when there is a block which is using on Node1 and at the same time, there is a request on Node2 for the same block.
DLM Master catch the status and let DBWR(DB Writer) write the block to Database file and this block is read by node2.
At last, if check point occurs on node1 , but the last version of block b is on node2 , and the old image of block b is on node1, then node1 notify the GCS and GCS send write-disk command to node2 . Node2 write block b on its local buffer to disk and then notify GCS . Finally , the GCS notify the other nodes to delete the past image (PI) .
来自 “ ITPUB博客 ” ，链接：http://blog.itpub.net/35489/viewspace-84900/，如需转载，请注明出处，否则将追究法律责任。