Memory allocation issue while reading many large files

[enp]

Hi,

Reading https://github.com/yandex-cloud/geesefs?tab=readme-ov-file#memory-limit about ENOMEM error for cases where read-ahead-large * processes count > memory-limit

Looks as not good limitation for production cases where we can't limit processes count

Maybe it is possible to check if read-ahead-large can be really allocated and skip allocation to allow at least slow reading without read ahead?

[vitalif]

Hi, I'm not sure it's easy to implement it this way, but we can think about other ways to solve this problem
For example we can dynamically increase memory limit when there are too many readers, or maybe just allow readers to wait for buffers to be unlocked
I.e. now readers may get ENOMEM when, at the moment of reading, memory cache is full and all data in memory cache is marked as required for some read or write requests. It's probably possible to just make it to wait for this data to be unmarked...

[enp]

Any solution above looks better than ENOMEM really :)