Has this actually happened on any instance? Asking since
posix_fallocate usually falls back to zero-writing on filesystems
where fallocate is not supported (which is why fallocate is less
used compared to posix_allocate in certain cases, since former
fails).
However, this may again fail but will be on platforms with very
old glibc (need to be really old AFAIK) where this fallback may
not have been implemented.
@Alex,
Has this actually happened on any instance? Asking since
posix_fallocate usually falls back to zero-writing on filesystems
where fallocate is not supported (which is why fallocate is less
used compared to posix_allocate in certain cases, since former
fails).
However, this may again fail but will be on platforms with very
old glibc (need to be really old AFAIK) where this fallback may
not have been implemented.
I tested with
wsrep_provider_ options = "gmcast. listen_ addr=tcp: //127.0. 0.1:4010; gcache. dir=/dev/ shm/gcache; gcache. page_size= 134217728; gcache. size=134217728; debug=1"
But was not able to get any crash here (it also changed the size
to a different value, not an exact multiple of 4096).