Between approximately 2016-02-16T17:00:00 UTC and 2016-02-19T17:00:00 UTC our Hadoop backend experienced a single data infrastructure failure, which caused the large data processing pipeline to become backed up. Since the problem has already been resolved late last week, the backed up data issues should be already resolved or will soon be resolved. No data was lost and primary effect was slowdown of data jobs to process and refresh data internally.
We can confirm this was causing the following adverse problems:
- TA List (audience) uploads were delayed and stuck in processing state.
- Conversion data may have appeared to have holes or gaps in data for times of > 24 hours ago (data considered to be realtime was not affected). These holes would appear in both data retrieved by Ads API and by viewing conversion graphs on ads.twitter.com.
We recommend to re-fetch conversion data from the past week and it is also a standard recommendation to re-fetch conversion data periodically for the duration of the app event tag window length (e.g. 14 days).
If you saw any issues related to the above two points in the past week, it was likely to have been related to the issue explained here. Our engineering team did all they can to speed up and optimize the data jobs to resolve these issues as quickly as possible.
We sincerely apologize for these issues and thank you for your understanding.