Handling delayed dataedit

Delayed data are documents that are indexed late. That is to say, it is data related to a time that your datafeed has already processed and it is therefore never analyzed by your anomaly detection job.

When you create a datafeed, you can specify a query_delay setting. This setting enables the datafeed to wait for some time past real-time, which means any "late" data in this period is fully indexed before the datafeed tries to gather it. However, if the setting is set too low, the datafeed may query for data before it has been indexed and consequently miss that document. Conversely, if it is set too high, analysis drifts farther away from real-time. The balance that is struck depends upon each use case and the environmental factors of the cluster.

If you get an error that says Datafeed missed XXXX documents due to ingest latency, consider increasing the value of ‘query_delay’. If it doesn’t help, investigate the ingest latency and its cause. You can do this by comparing event and ingest timestamps. High latency is often caused by bursts of ingested documents, misconfiguration of the ingest pipeline, or misalignment of system clocks.

Why worry about delayed data?edit

If data are delayed randomly (and consequently are missing from analysis), the results of certain types of functions are not really affected. In these situations, it all comes out okay in the end as the delayed data is distributed randomly. An example would be a mean metric for a field in a large collection of data. In this case, checking for delayed data may not provide much benefit. If data are consistently delayed, however, anomaly detection jobs with a low_count function may provide false positives. In this situation, it would be useful to see if data comes in after an anomaly is recorded so that you can determine a next course of action.

How do we detect delayed data?edit

In addition to the query_delay field, there is a delayed data check config, which enables you to configure the datafeed to look in the past for delayed data. Every 15 minutes or every check_window, whichever is smaller, the datafeed triggers a document search over the configured indices. This search looks over a time span with a length of check_window ending with the latest finalized bucket. That time span is partitioned into buckets, whose length equals the bucket span of the associated anomaly detection job. The doc_count of those buckets are then compared with the job’s finalized analysis buckets to see whether any data has arrived since the analysis. If there is indeed missing data due to their ingest delay, the end user is notified. For example, you can see annotations in Kibana for the periods where these delays occur:

Delayed data annotations in the Single Metric Viewer

As the doc_count from an aggregation is compared with the bucket results of the job, the delayed data check will not work correctly in the following cases:

  • if the datafeed uses aggregations and the job’s analysis_config does not have its summary_count_field_name set to doc_count,
  • if the datafeed is not using aggregations and summary_count_field_name is set to any value.

If the datafeed is using aggregations then it’s highly likely that the job’s summary_count_field_name should be set to doc_count. If summary_count_field_name is set to any value other than doc_count, the delayed data check for the datafeed must be disabled.

There is another tool for visualizing the delayed data on the Annotations tab in the anomaly detection job management page:

Delayed data in the datafeed chart

What to do about delayed data?edit

The most common course of action is to simply to do nothing. For many functions and situations, ignoring the data is acceptable. However, if the amount of delayed data is too great or the situation calls for it, the next course of action to consider is to increase the query_delay of the datafeed. This increased delay allows more time for data to be indexed. If you have real-time constraints, however, an increased delay might not be desirable. In which case, you would have to tune for better indexing speed.