PostgreSQL at low Level: stay curious! | Dmitrii Dolgov

Explore advanced performance analysis techniques for PostgreSQL, including BPF, Linux kernel tools, and custom solutions, to optimize performance and stay ahead of challenges posed by Kubernetes, Docker, and hypervisors.

Key takeaways

Use huge pages to reduce CPU overhead and improve performance
BPF can be used to profile and measure performance of a system and applications
Use Ple to measure latency and predict when issues will occur
Developing a custom solution to profile and analyze performance can be more effective than relying on existing tools
Linux kernel has various features and tools such as perf, bcc, and PGB that can be used for performance analysis
Kubernetes and Docker can pose challenges when profiling and analyzing performance due to their complexity and scalability
Hypervisors such as KVM and Xen can introduce additional complexity and overhead when profiling and analyzing performance
Subsystem profiling can be used to analyze performance at the subsystem level, such as the Postgres query analyzer
Stateless profiling can be used to analyze performance without maintaining state between measurements
Cache misses can be significant and can impact performance
Understanding and analyzing system calls and tracing can be important for performance analysis
M5 and M4 generations of instances on AWS have different hypervisors which can impact performance
Postgres has various features and tools such as Pgstat and Peak that can be used for performance analysis
BPF can be used to analyze and measure performance at the user-space and kernel-space level
Secomp can be used to analyze and measure performance of system calls
Kubernetes and Docker can be challenging to profile and analyze due to their complexity and scalability
Patrone is a successful solution for profiling and analyzing performance in a Kubernetes environment
Custom solutions can be developed to profile and analyze performance in a specific environment or application
Perf can be used to profile and analyze performance at the system and application level
Bcc can be used to analyze and measure performance of the Linux kernel and applications
GDB can be used to analyze and debug performance issues in applications

PostgreSQL at low Level: stay curious! | Dmitrii Dolgov

More talks