"The Coordinator registers each index spec by name that comes in, before the indexes are set up in the index catalog as in-progress. Then subsequent requests with the same index names hit an error in the Coordinator. Without the Coordinator, spec requests are normally automatically filtered out if they're already found to be built or building via the index catalog.

The concurrency test has 10 threads, potentially all running the same createIndexes requests at the same time, so "There's already an index with name..." errors make sense – previously they'd get filtered out and the redundant commands return OK.

Perhaps a new error code IndexBuildAlreadyInProgressForName to fix tests, and we can consider the final behavior changes later, and easily identify the changes by error code if we want to undo them. This is related to our problem regarding spec checking being scattered all over the place at different levels of the code."