Elasticsearch X-Pack version 6.1.0

Breaking Changes

  • The certgen command now returns validation errors when it encounters problems reading from an input file (with the -in command option). Previously these errors might have been ignored or caused the command to abort with unclear messages. For more information, see certgen.

New Features

Machine Learning
  • Added the ability to create job forecasts. This feature enables you to use historical behavior to predict the future behavior of your time series. You can create forecasts in Kibana or by using the forecast jobs API.

    You cannot create forecasts for jobs that were created in previous versions; this functionality is available only for jobs created in 6.1 or later.

  • Added overall buckets, which summarize bucket results for multiple jobs. For more information, see the get overall buckets API.
  • Added job groups, which you can use to manage or retrieve information from multiple jobs at once. Also updated many machine learning APIs to support groups and wildcard expressions in the job identifier.


Machine Learning
  • Improved the way machine learning jobs are allocated to nodes, such that it is primarily determined by the estimated memory requirement of the job. If there is insufficient information about the job’s memory requirements, the allocation decision is based on job counts per node.
  • Increased the default value of the xpack.ml.max_open_jobs setting from 10 to 20. The allocation of jobs to nodes now considers memory usage as well as job counts, so it’s reasonable to permit more small jobs on a single node. For more information, see Machine Learning Settings.
  • Decreased the default model_memory_limit property value to 1 GB for new jobs. If you want to create a job that analyzes high cardinality fields, you can increase this property value. For more information, see Analysis Limits.
  • Improved analytics related to decay rates when predictions are very accurate.
  • Improved analytics related to detecting non-negative quantities and using this information to constrain analysis, predictions, and confidence intervals.
  • Improved periodic trough or spike detection.
  • Improved the speed of the aggregation of machine learning results.
  • Improved probability calculation performance.
  • Expedited bucket processing time in very large populations by determining when there are nearly duplicate values in a bucket and de-duplicating the samples that are added to the model.
  • Improved handling of periodically missing values.
  • Improved analytics related to diurnal periodicity.
  • Reduced memory usage during population analysis by releasing redundant memory after the bucket results are written.
  • Improved modeling of long periodic components, particularly when there is a long bucket span.
  • Added the new interval_ms field to monitoring documents. This field indicates the current collection interval for Elasticsearch or external monitored systems.
  • Added the manage_index_templates cluster privilege to the built-in role kibana_system. For more information, see Cluster Privileges and Built-in Roles.
  • Newly created or updated watches execute with the privileges of the user that last modified the watch.
  • Added log messages when a PEM key is found when a PEM certificate was expected (or vice versa) in the xpack.ssl.key or xpack.ssl.certificate settings.
  • Added the new certutil command to simplify the creation of certificates for use with the Elastic stack. For more information, see certutil.
  • Added automatic detection of support for AES 256 bit TLS ciphers and enabled their use when the JVM supports them.
  • Improved error messages when there are no accounts configured for Watcher.
  • Added thread pool rejection information to execution state, which makes it easier to debug execution failures.
  • Added execution state information to watch status details. It is stored in the status.execution_state field.
  • Enabled the account monitoring url field in the xpack.notification.jira setting to support customized paths. For more information about configuring Jira accounts for use with watches, see Jira Action.
  • Improved handling of exceptions in Watcher to make it easier to debug problems.

Bug Fixes

Machine Learning
  • Improved handling of scenarios where there are insufficient values to interpolate trend components.
  • Improved calculation of confidence intervals.
  • Fixed degrees of freedom calculation that could lead to excessive error logging.
  • Improved trend modeling with long bucket spans.
  • Fixed timing of when model size statistics are written. Previously, if there were multiple partitions, there could be multiple model size stats docs written within the same bucket.
  • Updated the calculation of the model memory to include the memory used by partition, over, by, or influencer fields.
  • Fixed calculation of the frequency property value for datafeeds that use aggregations. The value must be a multiple of the histogram interval. For more information, see Aggregating Data for Faster Performance.
  • Removed unnecessary messages from logs when a job is forcefully closed.
  • Fixed REST requests that required a body but did not validate it, resulting in null pointer exceptions.
  • Fixed handling of watcher templates. Missing watcher templates can be added by any node if that node has a higher version than the master node.