disk buckets: refactor (#21972)

bucket_map/src/bucket.rs

@@ -198,14 +198,10 @@ impl<T: Clone + Copy> Bucket<T> {
                 continue;
             }
             index.allocate(ii, elem_uid, is_resizing).unwrap();
-            let mut elem: &mut IndexEntry = index.get_mut(ii);
+            let elem: &mut IndexEntry = index.get_mut(ii);
-            elem.key = *key;
+            // These fields will be overwritten after allocation by callers.
-            // These will be overwritten after allocation by callers.
             // Since this part of the mmapped file could have previously been used by someone else, there can be garbage here.
-            elem.ref_count = 0;
+            elem.init(key);
-            elem.storage_offset = 0;
-            elem.storage_capacity_when_created_pow2 = 0;
-            elem.num_slots = 0;
             //debug!(                "INDEX ALLOC {:?} {} {} {}",                key, ii, index.capacity, elem_uid            );
             return Ok(ii);
         }
@@ -285,8 +281,8 @@ impl<T: Clone + Copy> Bucket<T> {
                     if elem.num_slots > 0 {
                         current_bucket.free(elem_loc, elem_uid);
                     }
-                    elem.storage_offset = ix;
+                    elem.set_storage_offset(ix);
-                    elem.storage_capacity_when_created_pow2 = best_bucket.capacity_pow2;
+                    elem.set_storage_capacity_when_created_pow2(best_bucket.capacity_pow2);
                     elem.num_slots = data.len() as u64;
                     //debug!(                        "DATA ALLOC {:?} {} {} {}",                        key, elem.data_location, best_bucket.capacity, elem_uid                    );
                     if elem.num_slots > 0 {

bucket_map/src/index_entry.rs

@@ -19,13 +19,32 @@ use {
 pub struct IndexEntry {
     pub key: Pubkey, // can this be smaller if we have reduced the keys into buckets already?
     pub ref_count: RefCount, // can this be smaller? Do we ever need more than 4B refcounts?
-    pub storage_offset: u64, // smaller? since these are variably sized, this could get tricky. well, actually accountinfo is not variable sized...
+    storage_offset: u64, // smaller? since these are variably sized, this could get tricky. well, actually accountinfo is not variable sized...
     // if the bucket doubled, the index can be recomputed using create_bucket_capacity_pow2
-    pub storage_capacity_when_created_pow2: u8, // see data_location
+    storage_capacity_when_created_pow2: u8, // see data_location
     pub num_slots: Slot, // can this be smaller? epoch size should ~ be the max len. this is the num elements in the slot list
 }
 
 impl IndexEntry {
+    pub fn init(&mut self, pubkey: &Pubkey) {
+        self.key = *pubkey;
+        self.ref_count = 0;
+        self.storage_offset = 0;
+        self.storage_capacity_when_created_pow2 = 0;
+        self.num_slots = 0;
+    }
+
+    pub fn set_storage_capacity_when_created_pow2(
+        &mut self,
+        storage_capacity_when_created_pow2: u8,
+    ) {
+        self.storage_capacity_when_created_pow2 = storage_capacity_when_created_pow2;
+    }
+
+    pub fn set_storage_offset(&mut self, storage_offset: u64) {
+        self.storage_offset = storage_offset;
+    }
+
     pub fn data_bucket_from_num_slots(num_slots: Slot) -> u64 {
         (num_slots as f64).log2().ceil() as u64 // use int log here?
     }