Masahiko Takahashi <masahiko at linux-foundation.org> writes: > Hi everyone, > > I would like to post a small patch that implements pipe checkpoint > and restart functionality. > > The patch supports pipe buffers used with splice/vmsplice systemcall. > Usually, spliced buffers' pages are in memory cache because their data > are read from an filesystem. However the patch doesn't treat buffers' > pages as cached memory when restoring. If trying to treat as cached > memory, each buffer's page should be tagged with its filename when > checkpointing. This is not a difficult implementation, but I'm still > wondering this is worth implementing... Rewind a little bit please. I can clearly see where checkpoint/restart fit into the containers conversation. However I don't see where this checkpoint/restart patchset fits into the checkpoint restart picture. What ideas are you trying to discuss? I will say that I am not yet convinced that we need the kernel saving and restoring the checkpoint. Regardless of the approach when we start seriously discussing checkpoint/restart one of the big issues is how do we maintain an ABI to user space for all of the data so we can migrate applications across kernel version. Eric