A few years ago, the company I work at switched to using Alpine-based docker images for most containerized things. One side effect was that our Ansible playbook

A few years ago, the company I work at switched to using Alpine-based docker images for most containerized things. One side effect was that our Ansible playbooks (running from inside one of these containers) would fail with inconsistent network timeouts when targeting a couple thousand servers. It turned out that the issue wasn't with the network nor with Ansible. The way that Ansible invoked some library functions for keeping track of SSH connections caused it to create a bunch of POSIX semaphores via sem_open(). glibc had a dynamically allocated data structure to keep track of semaphores, but musl libc only had a fixed-size 256 element array. When the semaphore limit was exhausted, Ansible would fail to keep track of the connections, resulting in a network timeout error message. I fixed the problem by forking musl's semaphore functions, making the array resizable, and loading the implementation with LD_PRELOAD: https://github.com/chenxiaolong/musl-sem-ext https://github.com/chenxiaolong/musl-sem-ext. Worked perfectly for 6 years until we decommed our data center :)