blockworks-foundation
/
solana-with-rpc-optimizations
mirror of https://github.com/blockworks-foundation/solana-with-rpc-optimizations.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

when writing to disk bucket index, tune towards packing tighter (#30761) 

* when writing to disk bucket index, tune towards packing tighter

* switch to min
This commit is contained in:
Jeff Washington (jwash) 2023-03-17 14:34:56 -05:00 committed by GitHub
parent 1a71cd7fe2
commit 6dd5a22926
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 9 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
bucket_map/src/bucket.rs
View File

@ -308,7 +308,15 @@ impl<'b, T: Clone + Copy + 'static> Bucket<T> {
let cap_power = best_bucket.capacity_pow2;
let cap = best_bucket.capacity();
let pos = thread_rng().gen_range(0, cap);
for i in pos..pos + self.index.max_search() {
// max search is increased here by a lot for this search. The idea is that we just have to find an empty bucket somewhere.
// We don't mind waiting on a new write (by searching longer). Writing is done in the background only.
// Wasting space by doubling the bucket size is worse behavior. We expect more
// updates and fewer inserts, so we optimize for more compact data.
// We can accomplish this by increasing how many locations we're willing to search for an empty data cell.
// For the index bucket, it is more like a hash table and we have to exhaustively search 'max_search' to prove an item does not exist.
// And we do have to support the 'does not exist' case with good performance. So, it makes sense to grow the index bucket when it is too large.
// For data buckets, the offset is stored in the index, so it is directly looked up. So, the only search is on INSERT or update to a new sized value.
for i in pos..pos + (self.index.max_search() * 10).min(cap) {
let ix = i % cap;
if best_bucket.is_free(ix) {
let elem_loc = elem.data_loc(current_bucket);