Скачать с ютуб Pinpointing memory leaks in production Haskell deployments by Adithya Kumar & Harendra в хорошем качестве

Pinpointing memory leaks in production Haskell deployments by Adithya Kumar & Harendra

Memory leaks are often difficult to debug, especially in production deployments. In this talk, we describe a GHC RTS patch that introduces a lightweight profiling mode specifically designed to identify memory leaks and memory fragmentation. This mode makes it easier to debug observed, steady increases in memory consumption and can be enabled in production with minimal performance overhead. The new profiling mode provides precise, top-down accounting of all memory used by the Haskell application at a given point in time. It leverages GHC's existing retainer profiling and era profiling infrastructure while keeping the profiling instrumentation to a minimum. This mode helps identify heap objects that are being held for extended durations, and it prints these objects along with the chain of references leading back to the root of the heap object tree. Additionally, the objects can be mapped to source locations, helping to pinpoint potential sources of memory leaks. Memory fragmentation can contribute to gradual increases in memory usage in production environments. By traversing the object tree, the profiler can report exactly where each byte of memory is being used, allowing it to detect fragmentation due to dead pinned objects that cannot be compacted. More details: https://confengine.com/conferences/fu... Conference Link: https://functionalconf.com