Write enough to send at least one BRW RPC to an OST, otherwise the client
might cache the whole write and the VM not flush it before we re-enable
BRWs. When that happens the client isn't evicted and our file compares OK.
b=5595
# OBD_FAIL_PTLRPC_BULK_GET_NET 0x0503 | OBD_FAIL_ONCE
# client will get evicted here
sysctl -w lustre.fail_loc=0x80000503
- do_facet client cp /etc/termcap $DIR/$tfile
+ # need to write enough to ensure we send an RPC
+ do_facet client dd if=/dev/zero of=$DIR/$tfile bs=1024k count=2
sleep $TIMEOUT
sysctl -w lustre.fail_loc=0