As a promising paradigm for fifth generation (5G) wireless communication systems, cloud radio access networks (C-RANs) have been shown to reduce both capital and operating expenditures, as well as to provide high spectral efficiency (SE) and energy efficiency (EE). The fronthaul in such networks, defined as the transmission link between a baseband unit (BBU) and a remote radio head (RRH), requires high capacity, but is often constrained. This article comprehensively surveys recent advances in fronthaul-constrained C-RANs, including system architectures and key techniques. In particular, key techniques for alleviating the impact of constrained fronthaul on SE/EE and quality of service for users, including compression and quantization, large-scale coordinated processing and clustering, and resource allocation optimization, are discussed. Open issues in terms of software-defined networking, network function virtualization, and partial centralization are also identified.