That's exactly what I was looking for. Thank you. If allowed, I would greatly appreciate a review of the attached code for a worker. I have 3 goals using zookeeper. 1, make it easy to know what workers are active. 2, have at most 1 worker processing a given filename at any time. 3, a worker will never process a filename which was previously recorded as completed in zookeeper. Thanks again
On Sat, Feb 6, 2016 at 12:07 PM, Patrick Hunt <[email protected]> wrote: > Perhaps section 2.3 of the formal paper? > > http://static.cs.brown.edu/courses/cs227/archives/2012/papers/replication/hunt.pdf > > Patrick > > On Sat, Feb 6, 2016 at 8:24 AM, Matthew Harrigan > <[email protected]> wrote: > > Hello, > > > > Is there a comprehensive and formal set of all guarantees zookeeper > > offers? I know of this > > < > http://zookeeper.apache.org/doc/trunk/zookeeperProgrammers.html#ch_zkGuarantees > >documentation, > > but it strikes me as more buzzwordy than formal and comprehensive. If > > other documentation exists, please let me know and disregard the rest of > > this email. > > > > I recently asked about one guarantee which I do not think is deducible > from > > the above linked documentation, namely that no two clients can > successfully > > create the same znode. A similar guarantee which I suspect is honored is > > that if a client successfully creates a regular znode and no one ever > > deletes the node, all future calls by any client checking the existent of > > that node will not return false, even if they happen immediately after > the > > acknowledgement to the node creator. I suspect that more possible > > guarantees could be formulated. I humbly suggest creating additional > > documentation with a comprehensive and formal set of guarantees. > > > > Thank you >
