Investigating Increased Snowflake Errors in the US Region

We are encountering increased error rates from Snowflake in the US Region and are currently investigating the cause.
We will update here again as soon as we have relevant information or if 60 minutes have passed.

Update Oct 27, 2020 17:17 UTC:  The error rate has returned to normal since 16:55 UTC.  We are continuing to investigate the root cause.

Update Oct 27, 2020 18:57 UTC: Snowflake has released a status report.  The errors continue to occur, but at a reduced rate.

Update Oct 27, 2020 19:29 UTC: Snowflake has reported the incident as resolved.

Investigating Snowflake Performance in the EU Region

We are experiencing a minor slow down of performance on Snowflake in the EU Region and are currently investigating.
There are no job failures or increased queue backlog but everything seems to run slightly slower.
The next update will be in 120 minutes or as new information becomes available.

Update October 26 2020, 12:50 UTC: Performance has for the most part recovered, the rate of long execution of some queries has dropped significantly, but hasn't quite completely disappeared.  We will update again if the issue re-surfaces.

Week in Review - October 20th, 2020

New Components

  • POP3 Attachment extractor – allows you to download email attachments from a POP3 mailbox. There is no restriction on attachment type, so you can also download XLS files or images.

Updated Components

  • MongoDB extractor – added an option to “Include parent in Primary Key” in Mapping mode. This will allow better PK generation and the behavior will be much more consistent with Generic Extractor.
  • Looker writer – added support for Google BigQuery as the underlying database.

UI Improvements

Selecting a whole bucket in Input Mapping is now possible even if there are collisions between table names, or if there’s already a table from the specific bucket:
  • If there's a name collision, a numeric suffix is added (for example, if there is an account table from another bucket and you add a new bucket which has a table with the same name, it will be named account_2).
  • If there are already tables from the bucket, only the remaining tables will be added.

Force Unlink

There’s a new option to “force unlink” a bucket from projects that have it linked. When you disable sharing, it will “force unlink” the bucket automatically.

Edit Bucket Sharing

From now on, if there are projects that have the bucket linked, you can only add more projects/users. Edit sharing is available from “shared from this projects” list and from “bucket detail in data catalog” pages (if you have sufficient privileges). Until the bucket is linked, sharing can be changed without limitations.






Investigating EU overload

We experienced overload in EU

Since 2020-10-15 10:02 UTC Start investigating overload in EU. Next update in 1 hour or as new information

UPDATE 2020-10-15 11:14 UTC: All operations are back to normal and everything is fully working. No jobs should be affected.

Investigating EU overload

We experienced overload in EU

Since 2020-10-12 11:40 UTC Start investigating overload in EU. Next update in 1 hour or as new information becomes available.

UPDATE 2020-10-12 12:56 UTC:  Everything should be back to normal. No jobs should be affected. We'll keep monitoring our platform closely.

UPDATE 2020-10-12 13:57 UTC: All operations are back to normal and everything is fully working.

Telemetry Data Fixes and Upcoming Telemetry Component

Dear customers,

We are in the final phase of shifting and finalizing the calculation of time credits. We thank you for your patience and partnership in working with us and transitioning your contracts over to the new credit system.

As a part of finalizing, we are implementing some changes that weren’t applied previously and will begin to calculate starting now.

  • Transformations with multiple transformation backends - If you have a transformation that runs an SQL transformation with a R/Python transformation following it, in the same bucket, the SQL transformation is being calculated until the R/Python transformations are finished. As a result, these SQL jobs will have its child jobs run time deducted from their own, resulting in lower time credits consumption of affected clients. The change will cover jobs since the beginning of 2020.
  • SQL sandboxes time credits usage - Most of our clients don’t have SQL sandboxes consumption added to their overall consumption. In our analysis, this is a fraction of total consumption, so the impact is not very material for the majority of clients. For those clients that have a material impact, your CSM will be reaching out to you to let you know the impact, otherwise you can assume your usage is not going to impact you.
  • COVID-19 Error jobs - We haven’t been calculating credits for error jobs since April 2020 due to Covid and a higher than normal rate of Snowflake related issues. We will begin adding the consumption from job errors starting October.



Telemetry Component

We’re planning to release a new Keboola Connection component, which will allow the users to get telemetry data about their project or organization. That will eventually replace our temporary solutions like GoodData telemetry dashboard or direct writing of telemetry data to some projects. In the end, all of the Keboola Connection users will have easy access to documented telemetry data on demand.

Snowflake issues in US region

We're investigating an issue with Snowflake in US region which causes some Storage operations with tables to be stuck in processing state. This can cause jobs to be executing longer than expected or seemingly "forever". Terminating and restarting the job does not help in such a case. Only certain projects are affected.

Next update in 1 hour or as new information becomes available.

Update 10:50 UTC: The stuck jobs are unblocked now and should be finishing, we're monitoring the situation if the issue reappears. A post mortem will be published once we get an RCA from Snowflake.


Keboola-provided credentials for Snowflake and Redshift database writers

When configuring a Snowflake or Redshift database writer, you can use a Keboola-provided database.

In the past, when you selected this option, the credentials were stored in a configuration in a plain-text format. Storing the credentials this way allowed you to copy the password and use it in your favorite database client (or another system) even if you didn't copy it right after the creation.

To improve the overall security, we decided to show you your password only once and store it encrypted. From now on, when you create a new Keboola-provided database (Snowflake or Redshift), you will see the password only once, right after its creation.

Backward compatibility

The existing credentials will remain untouched. But if you delete them, there's no option to create them the old way.

Week in Review - September 14th, 2020

New Components

  • LiveRamp Identity Resolution application - solving some of the main challenges with customer and prospect data by returning people-based identifiers and metadata for your consumer records

  • KBC Project Metadata extractor - Keboola metadata extractor downloads metadata about all objects in your Keboola project.

  • Avro2CSV processor - Avro is a row-oriented remote procedure call and data serialization framework developed within Apache's Hadoop project. It uses JSON for defining data types and protocols, and serializes data in a compact binary format.

Updated Components

  • Generic extractor - added option "caCertificate" which allows you configure custom certificate authority bundle in crt/pem format. (documentation)

Minor Improvements

  • Google BigQueryupdated google-cloud-bigquery package

  • Python updated to 3.8.5

  • Julia updated to 1.5.0