So we were trying to re-use a single flyteagent docker image that has our custom connectors on it across multiple deployments. A main reason was for resource isolation. Rather than trying to create multiple repositories, each producing their own docker image, we opted to reuse a single image and rely on routing in the config. Unfortunately, the way the connectors are registered by their endpoints, all connectors that are loaded are made available. So we either needed to explicitly map each task type to a connector(including default connector) or find a way to not load connectors on deployments. We opted to not register our connector through

[project.entry-points."flytekit.plugins"]

anymore and instead create module files that we can point to on connector startup like

pyflyte serve agent --modules com.main_module

I'm not sure if this is the best solution, but it works for now. Would love to know how others have solved this