evmap

A lock-free, eventually consistent, concurrent multi-value map.

This map implementation allows reads and writes to execute entirely in parallel, with no
implicit synchronization overhead. Reads never take locks on their critical path, and neither
do writes assuming there is a single writer (multi-writer is possible using a Mutex), which
significantly improves performance under contention.

The trade-off exposed by this module is one of eventual consistency: writes are not visible to
readers except following explicit synchronization. Specifically, readers only see the
operations that preceeded the last call to WriteHandle::refresh by a writer. This lets
writers decide how stale they are willing to let reads get. They can refresh the map after
every write to emulate a regular concurrent HashMap, or they can refresh only occasionally to
reduce the synchronization overhead at the cost of stale reads.

For read-heavy workloads, the scheme used by this module is particularly useful. Writers can
afford to refresh after every write, which provides up-to-date reads, and readers remain fast
as they do not need to ever take locks.

The map is multi-value, meaning that every key maps to a collection of values. This
introduces some memory cost by adding a layer of indirection through a Vec for each value,
but enables more advanced use. This choice was made as it would not be possible to emulate such
functionality on top of the semantics of this map (think about it -- what would the operational
log contain?).

To faciliate more advanced use-cases, each of the two maps also carry some customizeable
meta-information. The writers may update this at will, and when a refresh happens, the current
meta will also be made visible to readers. This could be useful, for example, to indicate what
time the refresh happened.

Examples

Single-reader, single-writer

// new will use the default HashMap hasher, and a meta of ()
// note that we get separate read and write handles
// the read handle can be cloned to have more readers
let (book_reviews_r, mut book_reviews_w) = evmap::new();

// review some books.
book_reviews_w.insert("Adventures of Huckleberry Finn",    "My favorite book.");
book_reviews_w.insert("Grimms' Fairy Tales",               "Masterpiece.");
book_reviews_w.insert("Pride and Prejudice",               "Very enjoyable.");
book_reviews_w.insert("The Adventures of Sherlock Holmes", "Eye lyked it alot.");

// at this point, reads from book_reviews_r will not see any of the reviews!
assert_eq!(book_reviews_r.len(), 0);
// we need to refresh first to make the writes visible
book_reviews_w.refresh();
assert_eq!(book_reviews_r.len(), 4);
// reads will now return Some() because the map has been initialized
assert_eq!(book_reviews_r.get_and("Grimms' Fairy Tales",, rs, rs.len()), Some(1));

// remember, this is a multi-value map, so we can have many reviews
book_reviews_w.insert("Grimms' Fairy Tales",               "Eh, the title seemed weird.");
book_reviews_w.insert("Pride and Prejudice",               "Too many words.");

// but again, new writes are not yet visible
assert_eq!(book_reviews_r.get_and("Grimms' Fairy Tales",, rs, rs.len()), Some(1));

// we need to refresh first
book_reviews_w.refresh();
assert_eq!(book_reviews_r.get_and("Grimms' Fairy Tales",, rs, rs.len()), Some(2));

// oops, this review has a lot of spelling mistakes, let's delete it.
// empty deletes *all* reviews (though in this case, just one)
book_reviews_w.empty("The Adventures of Sherlock Holmes");
// but again, it's not visible to readers until we refresh
assert_eq!(book_reviews_r.get_and("The Adventures of Sherlock Holmes",, rs, rs.len()), Some(1));
book_reviews_w.refresh();
assert_eq!(book_reviews_r.get_and("The Adventures of Sherlock Holmes",, rs, rs.len()), None);

// look up the values associated with some keys.
let to_find = ["Pride and Prejudice", "Alice's Adventure in Wonderland"];
for book in &to_find {
    let reviewed = book_reviews_r.get_and(book,, reviews, {
        for review in reviews {
            println!("{}: {}", book, review);
        }
    });
    if reviewed.is_none() {
        println!("{} is unreviewed.", book);
    }
}

// iterate over everything.
book_reviews_r.for_each(, book, reviews, {
    for review in reviews {
        println!("{}: \"{}\"", book, review);
    }
});

Reads from multiple threads are possible by cloning the ReadHandle.

use std::thread;
let (book_reviews_r, mut book_reviews_w) = evmap::new();

// start some readers
let readers: Vec<_> = (0..4).map(, _, {
    let r = book_reviews_r.clone();
    thread::spawn(move, {
        loop {
            let l = r.len();
            if l == 0 {
                thread::yield_now();
            } else {
                // the reader will either see all the reviews,
                // or none of them, since refresh() is atomic.
                assert_eq!(l, 4);
                break;
            }
        }
    })
}).collect();

// do some writes
book_reviews_w.insert("Adventures of Huckleberry Finn",    "My favorite book.");
book_reviews_w.insert("Grimms' Fairy Tales",               "Masterpiece.");
book_reviews_w.insert("Pride and Prejudice",               "Very enjoyable.");
book_reviews_w.insert("The Adventures of Sherlock Holmes", "Eye lyked it alot.");
// expose the writes
book_reviews_w.refresh();

// you can read through the write handle
assert_eq!(book_reviews_w.len(), 4);

// the original read handle still works too
assert_eq!(book_reviews_r.len(), 4);

// all the threads should eventually see .len() == 4
for r in readers.into_iter() {
    assert!(r.join().is_ok());
}

If multiple writers are needed, the WriteHandle must be protected by a Mutex.

use std::thread;
use std::sync::{Arc, Mutex};
let (book_reviews_r, mut book_reviews_w) = evmap::new();

// start some writers.
// since evmap does not support concurrent writes, we need
// to protect the write handle by a mutex.
let w = Arc::new(Mutex::new(book_reviews_w));
let writers: Vec<_> = (0..4).map(, i, {
    let w = w.clone();
    thread::spawn(move, {
        let mut w = w.lock().unwrap();
        w.insert(i, true);
        w.refresh();
    })
}).collect();

// eventually we should see all the writes
while book_reviews_r.len() < 4 { thread::yield_now(); };

// all the threads should eventually finish writing
for w in writers.into_iter() {
    assert!(w.join().is_ok());
}

Implementation

Under the hood, the map is implemented using two regular HashMaps, an operational log,
epoch counting, and some pointer magic. There is a single pointer through which all readers
go. It points to a HashMap, which the readers access in order to read data. Every time a read
has accessed the pointer, they increment a local epoch counter, and they update it again when
they have finished the read (see #3 for more information). When a write occurs, the writer
updates the other HashMap (for which there are no readers), and also stores a copy of the
change in a log (hence the need for Clone on the keys and values). When
WriteHandle::refresh is called, the writer, atomically swaps the reader pointer to point to
the other map. It then waits for the epochs of all current readers to change, and then replays
the operational log to bring the stale map up to date.

Since the implementation uses regular HashMaps under the hood, table resizing is fully
supported. It does, however, also mean that the memory usage of this implementation is
approximately twice of that of a regular HashMap, and more if writes rarely refresh after
writing.

Small Vector Optimization

By default, the value-set for each key in the map uses the smallvec crate to keep a
maximum of one element stored inline with the map, as opposed to separately heap-allocated
with a plain Vec. Operations such as Fit and Replace will automatically switch
back to the inline storage if possible. This is ideal for maps that mostly use one
element per key, as it can improvate memory locality with less indirection.

If this is undesirable, simple set:

default-features = false

in the evmap dependency entry, and Vec will always be used internally.

Performance

I've run some benchmarks of evmap against a standard Rust HashMap protected
by a reader-writer
lock, as well as
against chashmap — a crate which provides
"concurrent hash maps, based on bucket-level multi-reader locks". The
benchmarks were run using the binary in benchmark/ on
a 40-core machine with Intel(R) Xeon(R) CPU E5-2660 v3 @ 2.60GHz CPUs.

The benchmark runs a number of reader and writer threads in tight loops, each
of which does a read or write to a random key in the map respectively. Results
for both uniform and skewed distributions are provided below. The benchmark
measures the average number of reads and writes per second as the number of
readers and writers increases.

Preliminary results show that evmap performs well under contention,
especially on the read side. This benchmark represents the worst-case usage of
evmap in which every write also does a refresh. If the map is refreshed
less often, performance increases (see bottom plot).