I think the issue would be that the compressor works on whole memory pages (I assume), so you'd need to know that the whole page could be freed. There might be kernel VM functions to handle that functionality manually, like if you were building a zone allocator or something?
I think it’s kind of related to how the oomkiller really needs to be integrated with the applications - the application needs to be able to identify memory as “critical, nice to have, ehhhhh, and cache” and then let the kernel throw them away as necessary. Once you start having to throw away critical pages you oomkill the process.
Doesn't good old uncompressed swap partition also have the same problem?
Page is swapped out but it has to be paged in before the process owning it can evaluate what's where and release the junk?
