Ceph » CephFS

LazyIO synchronize fails to do the task of making the propagated writes by other clients/fds visible to the current fd in the following situation:

1) write() by fda
2) lazyio_propagate() by fda
3) lazyio_synchronize() by fdb.
4) read() by fdb

In this scenario fdb is expected to read whatever is written by fda but it does not read anything (0 bytes returned). Upon debugging, it seems the read fails because the file size( in->size ) is still 0 during read() by fdb and hence returns without reading in this conditional in Client::_read_async():

if( off > in->size )
return 0;

Ideally the lazyio_synchronize() call before the read() should have taken care of this and hence should call a getattr(CEPH_STAT_CAP_SIZE) to update the in->size.