Fix flaky custom metric test.

Change-Id: Ia2628f205537099d28415412fe302763f17c3081

samples/snippets/v3/custom_metric_test.py

@@ -25,6 +25,7 @@
 
 from custom_metric import create_custom_metric, get_custom_metric
 from custom_metric import read_timeseries, write_timeseries_value
+from gcp.testing import eventually_consistent
 from gcp.testing.flaky import flaky
 import list_resources
 
@@ -53,9 +54,10 @@ def test_custom_metric(cloud_config):
 
  create_custom_metric(
  client, PROJECT_RESOURCE, METRIC_RESOURCE, METRIC_KIND)
- custom_metric = None
+
  # wait until metric has been created, use the get call to wait until
  # a response comes back with the new metric
+ custom_metric = None
  while not custom_metric:
  time.sleep(1)
  custom_metric = get_custom_metric(
@@ -64,12 +66,13 @@ def test_custom_metric(cloud_config):
  write_timeseries_value(client, PROJECT_RESOURCE,
  METRIC_RESOURCE, INSTANCE_ID,
  METRIC_KIND)
+
  # Sometimes on new metric descriptors, writes have a delay in being
- # read back. 3 seconds should be enough to make sure our read call
- # picks up the write
- time.sleep(3)
- response = read_timeseries(client, PROJECT_RESOURCE, METRIC_RESOURCE)
- value = int(
- response['timeSeries'][0]['points'][0]['value']['int64Value'])
- # using seed of 1 will create a value of 1
- assert value == pseudo_random_value
+ # read back. Use eventually_consistent to account for this.
+ @eventually_consistent.call
+ def _():
+  response = read_timeseries(client, PROJECT_RESOURCE, METRIC_RESOURCE)
+  value = int(
+  response['timeSeries'][0]['points'][0]['value']['int64Value'])
+  # using seed of 1 will create a value of 1
+  assert value == pseudo_random_value