No statistics error messages

[davidkyle]

I spotted these errors in a test with the following set up:
detector sum(total) - 'total' is the field name
bucketspan 1s

The test generates 1 data point a second for 30 seconds, processing the data produces these errors in the log file.

[CCountingModel.cc@120] No statistics at 0, current bucket = [1,2)
[CJsonOutputWriter.cc@174] Simple count detector has no current rate
[CCountingModel.cc@120] No statistics at 1, current bucket = [2,3)
[CJsonOutputWriter.cc@174] Simple count detector has no current rate
[CCountingModel.cc@120] No statistics at 2, current bucket = [3,4)
...
[CCountingModel.cc@120] No statistics at 28, current bucket = [29,30)
[CJsonOutputWriter.cc@174] Simple count detector has no current rate

Beyond the missing rate field there doesn't appear to be any harmful effects.

Reproduce
In the root of the elasticsearch repo open the file client/rest-high-level/src/test/java/org/elasticsearch/client/MachineLearningIT.java and changes line 433

configBuilder.setBucketSpan(new TimeValue(randomIntBetween(1, 10), TimeUnit.SECONDS));

to

configBuilder.setBucketSpan(new TimeValue(1, TimeUnit.SECONDS));

ie hard code the bucket span to 1 second.

Run the test with:

./gradlew :client:rest-high-level:integTest -Dtests.class=org.elasticsearch.client.documentation.MlClientDocumentationIT -Dtests.method=testDeleteForecast

The log file will appear at client/rest-high-level/build/cluster/integTestCluster node0/elasticsearch-7.0.0-alpha1-SNAPSHOT/logs/client_rest-high-level_integTestCluster.log

[davidkyle]

I saw this message again in MachineLearningIT#testStartDatafeed this time it spammed the log file with the message repeated thousands of times.

Once again it appears the job has a 1 second bucket span (from the log file, the test randomises the bucket span)
No statistics at 1537589302, current bucket = [1537589303,1537589304)

[benwtrent]

@dimitris-athanasiou has a good theory around this and partial buckets.

[benwtrent]

When doing some investigation, if the start of the data in the index is guaranteed to be at a second mark and each document is at one second intervals, the one second bucket-span poses no issues.