The technical product team of iQIYI
DPVS v1.9.0, the latest open‑source four‑layer network load balancer from iQIYI, now fully supports DPDK‑20.11 LTS, offers streamlined compilation, universal flow management, dynamic mbuf fields, enhanced per‑core scheduling and UDP health checks, and has been running stably in production for three months.
iQIYI’s Basic Data Platform unifies data exchange across dozens of business lines by providing massive storage, distribution, online query and offline analysis services, employing an access layer, unified management, fine‑grained governance, dual‑cluster ID generation, active‑standby HBase with MongoDB WAL, RocketMQ messaging with server‑side filtering, and horizontally scalable read replicas to ensure high availability and performance.
This article details iQIYI's unmanned monitoring platform, covering its design goals, overall architecture, core modules such as real‑time data collection, decision engine, and event‑processing engine, as well as the machine‑learning model used for production‑time prediction and the system's operational results and future roadmap.
The investigation of QLB‑4‑based TFServing calls revealed uneven traffic, stale routing after scaling, and idle servers due to layer‑4 hash routing, leading the team to replace QLB‑4 with a Consul‑driven client‑side load‑balancer that dynamically pools servers, eliminates restarts, and cuts GPU waste.
iQIYI’s iOS Cloud Recording and Playback Platform leverages a cloud‑device farm and a Swift‑based driver to capture user actions, generate Python scripts, and replay them across shared devices, dramatically cutting automation costs while providing fast DOM access, multi‑method element identification, CI integration, and visual reporting for reliable regression testing.
iQIYI built a Spring Cloud extension that adds zone‑ and region‑aware client‑side load balancing, metadata‑driven instance filtering, and delayed registration plus deregistration hooks, enabling proximity routing, graceful startup, and graceful shutdown for multi‑region, multi‑datacenter deployments and improving service stability and disaster‑recovery.
iQIYI’s generic ranking framework unifies feature production, replay, training, and ranking into modular, configurable phases that handle offline and real‑time data, support diverse models, provide automated monitoring, and have been deployed across all platforms, delivering over 20% higher watch time and doubling first‑play videos.
iQIYI’s QAV1 encoder, which combines the next‑generation AV1 codec with deep‑learning techniques, delivers 20‑42% bandwidth savings and up to 36% higher compression efficiency than x265 while maintaining ultrafast 60 fps encoding speeds, enabling high‑quality 4K/8K streaming and live broadcast across devices.
iQIYI integrated JaCoCo with its DevOps pipeline, using Git diff and Jira‑linked branch naming to automatically generate fine‑grained coverage reports at incremental, branch, and requirement levels, enforcing 90 % thresholds that lifted backend service coverage from under 70 % to over 90 % and markedly improved reliability.
iQIYI optimized a 4K video super-resolution model using TensorRT, employing split of graph, operator fusion, custom CUDA kernels, and int8 quantization, achieving tenfold speedup (≈180 ms per 1080p frame) and demonstrating deep customization potential for large‑scale production.
