I've been able to manually put the store into a consistent state. I did this by simply skipping the references to Relationship[2055394] that had somehow gotten into the relationship chain of Node[1]. I haven't, however, been able to reproduce or find the cause of this.
It would have been interesting to look at the logical log Neo4j created while working on your store. Could you in the future configure Neo4j with: keep_logical_logs=true That way all the logical logs will be kept and I could use that to analyze the problem, should this happen again, and you could use it to recreate your store from scratch to a consistent state. I'll continue my investigation a bit more, but I'm not sure I can get anywhere further since the faulty record has been removed. For my continued investigation I have a few questions: Does your application ever remove any relationships during normal operation? Or was the removal of the invalid relationship record the only that ever occurred? It looks like at least one other, Relationship[2055617], has been removed at some point. If this ever happens again, send the faulty store to me immediately, don't try to fix it on your own. I'd love to see the actual corrupted data. As it looks at the moment it is tempting to blame vmware (not going to do that without evidence though), I don't know how well they propagate filesystem flush() operations to the host file system, I'll have to investigate that. Was the file system in the vmware vm an actual partition on disk, or was it a virtual file system stored in an image file on the host? The non-clean shutdown that caused this, was that a JVM crash, a vmware vm crash, or a host machine crash? Cheers, Tobias On Tue, Feb 1, 2011 at 7:12 PM, Lars Hvile <[email protected]> wrote: > > -- directly after "successful" recovery > org.neo4j.kernel.impl.nioneo.store.InvalidRecordException: Node[1] is > neither firstNode[2055881] nor secondNode[2055876] for Relationship[2055394] > > -- after manual deletion of Relationship[2055394], in a desperate > recovery-attempt =) > Unfortunately I don't have a copy of the database at it's original corrupt > state, I do however have a copy of it after I deleted the the relationship.. > 788MiB zipped > > It really is a shame that you removed that relationship. It would have been super useful to look at what that record looked like. Although I think that record was fine, more likely it was the reference to it that was faulty. -- Tobias Ivarsson <[email protected]> Hacker, Neo Technology www.neotechnology.com Cellphone: +46 706 534857 _______________________________________________ Neo4j mailing list [email protected] https://lists.neo4j.org/mailman/listinfo/user
