When a client attempts parallel uploads of new objects (e.g. hundreds of unique small files) the upload latency quickly climbs to several seconds.
The backends have a global RWMutex. Per-object locks could reduce contention.
environment: fake-gcs-server and single client deployed in local minikube-on-docker
When a client attempts parallel uploads of new objects (e.g. hundreds of unique small files) the upload latency quickly climbs to several seconds.
The backends have a global RWMutex. Per-object locks could reduce contention.
environment: fake-gcs-server and single client deployed in local minikube-on-docker