5.4 KiB
2019.1.0
Cluster expiration emails sent for extended clusters
We're now sending cluster expiration emails when clusters are extended (for each extension). Previously the expiration email was only sent once for the initial cluster lifetime and not the extended lifetime(s).
2018.8.0
Larger EMR clusters for data engineers
For debugging purposes of larger datasets it's now possible to launch EMR clusters larger than the default of 30 nodes. This is especially useful for base datasets that are maintainer by the data pipeline team and therefor limited to that group of users at this time.
2017.10.0
Zeppelin notebooks in Spark jobs
In addition to uploading Jupyter/iPython Notebooks it's now also possible to upload Apache Zeppelin notebooks when scheduling Spark jobs.
2017.5.5
Spark jobs run history
For every scheduled Spark job you can now see the history of previous runs on the detail page in the "Runs" tab.
More accurate timestamps
For both on-demand clusters as well as scheduled Spark jobs we now record more details about the start, ready and finish time of the AWS EMR clusters. The timestamps are shown both on the dashboard as well as the cluster and Spark job detail pages.
2017.5.2
Site-wide announcements
Occasionally we may want to inform you about the current operational status of the Telemetry Analysis Service and will do so now using site-wide announcements at the top of every page.
For example in case we're facing partial degredation of service due to problems with upstream providers we'll post an update.
Run your scheduled Spark jobs right now
Every scheduled Spark job is now able to be run individually out-of-schedule with the new "Run now" button at the top of the Spark job detail page.
To prevent data loss by jobs that are run repeatedly we will skip the next scheduled run in case it hasn't finished by the scheduled time. Please make sure to plan accordingly. The Spark job detail page lists the time of the next run.
New unique identifiers for clusters
When creating new on-demand clusters or scheduling new Spark jobs we're now providing random and unique identifiers for better operational monitoring. They can of course still be overridden with custom names.
2017.5.1
Visual indicators for Spark job cluster status
We now show visual indicators for the current status of the cluster of scheduled Spark jobs on the dashboard and the detail page.
2017.4.3
Added ability to define and extend cluster lifetime
Backward-incompatible change!
Please note that running clusters are not affected, only new ones.On-demand Spark clusters have a new default lifetime of 8 hours.
You can increase the lifetime to 24 hours using the new optional Lifetime form field when launching a Spark cluster.
Additionally you can extend the lifetime of a running cluster with the
Extend button on the cluster detail page (next to Terminate now button).
The cluster expiration emails will also contain a link to extend the cluster if needed.
Experimental and deprecated EMR releases
When launching an on-demand Spark cluster or scheduling a Spark job, please be aware that some EMR releases may in the future be marked as experimental or deprecated to better explain the expectations around the support we can provide for those.
In case of a deprecated EMR release we'll announce a expected deprecation window for the version here and on the appropriate mailing lists.
Experimental EMR releases are marked as such to be able to vet them in production. They are available as any other release but are not recommended for critical Spark clusters or scheduled jobs.
New email notifications for expired and timed out Spark jobs
When a scheduled Spark job has expired we'll send out an email notifiying the owner of it. They may un-expire it by updating the end date of the Spark job again.
When a Spark job has timed out because it ran longer than the configured timeout period (up to 24 hours) we will terminate the Spark job and send an email to the owner requesting to modify the job to run less than the timeout, e.g. by increasing the number of cluster nodes, improving the Spark job notebook code etc.
2017.3.7
All Spark jobs visible for maintainers
Users who are members of the "Spark job maintainers" permission group can now optionally see the Spark jobs of all users in the dashboard.
2017.3.2
Persistent cluster storage
Good news everyone - files are now persisted between clusters! In fact, make changes to your own dotfiles in your home dir - they will live on between clusters as well.
We will be limiting your home directory size to 20GB, so don't go saving those big honkin' parquet files.
EMR 4.5.0 release removed
The EMR 4.5.0 release was removed and won't be available for new on-demand clusters and scheduled Spark jobs.
Email alerts on scheduled jobs failures
When ATMO discovers that a scheduled job resulted in a failed cluster state the creator of the scheduled job will receive an email alerting them of the situation.
New out-of-date notifications
The previous automatic page refresher that was active on the dashboard, the cluster and job detail pages has been replaced by an in-page-notification when the page has changed, e.g. when a cluster status is changed on AWS.