During a maintenance update to the regional Upstash Redis databases in AWS eu-west-1, several databases hosted in that region has unnecessarily triggered a full synchronisation between their primary and backup replicas.
A full synchronisation is the invalidation of the whole data in the target replica and starting a fresh re-population from the source replica. Under normal circumstances, full synchronisation is required only in cases where the data integrity is lost in one of the replicas, which was not the case here.
This incident impacted the performance of regional databases on AWS eu-west-1 only. Full synchronisation has caused a very high CPU load and caused a performance degradation on some of the databases that has a replica in this region. Moreover, our system throttles the databases that are going through this operation to allocate more CPU to the synchronisation to finish it sooner.
No data or consistency has been lost.
As a quick remediation, we have unthrottled affected databases on 15:06UTC and enabled more throughput, however high latency has still been observed until the full synchronisation is completed on 21:23UTC.
A fix has been prepared to avoid this unnecessary full synchronisation on regional databases, and will be deployed shortly.
This issue is not present on Upstash Global databases, which is our new generation infrastructure and is now our default offering. We will reach out to our regional users on how to migrate to Upstash Global going forward.