Use linxGnu/grocksdb instead of gorocksdb

[robert-zaremba]

Summary

We forked tecbot/gorocksdb (which doesn't have any activity any more) to add OptimisticTransactionDB::NewCheckpoint functionality.

There is https://github.com/linxGnu/grocksdb which is also a fork of tecbot/gorocksdb, is active and supports the latest Rocksdb (v7) and provides additional optimizations.

Objective

Update cosmos-sdk/db to support https://github.com/linxGnu/grocksdb or fork it into cosmos, add OptimisticTransactionDB::NewCheckpoint and use it instead of cosmos/gorocksdb.

Ref: osmosis-labs/osmosis#695

CC: @faddat , @roysc

---

For Admin Use

• Not duplicate issue
• Appropriate labels applied
• Appropriate contributors tagged
• Contributor assigned/self-assigned

[faddat]

Happy to work on this.

I should mention:

From every metric I can find (iavl benchmarks and state syncing osmo/gaia (tho both seem broke, both apply snapshots fastest with pebble)) PebbleDB is in fact preferable to RocksDB.

Trouble is, I think there is a bug in my implementation.

[robert-zaremba]

pebbledb is better event than linuxgnu/grocksdb?

[tac0turtle]

can we not upstream to the cosmos fork? im wary of switching forks every few months.

[alexanderbez]

can we not upstream to the cosmos fork? im wary of switching forks every few months.

This.

[tac0turtle]

Lets upstream and create a new release. Happy to coordinate this

[faddat]

It is not upstreamable

They are separate git histories.

I mean we could just replace our whole repo.... but that doesn't seem like a great idea.

linxGnu/grocksdb is very well-maintained.

So maybe instead of creating our own fork of it, we should consider a donation to its author?

[tac0turtle]

lets see what @roysc and @i-norden say. they are our db experts

[tac0turtle]

this is part of cosmos-db