By changing the instance types for the AWS Aurora cluster that powers the flame icon and content recommendations from r3.4xlarge to r3.xlarge we were able to save 75% on the EC2 instance monthly costs for that cluster.
Clean up logic around associate ids to make it easier to debug
We have updated our logic for skimlinks as a proof of concept for subtags:
We only get ASIN descriptions from Amazon’s earnings report, not via their Subtag reports. We’ve done some work on our end to take ASIN from the Subtag report and match with a description from the Earnings report and expose in Looker.
Ingestions were turned off previously, but data was still being stored in FMG warehouses. We’ve transferred all non-backfillable data to storage with UCI teams, and deleted what was stored on our end.
While the urls were correctly passing through associate-id, we were losing subtag data. All amazon urls should now be ignored by the skimlinks crawler.
Commerce insets whitelisted on uBlock (3/11) and AdBlock Plus (3/19) via EasyList
Our process for adding subtags happens in two places: once in the backend to store the hash and metadata, and once in the presentation layer to attach to the link.
Adding PVs by Device, and Sessions from GA
We learned that Walmart links come in a different format than expected, and have updated our logic to account for this when passing through utm parameters to Trackonomics
All tracking related to Cxense was removed from the Kinja codebase
Subtag data will only be valid as of this date for any posts on TheInventory.com
Impressions and clicks now go back to 2017, and splice counts and clicks go back to 2012
If a user clicks on an amazon link on the homepage stream excerpt, we are now sending source = “streamshare”
So that we can better track conversion for insets vs text links
We should now be able to provide more granular reporting for Trackonomics conversion