ZooKeeper does not offer atomic write semantic for transaction log file write - in other words, it is possible that partial writes happen to transaction log file, in cases like disk error or ZK server process faults (i.e. process get killed).
On Sat, Dec 10, 2016 at 1:18 PM, Jordan Zimmerman < [email protected]> wrote: > Writes in ZooKeeper are all or nothing. A write to a ZNode is atomic and > complete. You cannot write partially to a ZNode. Does this help? > > -Jordan > > > On Dec 9, 2016, at 7:16 AM, Peng Qu <[email protected]> wrote: > > > > I am trying to figure out how zookeeper implements transaction. I am > > reading source code in directory org/apache/zookeeper/server/ > persistence/ > > but I could not find how zookeeper manager no-integration WAL files. So > > could you offer a simple introduction for that to me? > > Thanks very much. > > -- Cheers Michael.
