The Case for DBMS Live Patching [Extended Version]

• DOI: 10.48550/arxiv.2410.09925
• Publication Date: 2024-10-13
• Authors: Michael Fruth; Stefanie Scherzinger

Traditionally, when the code of a database management system (DBMS) needs to be updated, is restarted and clients suffer downtime, or provider instantiates hot-standby instances rolls over workload. We investigate third option, live patching DBMS binary. For certain changes, allows modify application in memory, without restart. The memory state all client connections can maintained. Although has been explored operating systems research community, it remains blind spot research. In this Experiment, Analysis & Benchmark article, we systematically explore field from perspective. discuss what distinguishes generic multi-threaded applications comes patching. then propose domain-specific strategies for injecting quiescence points into source code, so that threads safely migrate patched process version. experimentally interplay between query workload different methods, monitoring both transaction throughput tail latencies. show viable option updating systems, since providers make informed decisions w.r.t. latency overhead on side.