0. change(db): Use Ribbon filters for database index lookups

[teor2345]

Motivation

1. We need Zebra to sync faster for the cached state lightwalletd tests.
2. We want to speed up Zebra's tests, and the slowest tests are sync tests.
3. This might help with locking bugs.

Using Ribbon filters for index lookups makes it about 7% faster. (Ribbon filters are like bloom filters.)

Specifications

RocksDB can do lookups using a Ribbon filter, which can improve read performance:
https://docs.rs/rocksdb/latest/rocksdb/struct.BlockBasedOptions.html#method.set_ribbon_filter

See also:
https://zhangyuchi.gitbooks.io/rocksdbbook/content/RocksDB-Tuning-Guide.html

Solution

• Use a Ribbon filter for database queries

This change doesn't require a database version increment, but it won't have the full performance impact until the database is rebuilt. (I tested locally, and the cached state test should also check this.)

Performance

I ran a full sync test to check performance:

• https://github.com/ZcashFoundation/zebra/actions/runs/2094282197

It finished in 5h43m, compared with most syncs that are around 97% - 99% at 6h.
That's around 7% faster.

(The test was after the lightwalletd database changes, but it should also be faster before them.)

Review

Anyone can review this PR.

I'd like to merge this PR before we merge any more database changes, to make merging them faster.

Reviewer Checklist

• Code makes sense
• Existing Tests Pass
• Sync is faster

Follow Up Tasks

Try a database format upgrade with these changes:

• like change(db): Use new bloom filter format for database index lookups #4041, but with Ribbon filters

[conradoplg]

Looks good! I'm wondering why 9.9 was chosen, but I'll go ahead with this to unblock the rest

[teor2345]

macOS timed out fetching the Zcash parameters

[teor2345]

@Mergifyio refresh

[mergify]

refresh

✅ Pull request refreshed