MR discussion diffs require compression/decompression when accessing Redis
In gitlab-com/gl-infra/scalability#418 (comment 371299552), it was discovered that
Projects::MergeRequestsController#discussions was the top Redis
cache reader (taking around 45% in the context of the top 3 actions), reading an average of 20gb per day. This was reported in #225600 (closed).
The outcome of #225600 (closed) was to compress the data before it is written to Redis, and then decompress the data on read. While this mitigated the problem of how much data was being persisted in Redis, the longer-term problem of being more efficient with the data (and moving the need for compression) needs to be addressed.
Thelabel appears on this issue so that we can track the work, but the DRI remains with .