Update to newest YCSB version

[koppor]

YCSB has moved almost 700 commits since forking.

Integrate the new changes

[Vogel612]

Those 700 commits correspond to 9 releases and a bit.

Especially interesting for us are integrations of cassandra and redis (see #1), maybe dynamo. In general high-availability databases have been added or updated (elasticsearch, hbase).

AFAICT it should be possible to reintegrate the YCSB-TS fork into YCSB. That would eliminate the need to maintain this fork independently. It would also allow comparing Timeseries databases against general-purpose key-value stores without further effort.

Making the integration more difficult is the fact that apparently there have been changes to the indentation conventions of upstream, which results in huge whitespace conflicts that may or may not carry additional changes which are hard to spot.

I propose following route to integrating the changes:

• Integrate the changes of checkstlye.xml and non-java binaries from upstream
• Eliminate checkstyle warnings in both YCSB-TS and YCSB
• Integrate changes from upstream module-wise
• Finally reintegrate YCSB-TS into YCSB (?)

[baderas]

The data model/data structure is completely different between YCSB and YCSB-TS. Many TS features are based on that change in the underlying data/query structure.
I think a first step would be to check if they would accept some kind of "second model" for TS. The result would be that each client has to be changed/adopted to fit to both models and answer queries for both in a different way.
The Cassandra, Redis, Elasticsearch, Hbase integrations do not implement any TS data, only the default YCSB data model and its queries.
Maybe the TS model can somehow be fit on the original model? But i think its hardly possible.

[Vogel612]

@baderas YCSB is currently moving to support Timeseries Databases, see also brianfrankcooper#1008 Someone put a lot of thought into shoehorning the different requirements of a TS model into the default YCSB model. After a few hours of work, it seems very doable to adapt our Clients to the model (with some modifications of course). I'm currently doing a PoC with influxdb.

[baderas]

ah nice, I did not know that. Since the model is quite similiar, it seems to be possible to adopt our clients. It should also be checked if any ideas or code from us is used.