As part of the infrastructure improvements that we started in June, we’re going to be merging some of the clusters of EC2 instances that host MemCachier caches, starting in the next couple of weeks. This will not affect all customer caches, and the impact during the merge process will be minimal, but we wanted to be up-front about the fact that this is happening.
The main purpose of this activity is to take advantage of some of the newer instance types offered by AWS, and in particular to improve the networking performance of the instances we use. Merging clusters also gives us the opportunity to launch new sets of instances for the new merged clusters, which means we can set these instances up in placement groups, which provide additional benefits in terms of network latency and throughput.
The only negative impact that customers may notice during the cluster merge process is that small fractions of their caches’ keyspace will be evicted as storage backends are migrated into the new merged cluster. We’ll follow the same kind of procedure for this that we’ve used for earlier migrations, where we move backends over the course of a couple of weeks. This has been successful in the past and hasn’t led to any noticeable performance degradation for users.
The positive impact of these changes is that all user caches in the merged clusters should exhibit lower latency once the merge is complete!
If you have any questions about this process, as usual please contact us at firstname.lastname@example.org.