Re: GIT get corrupted on lustre

Thomas Rast wrote:

Taken together this should mean that the bug is in fact simply that the
calls do not *restart*.  They are (like you say) allowed to return EINTR
despite not being specified to, *but* SA_RESTART should restart it.

Now, does that make it a lustre bug or a glibc bug? :-)

The kernel takes care of SA_RESTART, if I remember correctly.  (See
arch/x86/kernel/signal.c::handle_signal() case -ERESTARTSYS.)