Istvan Toth
Istvan Toth
Note that new patch removes the SchemaUtil.processSplit() logic, which attempts to expend the split point to cover the whole PK. My reasons removing it: * It uses arbitrary lengths when...
Some more reasoning to remove SchemaUtil.processSplit() : AFAICT it tries to emulate how HBase would automatically split a region. However, I could not find any reason why we would want...
There two outstanding problems: * Upgrade tests fail, because the split points on the System tables use the old longer split points, and the update code errors out. * org.apache.phoenix.end2end.LocalIndexSplitMergeIT.testLocalIndexScanWithMergeSpecialCase()...
Both could probably be fixed by re-adding the split point extesion code, but I'd like to understand it more first.
org.apache.phoenix.end2end.LocalIndexSplitMergeIT.testLocalIndexScanWithMergeSpecialCase() doesn't even use salted tables. I must have broken something else in the patch.
Turns out that split processing is not necessary for salted tables, but rather for all tables that have local indexes, which LocalIndexSplitMergeIT.testLocalIndexScanWithMergeSpecialCase() has luckily caught. This of course means that...
Could you please review the new version @dbwong ?
However, I need to update the ticket and patch description, as the patch does kind of the opposite of what it says.
Squashed and renamed the patch.
Yes, I tested 4 different credentials, but none of them worked :(