Sometimes a "mv" of a large file (> 1GB) in the same directory takes a long time on linux. Sometimes it doesn't. Why is that?
Ok, more info: gist.github.com/jamesls/9d52c7… Current theory is something is forcing rename() to wait on IO but I don't know why.
Screw it, we're going into the kernel...you don't want me in the kernel. I have no idea what I'm doing there. But I really want to know why this rename() call is so slow, it's really bugging me.
Updates! Digging into this more, I have confirmed that a rename() is in fact triggering IO via ftrace: gist.github.com/jamesls/345705…
And now the conclusion! This behavior is happening due to auto_da_alloc (kernel.org/doc/Documentat…). Mounting with noauto_da_alloc "fixes" this behavior: gist.github.com/jamesls/1df5d5…. Thanks to @_msw_ for pointing me in the right direction. That was fun.
