Skip to content

Commit

Permalink
updated readme
Browse files Browse the repository at this point in the history
  • Loading branch information
@tomfran
tomfran committed Oct 27, 2023
1 parent 375c039 commit e4a8259
Show file tree
Hide file tree
Showing 8 changed files with 624 additions and 80 deletions.
125 changes: 69 additions & 56 deletions README.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -12,31 +12,33 @@ An implementation of the Log-Structured Merge Tree (LSM tree) data structure in
1. [SSTable](#sstable-1)
2. [Skip-List](#skip-list-1)
3. [Tree](#tree-1)
5. [Implementation status](#Implementation-status)
5. [Possible future improvements](#possible-improvements)
6. [References](#references)

## Console
### Console

To interact with a toy tree you can use `./gradlew run -q` to spawn a console.

![console.png](misc%2Fconsole.png)

---

# Architecture

Architecture overview, from SSTables, which are the disk-resident portion of the database, Skip Lists, used
as memory buffers, and finally to the combination of the twos to create insertion, lookup and deletion primitives.

## SSTable

Sorted String Table (SSTable) is a collection of files modelling key-value pairs in sorted order by key.
It is used as a persistent storage for the LSM tree.

### Components
**Components**

- _Data_: key-value pairs in sorted order by key, stored in a file;
- _Sparse index_: sparse index containing key and offset of the corresponding key-value pair in the data;
- _Bloom filter_: a [probabilistic data structure](https://en.wikipedia.org/wiki/Bloom_filter) used to test whether a
key is in the SSTable.

### Key lookup
**Key lookup**

The basic idea is to use the sparse index to find the key-value pair in the data file.
The steps are:
Expand All @@ -49,20 +51,20 @@ The steps are:
The search is as lazy as possible, meaning that we read the minimum amount of data from disk,
for instance, if the next key length is smaller than the one we are looking for, we can skip the whole key-value pair.

### Persistence
**Persistence**

A table is persisted to disk when it is created. A base filename is defined, and three files are present:

- `<base_filename>.data`: data file;
- `<base_filename>.index`: index file;
- `<base_filename>.bloom`: bloom filter file.

**Data format**
Data format:

- `n`: number of key-value pairs;
- `<key_len_1, value_len_1, key_1, value_1, ... key_n, value_n>`: key-value pairs.

**Index format**
Index format:

- `s`: number of entries in the whole table;
- `n`: number of entries in the index;
Expand All @@ -71,7 +73,7 @@ A table is persisted to disk when it is created. A base filename is defined, and
- `s_1, s_2, ..., s_n`: remaining keys after a sparse index entry, used to exit from search;
- `<key_len_1, key_1, ... key_len_n, key_n>`: keys in the index.

**Filter format**
Filter format:

- `m`: number of bits in the bloom filter;
- `k`: number of hash functions;
Expand All @@ -82,8 +84,6 @@ To save space, all integers are stored
in [variable-length encoding](https://nlp.stanford.edu/IR-book/html/htmledition/variable-byte-codes-1.html),
and offsets in the index are stored as [deltas](https://en.wikipedia.org/wiki/Delta_encoding).

---

## Skip-List

A [skip-list](https://en.wikipedia.org/wiki/Skip_list) is a probabilistic data structure that allows fast search,
Expand All @@ -92,7 +92,7 @@ insertion and deletion of elements in a sorted sequence.
In the LSM tree, it is used as an in-memory data structure to store key-value pairs in sorted order by key.
Once the skip-list reaches a certain size, it is flushed to disk as an SSTable.

### Operations details
**Operations details**

The idea of a skip list is similar to a classic linked list. We have nodes with forward pointers, but also levels. We
can think about a
Expand All @@ -108,40 +108,57 @@ we are looking for. Then we move down to the next level and repeat the process u
Insertions, deletions, and updates are done by first locating the element, then performing
the operation on the node. All of them have an average time complexity of `O(log(n))`.

---

## Tree

...
Having defined SSTables and Skip Lists we can obtain the final structure as a combination of the two.
The main idea is to use the latter as an in-memory buffer, while the former efficiently stores flushed
buffers.

**Insertion**

### Components
Each insert goes directly to a Memtable, which is a Skip List under the hood, so the response time is quite fast.
There exists a threshold, over which the mutable structure is made immutable by appending it to the _immmutable
memtables LIFO list_ and replaced with a new mutable list.

...
The immutable memtable list is asynchronously consumed by a background thread, which takes the next available
list and create a disk-resident SSTable with its content.

### Insertion
**Lookup**

...
While looking for a key, we proceed as follows:

### Lookup
1. Look into the in-memory buffer, if the key is recently written it is likely here, if not present continue;
2. Look into the immutable memtables list, iterating from the most recent to the oldest, if not present continue;
3. Look into disk tables, iterating from the most recent one to the oldest, if not present return null.

...
**Deletions**

### Write-ahead logging
To delete a key, we do not need to delete all its replicas, from the on-disk tables, we just need a special
value called _tombstone_. Hence a deletion is the same as an insertion, but with a value set to null. While looking for
a key, if we encounter a null value we simply return null as a result.

...
**SSTable Compaction**

---
The most expensive operation while looking for a key is certainly the disk search, and this is why bloom filters are
crucial for negative
lookup on SSTables. But no bloom filter can save us if too many tables are available to search, hence we need
_compaction_.

When flushing a Memtable, we create an SSTable of level one. When the first level reaches a certain threshold,
all its tables are merged into a level-two table, and so on. This permits us to save storage and query fewer
tables in lookups.

Note that this style of compaction is not standard, there are various sophisticated techniques, but for the sake of
this project this simple level-like compaction works wonders.

# Benchmarks

I am using [JMH](https://openjdk.java.net/projects/code-tools/jmh/) to run benchmarks,
the results are obtained on AMD Ryzen™ 5 4600H with 16GB of RAM and 512GB SSD.

> Take those with a grain of salt, development is still in progress.
To run them use `./gradlew jmh`.

## SSTable
**SSTable**

- Negative access: the key is not present in the table, hence the Bloom filter will likely stop the search;
- Random access: the key is present in the table, the order of the keys is random.
Expand All @@ -154,7 +171,7 @@ c.t.l.sstable.SSTableBenchmark.randomAccess thrpt 5 7989.945 ± 40
```

## Bloom filter
**Bloom filter**

- Add: add keys to a 1M keys Bloom filter with 0.01 false positive rate;
- Contains: test whether the keys are present in the Bloom filter.
Expand All @@ -166,7 +183,7 @@ c.t.l.bloom.BloomFilterBenchmark.contains thrpt 5 3567392.634 ± 220377
```

## Skip-List
**Skip-List**

- Get: get keys from a 100k keys skip-list;
- Add/Remove: add and remove keys from a 100k keys skip-list.
Expand All @@ -179,7 +196,7 @@ c.t.l.memtable.SkipListBenchmark.get thrpt 5 487265.620 ± 8201
```

## Tree
**Tree**

- Get: get elements from a tree with 1M keys;
- Add: add 1M distinct elements to a tree with a memtable size of 2^18
Expand All @@ -191,28 +208,24 @@ c.t.l.tree.LSMTreeGetBenchmark.get thrpt 5 9426.951 ± 241
```

---

## Implementation status

- [x] SSTable
- [x] Init
- [x] Read
- [x] Compaction
- [x] Ints compression
- [x] Bloom filter
- [x] Indexes persistence
- [x] File initialization
- [x] Skip-List
- [x] Operations
- [x] Iterator
- [x] Tree
- [x] Operations
- [x] Background flush
- [x] Background compaction
- [ ] Write ahead log
- [x] Benchmarks
- [x] SSTable
- [x] Bloom filter
- [x] Skip-List
- [x] Tree
## Possible improvements

There is certainly space for improvement on this project:

1. Blocked bloom filters: its a variant of a classic array-like bloom filter which is more cache efficient;
2. Search fingers in the Skip list: the idea is to keep a pointer to the last search, and start from there with
subsequent queries;
3. Proper level compaction in the LSM tree;
4. Write ahead log for the insertions, without this, a crash makes all the in-memory writes disappear;
5. Proper recovery: handle crashes and reboots, using existing SSTables and the write-ahead log.

I don't have the practical time to do all of this, perhaps the first two points will be handled in the future.

## References

- [Database Internals](https://www.databass.dev/) by Alex Petrov, specifically chapters about Log-Structured Storage and
File Formats;
- [A Skip List Cookbook](https://api.drum.lib.umd.edu/server/api/core/bitstreams/17176ef8-8330-4a6c-8b75-4cd18c570bec/content)
by William Pugh.

_If you found this useful or interesting do not hesitate to ask clarifying questions or get in touch!_
Loading

0 comments on commit e4a8259

Please sign in to comment.