}'
client.put_dashboard(DashboardName="test1", DashboardBody=widget)
resp = client.get_dashboard(DashboardName="test1")
resp.should.contain("DashboardArn")
resp.should.contain("DashboardBody")
resp["DashboardName"].should.equal("test1")
@mock_cloudwatch
def test_get_dashboard_fail():
client = boto3.client("cloudwatch", region_name="eu-central-1")
try:
client.get_dashboard(DashboardName="test1")
except ClientError as err:
err.response["Error"]["Code"].should.equal("ResourceNotFound")
else:
raise RuntimeError("Should of raised error")
@mock_cloudwatch
def test_alarm_state():
client = boto3.client("cloudwatch", region_name="eu-central-1")
client.put_metric_alarm(
AlarmName="testalarm1",
MetricName="cpu",
Namespace="blah",
Period=10,
EvaluationPeriods=5,
Statistic="Average",
Threshold=2,
ComparisonOperator="GreaterThanThreshold",
)
client.put_metric_alarm(
AlarmName="testalarm2",
MetricName="cpu",
Namespace="blah",
Period=10,
EvaluationPeriods=5,
Statistic="Average",
Threshold=2,
ComparisonOperator="GreaterThanThreshold",
)
# This is tested implicitly as if it doesnt work the rest will die
client.set_alarm_state(
AlarmName="testalarm1",
StateValue="ALARM",
StateReason="testreason",
StateReasonData='{"some": "json_data"}',
)
resp = client.describe_alarms(StateValue="ALARM")
len(resp["MetricAlarms"]).should.equal(1)
resp["MetricAlarms"][0]["AlarmName"].should.equal("testalarm1")
resp["MetricAlarms"][0]["StateValue"].should.equal("ALARM")
resp["MetricAlarms"][0]["ActionsEnabled"].should.equal("True")
resp = client.describe_alarms(StateValue="OK")
len(resp["MetricAlarms"]).should.equal(1)
resp["MetricAlarms"][0]["AlarmName"].should.equal("testalarm2")
resp["MetricAlarms"][0]["StateValue"].should.equal("OK")
resp["MetricAlarms"][0]["ActionsEnabled"].should.equal("True")
# Just for sanity
resp = client.describe_alarms()
len(resp["MetricAlarms"]).should.equal(2)
@mock_cloudwatch
def test_put_metric_data_no_dimensions():
conn = boto3.client("cloudwatch", region_name="us-east-1")
conn.put_metric_data(
Namespace="tester", MetricData=[dict(MetricName="metric", Value=1.5)]
)
metrics = conn.list_metrics()["Metrics"]
metrics.should.have.length_of(1)
metric = metrics[0]
metric["Namespace"].should.equal("tester")
metric["MetricName"].should.equal("metric")
@mock_cloudwatch
def test_put_metric_data_with_statistics():
conn = boto3.client("cloudwatch", region_name="us-east-1")
utc_now = datetime.now(tz=pytz.utc)
conn.put_metric_data(
Namespace="tester",
MetricData=[
dict(
MetricName="statmetric",
Timestamp=utc_now,
# no Value to test https://github.com/spulec/moto/issues/1615
StatisticValues=dict(
SampleCount=123.0, Sum=123.0, Minimum=123.0, Maximum=123.0
),
Unit="Milliseconds",
StorageResolution=123,
)
],
)
metrics = conn.list_metrics()["Metrics"]
metrics.should.have.length_of(1)
metric = metrics[0]
metric["Namespace"].should.equal("tester")
metric["MetricName"].should.equal("statmetric")
# TODO: test statistics - https://github.com/spulec/moto/issues/1615
@mock_cloudwatch
def test_get_metric_statistics():
conn = boto3.client("cloudwatch", region_name="us-east-1")
utc_now = datetime.now(tz=pytz.utc)
conn.put_metric_data(
Namespace="tester",
MetricData=[dict(MetricName="metric", Value=1.5, Timestamp=utc_now)],
)
stats = conn.get_metric_statistics(
Namespace="tester",
MetricName="metric",
StartTime=utc_now - timedelta(seconds=60),
EndTime=utc_now + timedelta(seconds=60),
Period=60,
Statistics=["SampleCount", "Sum"],
)
stats["Datapoints"].should.have.length_of(1)
datapoint = stats["Datapoints"][0]
datapoint["SampleCount"].should.equal(1.0)
datapoint["Sum"].should.equal(1.5)
@mock_cloudwatch
def test_list_metrics():
cloudwatch = boto3.client("cloudwatch", "eu-west-1")
# Verify namespace has to exist
res = cloudwatch.list_metrics(Namespace="unknown/")["Metrics"]
res.should.be.empty
# Create some metrics to filter on
create_metrics(cloudwatch, namespace="list_test_1/", metrics=4, data_points=2)
create_metrics(cloudwatch, namespace="list_test_2/", metrics=4, data_points=2)
# Verify we can retrieve everything
res = cloudwatch.list_metrics()["Metrics"]
len(res).should.equal(16) # 2 namespaces * 4 metrics * 2 data points
# Verify we can filter by namespace/metric name
res = cloudwatch.list_metrics(Namespace="list_test_1/")["Metrics"]
len(res).should.equal(8) # 1 namespace * 4 metrics * 2 data points
res = cloudwatch.list_metrics(Namespace="list_test_1/", MetricName="metric1")[
"Metrics"
]
len(res).should.equal(2) # 1 namespace * 1 metrics * 2 data points
# Verify format
res.should.equal(
[
{u"Namespace": "list_test_1/", u"Dimensions": [], u"MetricName": "metric1"},
{u"Namespace": "list_test_1/", u"Dimensions": [], u"MetricName": "metric1"},
]
)
# Verify unknown namespace still has no results
res = cloudwatch.list_metrics(Namespace="unknown/")["Metrics"]
res.should.be.empty
@mock_cloudwatch
def test_list_metrics_paginated():
cloudwatch = boto3.client("cloudwatch", "eu-west-1")
# Verify that only a single page of metrics is returned
cloudwatch.list_metrics()["Metrics"].should.be.empty
# Verify we can't pass a random NextToken
with assert_raises(ClientError) as e:
cloudwatch.list_metrics(NextToken=str(uuid4()))
e.exception.response["Error"]["Message"].should.equal(
"Request parameter NextToken is invalid"
)
# Add a boatload of metrics
create_metrics(cloudwatch, namespace="test", metrics=100, data_points=1)
# Verify that a single page is returned until we've reached 500
first_page = cloudwatch.list_metrics()
first_page["Metrics"].shouldnt.be.empty
len(first_page["Metrics"]).should.equal(100)
create_metrics(cloudwatch, namespace="test", metrics=200, data_points=2)
first_page = cloudwatch.list_metrics()
len(first_page["Metrics"]).should.equal(500)
first_page.shouldnt.contain("NextToken")
# Verify that adding more data points results in pagination
create_metrics(cloudwatch, namespace="test", metrics=60, data_points=10)
first_page = cloudwatch.list_metrics()
len(first_page["Metrics"]).should.equal(500)
first_page["NextToken"].shouldnt.be.empty
# Retrieve second page - and verify there's more where that came from
second_page = cloudwatch.list_metrics(NextToken=first_page["NextToken"])
len(second_page["Metrics"]).should.equal(500)
second_page.should.contain("NextToken")
# Last page should only have the last 100 results, and no NextToken (indicating that pagination is finished)
third_page = cloudwatch.list_metrics(NextToken=second_page["NextToken"])
len(third_page["Metrics"]).should.equal(100)
third_page.shouldnt.contain("NextToken")
# Verify that we can't reuse an existing token
with assert_raises(ClientError) as e:
cloudwatch.list_metrics(NextToken=first_page["NextToken"])
e.exception.response["Error"]["Message"].should.equal(
"Request parameter NextToken is invalid"
)
def create_metrics(cloudwatch, namespace, metrics=5, data_points=5):
for i in range(0, metrics):
metric_name = "metric" + str(i)
for j in range(0, data_points):
cloudwatch.put_metric_data(
Namespace=namespace,
MetricData=[{"MetricName": metric_name, "Value": j, "Unit": "Seconds"}],
)