Gitaly: Get perf data on clone performance (shallow vs partial clones)
On this issue, we'd like to get a baseline for what we currently have (re our repo cloning performance). Once we have the benchmark, we could look into what would give us a performance benefit. Example: What is the performance difference between partial and shallow clone? We are using shallow clone now for most operations.
From @nnelson : We have flame graphs for some nodes now. We should work with Observability and Igor to get those for at least one experimental gitaly node to establish firm metrics on performance for such operations.
Alb: We could also leverage Jaeger: https://jaeger.gstg.gitlab.net/
Nels: We can easily establish an elapsed time duration for such clone operations on large projects like gitlab-org/gitlab
and gitlab-com/www-gitlab-com
by creating a single VM instance in GCP in the same network as a gitaly shard with a large repository for a project like gitlab-org/gitlab
and executing the different clone operations there.
Analyzing round-trip durations for off-network cloning is worthless, at least in this context.