YBK-183: Garbage collector deletes active ledgers present in zookeeper

[jai1]

When all level 3 nodes are deleted from a given level 2 node, the hasNext returns null which compels the bookkeeper to incorrectly GC the ledgers

Based on the position of the level 2 node either of the two can happen:-
a. If the level2 node is the first level2 node scanned (smallest number) then the bookkeeper GCs all nodes herewith - Fixed in this PR

b. if level2 node is in the middle (not the first) the bookkeeper throws an exception - and GC activity stops altogether and those ledgers are never deleted - Needs to be analyzed and fixed

I am able to reproduce these scenarios manually - will see if I can add unit tests

@revans2

[merlimat]

@jai1 Change LGTM. Would be possible to test this by manually removing the intermediate z-nodes?

Also, please submit this on apache/bookkeeper as well so that it gets included into master.

[merlimat]

@jai1 @rdhabalia Is this getting targeted for 1.22 release?

[revans2]

@jai1 This change does look good. Are you going to address b? Also this code is out in the Apache repo as well and needs to be addressed there too. Do you have plans to put up a pull request there as well?

[jai1]

@revans2 - I don't have permissions to merge PRs on bookkeeper

I have raised a PR for fixing issue 2 (#23) but didn't get time to test it

Also this code is out in the Apache repo

Can you point me to the exact branch cause - I checked master and the code there doesn't use preload.