The new error label indicate when something is safe to retry, no matter the nature of the command, even getMore.

Currently, the mongos ignore retryability, most notably in all write ops. The following function must be evaluated if changing kNoRetry to kStrictlyNotIdenpotent:

• executeChildBatches (bulk_write_exec.cpp)
• exhaustCursorForReplyItems (bulk_write_exec.cpp
• executeChildBatches (batch_write_exec.cpp)
• WriteBatchExecutor::_execute (write_batch_executor.cpp)

This should make writes able to retry on error with the label RetryableError.