Hi Ohad, this is a known issue, documented here http://hadoop.apache.org/zookeeper/docs/current/zookeeperProgrammers.html#Sequence+Nodes+--+Unique+Naming the plan is to mitigate when we move to version 4.
Note that the seq number is "per znode", meaning that the value is not globally maintained. If this were ever to become an issue one could delete the "parent" znode and recreate it, thereby resetting the req number counter. (btw, mahadev did the calculation a while back and it was something like 4 years to overflow if seq znodes were created on a single parent znode at near our max write rate, we hope to put out v4 before this. ;-) )
Patrick Ohad Ben Porat wrote:
Just to make sure, if I use the sequential flag and at some point will reach its limit of 9999999999... ZK will assign the next new node with 0000000000 correct? (assuming the first 0000000000 node already been deleted from tree)