wasphin commented on code in PR #3668: URL: https://github.com/apache/brpc/pull/3668#discussion_r4390052951
########## src/bthread/stack.cpp: ########## @@ -37,6 +38,14 @@ DEFINE_int32(stack_size_large, 8388608, "size of large stacks"); DEFINE_int32(guard_page_size, 4096, "size of guard page, allocate stacks by malloc if it's 0(not recommended)"); DEFINE_int32(tc_stack_small, 32, "maximum small stacks cached by each thread"); DEFINE_int32(tc_stack_normal, 8, "maximum normal stacks cached by each thread"); +DEFINE_int32(bthread_stack_trim_idle_s, 600, Review Comment: Could we default `bthread_stack_trim_idle_s` to `0` and make trimming opt-in? For production deployments provisioned for peak concurrency, lowering idle RSS does not reduce the memory capacity that still needs to be reserved for the next peak. Reusing trimmed stacks also requires faulting in and zeroing pages again; retaining the virtual mappings does not guarantee that physical memory will be available when demand returns. Reclaiming idle pages can be useful on shared hosts, but the benefit seems workload-dependent and may be limited for dedicated instances. Defaulting to `0` would preserve the existing stack reuse behavior and let users explicitly choose this tradeoff. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
