MIRA performs amazing when the node is live, but it less than optimal when reindexing. We are working on a hybrid approach that will allow reindexing fully in memory for costly indices which will then be transferred to RocksDB after the reindex is finished. This will allow node operators the freedom of faster replay at the cost of more expensive hardware, or reindexing on cheaper hardware over a longer period of time.
You are viewing a single comment's thread from:
How about creating snapshots of indexed checkpoints? Load them directly into RocksDB instead of having to reindex them every time.
We would like to do that as well, but sometimes snapshotting is not enough. We have had major security fixes in the past that could only be fixed via reindexing. There are different reasons to reindex. Getting a node up for the first time is the most common and makes sense to load from a snapshot. But in the case of a security vulnerability or a chain halt that requires a reindex to fix, we may not be able to wait a longer time to reindex. We realize that there is not a "one size fits all" solution, which is why we are looking at the hybrid approach and are interested in snapshotting.
we would have to trust snapshot party.. and thats not decentralized