To enable MongoDB perf CI testing to run with unified infrastructure, STM would like to experiment with migrating sys-perf and perf tests on Evergreen owned machines.
Currently, we use terraform to create machines in the 10gen-cap account that run performance workloads. Most of the EC2 options that terraform uses are already available through host.create with the following exceptions:
- creating hosts in the same (aka "Cluster") placement group: to reduce network latency, the hosts for sys-perf are placed in the same placement group unless opted out. Other placement group options supported by EC2 are not needed at this time.
- create up to around 50 hosts in the same task or by the same user: the sharded cluster perf tests use a 3 shard + 3 mongos cluster + workload client totaling 3 * 3 + 3 + 3 + 1 = 16 nodes. Future topologies (e.g. elastic) may wish to run with even more hosts of smaller sizes.
This ticket is for requesting the above 2 features.