As the data demand for deep learning models increases, active learning becomes essential to strategically select samples for labeling, which maximizes data efficiency and reduces training costs.Recent work addresses important real-world considerations of active learning, such as handling out-of-distribution (OOD) data and online discovery of novel object categories. However, a combined analysis of these scenarios remains unexplored.To address this gap regarding real-world considerations, we propose a novel scenario, Open-Set Discovery Active Learning (OSDAL), which integrates OOD sample handling and novel category discovery.In contrast to previous methods, we construct a common feature space within a single model that aligns known and novel categories while separating OOD samples.This enables our approach, Joint Out-of-distribution filtering and data Discovery Active learning (Joda), to uniquely address both challenges simultaneously by filtering out OOD data before selecting candidates for labeling.Unlike previous work, Joda does not require auxiliary detection models for filtering or selection and is, therefore, effectively reducing the computational overhead.In extensive experiments on 15 configurations and 3 metrics, Joda achieves consistently the highest or equally high accuracy as state-of-the-art competitor approaches in 39 out of 45 cases.