Serco Business Greetings,
The team I work with is encountering an issue when starting an Accumulo 1.7.x cluster and when running troubleshooting commands such as bin/accumulo admin checkTablets. The primary symptom is a NumberFormatException thrown within ZookeeperLockChecker that occurs when parsing the tablet server session id (Long.parseLong) for an input string "ff804d767efe0004" (which is out of range when interpreting as a positive signed long). >From what I can gather, our zookeeper cluster has been running for such a long >time that the epoch component of the session id has grown to the point where >interpreting the session id as a signed long would be a negative value. Within >the ZooKeeper code, the session id is treated as an unsigned long (e.g., >Long.toHexString) which leads me to think that the Accumulo code is not >parsing the value correctly. This discrepancy is present in all versions since >the introduction of the ZookeeperLockChecker class. There does not appear to be an easy way to work around this problem. Currently, our best idea of how to recover the data from this cluster is to set up a separate zookeeper cluster, migrate the data we have in zookeeper to the new cluster, and then swap over configuration to point to the new zookeeper cluster. I would appreciate any ideas or suggestions from the community. Thanks, Jonathan
