Traditional methods for data analytic include two disparate database systems for different goals. One database system processes transactional workloads, known as Online Transactional Processing (OLTP), characterized by low latency but highly concurrent operations. The other one is a data warehouse that processes queries for data analysis, known as Online Analytical Processing (OLAP), characterized by long running operations such as joins and aggregates, involving computation on multiple tables and causing bulk reads and writes.Greenplum database is traditionally known as an OLAP data warehouse system without sufficient OLTP ability. In this paper, we aim to address the ever-growing demand for real-time data analysis and to enhance Greenplum with the hybrid capability to serve both OLTP and OLAP workloads, while maintaining the ACID properties and delivering satisfying performance. To build an efficient parallel HTAP system based on an OLAP system, we first identified engineering and performance challenges, such as the under-performing restrictive locking and the two-phase commit protocol. We proposed and implemented solutions to address these challenges in Greenplum, including a global deadlock detector which is used to increase the concurrency of query processing, and a one-phase commit protocol which is used to reduce table locking to speed up query processing. A further improvement which is named resource group is deployed to separate OLAP and OLTP workloads into more suitable query processing mode. Those ideas boost the OLTP performance without sacrificing the OLAP performance and ACID. Performance evaluation on the TPC-B and CH-benCHmark benchmarks shows a significant improvement.