Optimize MemoryDB getOrCreateBlockStore with atomic operation 'computeIfAbsent'

[mpalriwal-Netflix]

The separate get and computeIfAbsent calls can introduce race conditions in a multithreaded environment. If multiple threads check for the presence of the blkStore at the same time, they might all find it null and proceed to create new instances, leading to inconsistencies.

CHMap computeIfAbsent is atomic, ensuring that only one thread creates the blkStore if it is absent. This makes the operation thread-safe without additional synchronization.

[brharrington]

Updates are single threaded so there shouldn't be a race for updating the entry. Would need to verify this doesn't result in a lambda allocation per call, will check when I get some time.

[mpalriwal-Netflix]

If its single threaded we can simply avoid double checking and remove computeIfAbsent completely

private def getOrCreateBlockStore(id: ItemId, tags: Map[String, String]): BlockStore = {
  var blkStore = data.get(id)
  if (blkStore == null) {
    blkStore = new MemoryBlockStore(step, blockSize, numBlocks)
    data.put(id, blkStore)
    index.update(BlockStoreItem.create(id, tags, blkStore))
  }
  blkStore
}