RocksDB

[shargon]

Close #966

Default values taken from: https://github.com/facebook/rocksdb/blob/189f0c27aaecdf17ae7fc1f826a423a28b77984f/utilities/leveldb_options/leveldb_options.cc

Descriptions taken from: https://github.com/influxdata/rocksdb/blob/7adff3eb33001c471c48cdcadf1b55462920b123/options.go

TODO:

• Unit tests
• Use families instead of prefixes
• Update neo-cli (Allow neo-cli to switch between RocksDB and LevelDB neo-node#451)

[shargon]

@eryeer Could you move your LevelDB unit tests to this PR? (modified for RocksDB)

[vncoelho]

@shargon, I think that the discussion #966 was not yet decided, please check my comments there.

In the last comment of @erikzhang he mentioned that it was not yet decided when you asked.
I also had a conversation with @jsolman and he also did not say that RockDB would be better.

In this sense, I think that this PR could wait a little more.
It will require a considerable effort. This is not a priority for NEO 3 I think.

The DB could be even more flexible, this would really be a good change.
By having the DB more flexible it could even by defined by choice of the node runner.

I think that a change like this would be more plausible with a dedicated benchmark when we have defined a strategy for testing the Blockchain and throughput.

I am not in favor of this change right now.

However, if you really think it is better in terms of performance and the needs, I think that you should push this forward, because you are more expert in this topic than me.
On the other hand, I think that we currently are not really able to test all cases statistically. But maybe it is clear to you that this is better.

[vncoelho]

@shargon, with the possibility of switching between them (and perhaps, for now, leaving LevelDB as default) I am 100% in favor of this.

In addition, it is was surely not an easy task to you and since you already dedicated time to learn and push this, I believe that it should be merged.

[shargon]

@eryeer wait until merge this then, we can do the test for the three storages, memory #1064, RocksDB and LevelDB together

[eryeer]

@shargon Ok, after it merged, let's test them together.

[vncoelho]

Beautiful job @shargon!
Congratulations. Flexible and powerful.

This surely opens a wide path of possibilities.

[erikzhang]

Maybe we can move this to a plugin. Because I anticipate that we may have many different storages in the future, such as LevelDb, RocksDb, FASTER, and memory. If each one is integrated in the core, it is not necessary.

[vncoelho]

I agree, Erik.

We could merge @shargon refactor at least, yes?

[vncoelho]

The plugin sounds a good idea that @erikzhang suggested and even more flexible, promoting a compact code.

Maybe we can keep the refactoring of level leveldb.

@shargon, it is a great job and this was the most import step. Congratulations once more.
Thinking on Rocks, Level, Faster and Memory makes us more prone to consider as a plugin.

dismissing because of possibility for changing to plugins. However, PR had my previous approval.

[shargon]

After #1087 i will reduce this PR for preserve only the refactored part, and move th rest to neo-plugins

[shargon]

I will close this when IStorage is merged, in order to remember the improves. In favor of neo-project/neo-modules#144

[erikzhang]

Please review #1249 first.

[shargon]

LevelDb changes moved to #1308
RocksDb changes moved to neo-project/neo-modules#144