🐏 mogoz

Search

SearchSearch

sqlite

Jun 30, 2024, 7 min read

tags : Database

Basics

  • SQLite is a relational database management system contained in a C library. It’s the standard bearer of the “single process application” architecture.
  • sqlite currently is single-writer, multi-reader. There are ways to do multi-writer, see cr-sqlite (Local First Software (LoFi))

Gotchas

  • AUTOINCREMENT
    • INSERT, if a INTEGER PRIMARY KEY column is not explicitly given a value, then it will be filled automatically with an unused integer, usually one more than the largest ids in use.
      • This is true regardless use of AUTOINCREMENT
    • In sqlite, AUTOINCREMENT prevents the reuse of ids over the lifetime of the database
  • Weak Types
    • int column can take string
    • It does not enforce data type constraints. Data of any type can (usually) be inserted into any column. You can put arbitrary length strings into integer columns, floating point numbers in boolean columns, or dates in character columns.
    • But it does type affinity, if you try to insert a string which can be casted into an int, it’ll probaby try doing that

Internals

VFS

  • SQLite has a VFS interface, which can be used to implement a file system on any system

WAL

  • ???
    • checkpoint is an op which compacts WAL frames to drop ones that are out-of-date
    • checkpoint operation is one where the contents of the WAL file is applied to the database.
  • What happens
    • In WAL mode, write operations append to a log file stored alongside SQLite’s main database file. Readers check both the WAL file and the main database to satisfy queries.
    • SQLite automatically checkpoints pages from the WAL back to the main database.
  • Purpose of WAL
    • Durability (primary)
    • Allow concurrent access (secondary)
      • This is why you can implement WAL without shared memory

Page & Cache

  • database file consists of one or more “pages”.
  • The minimum size of an SQLite database is one page for each table and each index.
  • See https://www.sqlite.org/pgszchng2016.html
  • For any query, sqlite will do certain amount of page reads
  • Page types
    • Internal
    • Leaf
    • Overflow

Settings/Configuration

Concurrency

TypeROLLBACKWALWAL2(future)
multi-readallowedallowedallowed
read-writenot allowedallowedallowed
multi-writenot allowednot allowedallowed
  • Each connection can run one transaction at a time
  • Multiple connections = Multiple transactions = concurrency
    • May be allowed over multiple sqlite pages /within same sqlite page
  • WARNING: Some VFS (eg. OPFS sqlite vfs) don’t allow >1 connection
    • In those cases, no concurrency, everything sequential

Suggested settings

  • read-only: max open connection to some large no
  • read-write: max 1 open connection

Consistency

Timeouts

  • busy timeout to some large value, like 5000
  • Any other transactions that attempt to write at the same time will fail with SQLITE_BUSY.

Transaction

  • Use transactions for multi-query methods
  • read/write pool should also use BEGIN IMMEDITATE when starting transactions (_txlock=immediate), to avoid certain issues that can result in “database is locked” errors.

Replication

See Data Replication

  • litestream: Disaster recovery tool for SQLite. It runs as a background process and safely replicates changes incrementally to another file or S3.
    • It takes control of SQLite’s WAL-mode journaling by opening an indefinite read transaction that prevents automatic checkpoints that’s default for sqlite.
    • It then captures WAL updates, replicate them, and trigger the checkpointing.
  • litefs: LiteFS sets up a FUSE mount so once you have that in place then you can create a database within that mount directory and it is automatically replicated.
  • cr-sqlite: cr-sqlite adds multi-master replication and partition tolerance to SQLite via crdt and/or causally ordered event logs. (Also see Local First Software (LoFi))
  • maxpert/marmot : Multi-writer distributed SQLite based on NATS (See Message Queue (MQ))
  • clouflare D1: offers d1 sqlite w replication
  • Distributed sqlite

Using in applications

General use

  • We need to worry about state when writing our applications cuz
    • The primitives we have to express mutations
    • Support for transactions against in-memory data structures
    • Support for constraints on im-memory data
    • This sort of can be sidestepped by using sqlite as our data store for application state
  • SQLite is that it’s in-process so you avoid most per-query latency so you don’t have to worry as much about N+1 query performance issues.

Use on the web

See The Current State of SQLite Persistence on the Web

Technologies

  • SQL.js
  • wa-sqlite
    • came in 2021, to wasm
    • persistence to indexDB but slow
    • later adopted things from absurd-sql which made it faster.
      • absurd-sql is a backend for sql.js
      • absurd-sql was inspired by phiresky’s github pages experiment
  • sqlite wasm build : official wasm version from sqlite folks
  • WebSQL
    • built-in support in Chrome and older versions of Safari. It had too many limitations
    • phasing out

Storage

  • OPFS

    • Ideal for sqlite but currently API not super mature
    • Concurrency
      • Opening a file adds exclusive lock and no other read-write can happen in that connection.
      • OPFS currently only allows one connection, so you can’t really use sqlite WAL when using OPFS. Because it’s just one connection, you can’t even have concurrent reads.

  • IndexDB

    • Storing individual blocks of data as IndexedDB objects
    • Concurrency
      • Exclusive lock on an object store for any write. No concurrent r-w
      • Can support concurrent reads

Async

SQlite uses synchronous calls but the Javascript Runtime and Browser are async (See Concurrency)

  • Asyncify
    • Translates sync to async
    • more build size, performance reduction, bugs in asyncify could lead to sqlite
  • SharedArrayBuffer + Atomics API
    • Separate Worker process for the file system operations
  • OPFS syncAccessHandle
    • synchronous read and write access to files.
    • While operations on an open file are synchronous now, opening a file is still an asynchronous operation. Needs workaround for that.

On Github Pages

phiresky built a readonly VFS on top of HTTP Range requests with SQL.js

  • See Hosting SQLite databases on Github Pages
  • They set sqlite page_size to 1KiB
  • i.e each total response size(query result) is 270KiB
    • It’ll need to do 270 requests
    • Unless you prefetch, which can be less then. Like 20 requests will fetch 270KiB
    • “I implemented a pre-fetching system that tries to detect access patterns through three separate virtual read heads and exponentially increases the request size for sequential reads.”

On the edge

  • Turso does this
  • Turso is essentially a globally distributed single-write multi-read database. All writes must be made to a central primary instance and replicated to edge nodes. Before the replication completes, readers can see inconsistent results.

Extensions

run-time loadable extension

webassembly functions

libSQL

  • fork of sqlite w extended capabilities
  • It also implements the PostgreSQL wire protocol, so postgres tools can connect to it thinking it’s postgres, ofc not all commands will work
  • sqld can be connected via http and WebSockets (Hrana). See this

Graph View

Backlinks