pub struct ConcurrentMap<K, V, const FANOUT: usize = 64, const LOCAL_GC_BUFFER_SIZE: usize = 128>where
    K: 'static + Clone + Debug + Minimum + Ord + Send + Sync,
    V: 'static + Clone + Debug + Send + Sync,
{ /* private fields */ }
Expand description

A lock-free B+ tree.

Note that this structure is Send but NOT Sync, despite being a lock-free tree. This is because the inner reclamation system, provided by the ebr crate completely avoids atomic operations in its hot path for efficiency. If you want to share ConcurrentMap between threads, simply clone it, and this will set up a new efficient thread-local memory reclamation state.

If you want to use a custom key type, you must implement the Minimum trait, allowing the left-most side of the tree to be created before inserting any data.

The FANOUT const generic must be greater than 3. This const generic controls how large the fixed-size array for either child pointers (for index nodes) or values (for leaf nodes) will be. A higher value may make reads and scans faster, but writes will be slower because each modification performs a read-copy-update of the full node. In some cases, it may be preferable to wrap complex values in an Arc to avoid higher copy costs.

The LOCAL_GC_BUFFER_SIZE const generic must be greater than 0. This controls the epoch-based reclamation granularity. Garbage is placed into fixed-size arrays, and garbage collection only happens after this array fills up and a final timestamp is assigned to it. Lower values will cause replaced values to be dropped more quickly, but the efficiency will be lower. Values that are extremely high may cause undesirable memory usage because it will take more time to fill up each thread-local garbage segment.

Implementations

Atomically get a value out of the tree that is associated with this key.

Atomically insert a key-value pair into the tree, returning the previous value associated with this key if one existed.

Atomically remove the value associated with this key from the tree, returning the previous value if one existed.

Atomically compare and swap the value associated with this key from the old value to the new one. An old value of None means “only create this value if it does not already exist”. A new value of None means “delete this value, if it matches the provided old value”. If successful, returns the old value if it existed. If unsuccessful, returns the proposed new value.

Examples
let tree = concurrent_map::ConcurrentMap::<usize, usize>::default();

// key 1 does not yet exist
assert_eq!(tree.get(&1), None);

// uniquely create value 10
tree.cas(1, None, Some(10)).unwrap();

assert_eq!(tree.get(&1).unwrap(), 10);

// compare and swap from value 10 to value 20
tree.cas(1, Some(&10_usize), Some(20)).unwrap();

assert_eq!(tree.get(&1).unwrap(), 20);

// if we guess the wrong current value, a CasFailure is returned
// which will tell us what the actual current value is (which we
// failed to provide) and it will give us back our proposed new
// value.
let cas_result = tree.cas(1, Some(&999999_usize), Some(30));

let expected_cas_failure = Err(concurrent_map::CasFailure {
    actual: Some(20),
    returned_new_value: Some(30),
});

assert_eq!(cas_result, expected_cas_failure);

// conditionally delete
tree.cas(1, Some(&20_usize), None).unwrap();

assert_eq!(tree.get(&1), None);

Iterate over the tree.

This is not an atomic snapshot, and it caches B+tree leaf nodes as it iterates through them to achieve high throughput. As a result, the following behaviors are possible:

  • may (or may not!) return values that were concurrently added to the tree after the iterator was created
  • may (or may not!) return items that were concurrently deleted from the tree after the iterator was created
  • If a key’s value is changed from one value to another one after this iterator is created, this iterator might return the old or the new value.

But, you can be certain that any key that existed prior to the creation of this iterator, and was not changed during iteration, will be observed as expected.

Iterate over a range of the tree.

This is not an atomic snapshot, and it caches B+tree leaf nodes as it iterates through them to achieve high throughput. As a result, the following behaviors are possible:

  • may (or may not!) return values that were concurrently added to the tree after the iterator was created
  • may (or may not!) return items that were concurrently deleted from the tree after the iterator was created
  • If a key’s value is changed from one value to another one after this iterator is created, this iterator might return the old or the new value.

But, you can be certain that any key that existed prior to the creation of this iterator, and was not changed during iteration, will be observed as expected.

Trait Implementations

Returns a copy of the value. Read more
Performs copy-assignment from source. Read more
Returns the “default value” for a type. Read more

Auto Trait Implementations

Blanket Implementations

Gets the TypeId of self. Read more
Immutably borrows from an owned value. Read more
Mutably borrows from an owned value. Read more

Returns the argument unchanged.

Calls U::from(self).

That is, this conversion is whatever the implementation of From<T> for U chooses to do.

The resulting type after obtaining ownership.
Creates owned data from borrowed data, usually by cloning. Read more
Uses borrowed data to replace owned data, usually by cloning. Read more
The type returned in the event of a conversion error.
Performs the conversion.
The type returned in the event of a conversion error.
Performs the conversion.