[ML] Adding an open job to a group does not update autodetect with related scheduled events

[davidkyle]

Steps to reproduce:

1. Create a calendar with a scheduled event
2. Add a new job group to the calendar
3. Create a job and open it. In the UI this is achieved by creating the job but not starting the datafeed
4. Edit the job to add it to the new group
5. Run the job and see that is has not picked up the calendar associated with the group

The problem is that updating the job with a group does not update autodetect with scheduled events.

cc @pheyos

[elasticmachine]

Pinging @elastic/ml-core

[jucaleb4]

Hi, can I take a stab at this? @davidkyle @pheyos

[jucaleb4]

I tried reproducing your steps and got this, it looks like the job job_id2 picked up the calendar planned-outages when they shared a common job and running it. Am I missing a step, as it looks like this solves step 5?

[davidkyle]

Hi @jucaleb4 thank for your interest I'm delighted you want to take a stab at this.

For background on ML Calendars and Scheduled events see the docs, the basic idea is to define periods of time where anomalies won't be generated.

Below are the detailed steps to reproduce, you will need some sample data and a script into index it I've created a gist for that purpose.

1. Extract the files from the gist and index the farequote dataset by running python upload_farequote.py with the file farequote_epoch_ms.csv in the same directory

2. Create the index pattern farequote for the farequote index

3. In Maching Learning create a single metric job on the farequote index pattern.

4. Set the aggregation to count and run the job. Notice the single large anomaly on the 2016/06/25

5. Go to Machine Learning -> Settings -> Calendar Management

6. Create a new Calendar farequote

7. Add the new group farequoute-group to the calendar

8. Add a new event for a single day on 2016/06/25 with the description farequote anomaly

9. Save the calendar

10. Clone the original farequote job and name it farequote-with-calendar and add the Job Group farequote-group

11. Save the job and click Start datafeed then Start

12. View the job results and notice that there is no anomaly on 2016/06/25

This time the calendar worked as expected as no anomalies were created you should see results similar to this.

The bug occurs when you add the job to a group after the job has been created. To recreate the bug:

1. Clone the original farequote job again. Name it farequote-bug and don't add the job group

2. Click Save, click Ok to the influencers question then click Close instead of Start Datafeed. At this point the job is open.

3. Go back the the Job Management page and edit the job farequote-bug (edit is one of the actions on the right of the screen)

4. Add the job group farequote-group and click Update

5. Now start the datafeed for job farequote-bug (the first action under actions on the right of the screen)

6. View the results. The Calendar entry is not picked up and an anomaly is created on 2016/06/25.

The job is part of the job group which uses the farequote calendar. All jobs in the group should pick up the calendar entries.

The relevant code is in JobManager.updateJob, in particular when the job is changed by adding a group it needs to send an update to the native ML process

[jucaleb4]

This will help me get started, thanks @davidkyle. If I have anymore questions, I'll let you know along the way.

[jucaleb4]

@davidkyle Hi David, so the current elasticsearch source code is es7.0. However, the kibana download version I have is 6.3.0. So when I try to run ES, the Kibana complains about the type mismatch. Is there a way I can either a) downgrade es to 6.3.0 or b) upgrade kibana to 7.0 to test this code?

edit: figured it out!

[droberts195]

Hi @jucaleb4. Are you still planning to submit a PR for this?