Skip to main content
Warning This document has not been updated for a while now. It may be out of date.
Last updated: 20 Dec 2019

content-publisher: 15. Bulk data loading

Date: 2019-12-19


Content Publisher regularly has to present users with long lists of data that is loaded from the Publishing API, examples of these are organisations, contacts and the topic taxonomy. Content Publisher typically needs to download all the records of a particular datatype, which can involve multiple HTTP requests. When this is done during a web request the application is at an increased risk of returning a slow response or even timing out. The risks of this have been partially mitigated by caching.

These situations where Content Publisher needs all the data records for a particular dataset are colloquially referred to as bulk data. This is distinguished from other API look ups where we may need to access a single resource (such as looking up the topic tags) that have less risks of performance penalties.

This particular problem has been faced and solved for GOV.UK before. Whitehall takes an approach for the topic taxonomy where the data is loaded at runtime from Redis. A periodic background job then refreshes the data in Redis. This avoids the need to look up the data during a web request.


In Content Publisher we have chosen to implement a similar pattern to Whitehall. Data is read from Redis at runtime and a background process runs a periodic job to re-populate the Redis data.

This approach makes use of the Rails Redis cache store which is wrapped within an object, BulkData::Cache, for writing and accessing data. We have introduced a concept of repository classes that can be used to work with the loading and modelling of data that is stored in the cache.

On application boot and on a recurring 15 minute interval a background job will run to try to re-populate the data. In the event of an error occurring this job will log this issue and retry. If retries are exhausted a decision will be made whether to send the error to Sentry or not based on whether the error is likely to be a problem we will investigate.

The data within the bulk data cache is stored with an expiry of 24 hours, which is a lot longer than the interval we use to re-populate the cache. The reason for such a long time is to provide as a safety net for problems to occur and not to present any sign of issues to users until this period has expired. If a frontend request is made for a resource that requires bulk data, and the cache entry for that is empty, a 503 response is returned to the user and the job to re-populate the cache is enqueued.

If we hit a scenario where there are errors with the jobs there is a chance that we will slowly build up a large queue of this same job multiple times. To prevent this situation causing any significant problems (such as flooding the Publishing API with requests) the job will only populate data where the data is older than 5 minutes.




For Content Publisher to operate normally the bulk data cache will be expected to be populated. In the development environment this has the consequence that developers should always run the application with the background worker process. In a test environment the bulk data cache will need to be populated before and cleared after any test run that makes use of it, helpers have been provided for this task.

Data that is currently loaded at runtime (such as organisations, contacts and topics) should be migrated to use the bulk data approach to reduce the risk of timeouts and slow responses for users.

The switching of loading data from a runtime task to a background task allows Content Publisher to load richer data from the Publishing API. This is intended to be used to enable eventual features such as: acronyms of an organisation or time a minister was in appointment.

The use of a scheduled task increases the risk of sending Sentry unactionable error alerts. In the staging and integration environments there is a high chance of errors occurring during the data sync process. We intend to review any errors that come in as part of this and adjust reporting thresholds to try exclude them.