Keith Turner
Keith Turner
In #593 async commit was added. The actual asynchronous writes to Accumulo were done in a hacky way that works, but is sub-optimal. The reason the current method is sub-optimal...
This iterator needs a comprehensive unit test. The unit test does not have to use minifluo, it could just use the Accumulo SortedMapIterator.
This iterator needs a comprehensive unit test. The unit test does not have to use minifluo, it could just use the Accumulo SortedMapIterator.
When a transaction rolls forward or rolls back another transaction, it uses its start timestamp. If a transaction fails, then multiple other transactions could roll its columns forward with different...
In the work for #593 the commit steps that involved scanning were not converted to an async model. This caused #672 which created a seperate thread pool for these commit...
In #655 a change was made to drop rolled back data during GC. This change did not drop delete locks for the primary row/column of a transactions. I suspect it...
During a long run of webindex on EC2. The Oracle died and was not restarted. When I started looking into this I also noticed some workers were not being restarted....
Saw higher lock after changes for #593. Thinking for the case of observers triggered by weak notification that hardly collide, can obtain a start timestamp a bit before the observer...
With the change in #645 its possible that an observer could be run for a transactions that rolled back. This would happen in the case where a transaction writers notifications...
I noticed that `SpanUtil.toRowColumn(Key)` ignores the qualifier when the family is empty. Maybe this intended for the use within Span, not sure. However for other places that use this code,...