Right now, we do two passes to load pages: first an fadvise pass, then a pass where we actually fault the pages using mmap. If the OS is doing a non scan-resistant LRU, then we can get in a situation where we scan once, evict a few pages at the end of the scan, and then scan again. This would happen whenever a larger page cache is dumped from a larger machine and loaded on a smaller one.
Right now, we do two passes to load pages: first an fadvise pass, then a pass where we actually fault the pages using mmap. If the OS is doing a non scan-resistant LRU, then we can get in a situation where we scan once, evict a few pages at the end of the scan, and then scan again. This would happen whenever a larger page cache is dumped from a larger machine and loaded on a smaller one.