mirror of
https://gh.wpcy.net/https://github.com/discourse/discourse.git
synced 2026-06-18 23:54:54 +08:00
06b32204c0
Previously, the migrations tooling was a single flat `migrations/` tree, autoloaded by one global Zeitwerk loader and driven by a Thor CLI, so each planned next step had nowhere clean to land. This change splits it into four `path:`-referenced gems — `migrations-core`, `migrations-tooling`, `migrations-converters`, and `migrations-importer` — served by a single Samovar-based `disco` binary, without rewriting any domain logic. ### Why now The DSL refactor that replaced the IntermediateDB YAML config just landed, which is the cheapest moment to do this. Everything queued behind it — column coverage verification, the `discourse-migrations` validation plugin, the transformer framework, and private converter isolation — either has nowhere clean to land in the flat tree or would have to be retrofitted into a gem layout later. Doing the split now, while it's still a pure move (suite green, no domain logic touched), is far cheaper than after another round of features has built on the flat layout. ### What changes - **Four gems under `migrations/`**, all `path:`-referenced from the root `Gemfile` (nothing is published to RubyGems): `core` (CLI framework, UI, DB infrastructure, IntermediateDB, and the conversion framework), `tooling` (schema DSL and `schema` commands), `converters` (implementations and source adapters), and `importer` (row and uploads import). - **A single CLI binary:** `migrations/bin/cli` (Thor) becomes `disco` (Samovar), with each gem registering its own commands. Same surface — `convert`, `import`, `upload`, `schema generate|validate|…` — and Rails is still booted lazily. - **Isolated test suites:** each gem runs its own no-Rails specs in a new CI job, while the existing job keeps running the Rails-integration specs.
3 KiB
Vendored
3 KiB
Vendored
Benchmark Results
Here are the latest benchmark results. All benchmarks ran with ruby 3.2.3 (2024-01-18 revision 52bb2ac0a6) [x86_64-linux]
database_write.rb
Compares the INSERT speed of SQLite and DuckDB
| Database | User Time | System Time | Total Time | Real Time |
|---|---|---|---|---|
| SQLite3 | 99.212731 | 4.883932 | 104.096663 | 104.233575 |
| Extralite | 45.396666 | 4.457247 | 49.853913 | 50.065680 |
| DuckDB | 49.012140 | 10.210994 | 59.223134 | 52.095679 |
hash_vs_data.rb
Compares the INSERT speed when the data is bound as Hash or Data class
Extralite regular 868.703k (± 2.2%) i/s - 8.744M in 10.070511s
Extralite hash 579.753k (± 1.2%) i/s - 5.838M in 10.071266s
Extralite data 672.752k (± 0.8%) i/s - 6.790M in 10.093191s
Extralite data/array 826.296k (± 0.9%) i/s - 8.318M in 10.067518s
SQLite3 regular 362.037k (± 0.7%) i/s - 3.628M in 10.021699s
SQLite3 hash 308.647k (± 1.1%) i/s - 3.111M in 10.081159s
SQLite3 data/hash 288.747k (± 2.7%) i/s - 2.890M in 10.018335s
Comparison:
Extralite regular: 868702.8 i/s
Extralite data/array: 826295.7 i/s - 1.05x slower
Extralite data: 672752.0 i/s - 1.29x slower
Extralite hash: 579753.5 i/s - 1.50x slower
SQLite3 regular: 362037.0 i/s - 2.40x slower
SQLite3 hash: 308646.7 i/s - 2.81x slower
SQLite3 data/hash: 288747.1 i/s - 3.01x slower
parameter_binding.rb
A similar benchmark that looks at various parameter binding styles, especially in Extralite
Extralite regular 825.159 (± 0.6%) i/s - 8.316k in 10.078450s
Extralite named 571.135 (± 0.4%) i/s - 5.742k in 10.053796s
Extralite index 769.273 (± 1.0%) i/s - 7.742k in 10.065238s
Extralite array 860.549 (± 0.5%) i/s - 8.624k in 10.021749s
SQLite3 regular 361.745 (± 0.6%) i/s - 3.636k in 10.051588s
SQLite3 named 307.875 (± 0.6%) i/s - 3.090k in 10.036954s
Comparison:
Extralite array: 860.5 i/s
Extralite regular: 825.2 i/s - 1.04x slower
Extralite index: 769.3 i/s - 1.12x slower
Extralite named: 571.1 i/s - 1.51x slower
SQLite3 regular: 361.7 i/s - 2.38x slower
SQLite3 named: 307.9 i/s - 2.80x slower
time_formatting.rb
Fastest way of converting Time into String?
Time#iso8601 1.084M (± 0.9%) i/s - 10.875M in 10.033905s
Time#strftime 1.213M (± 1.4%) i/s - 12.200M in 10.056764s
DateTime#iso8601 2.419M (± 1.8%) i/s - 24.296M in 10.046295s
Comparison:
DateTime#iso8601: 2419162.1 i/s
Time#strftime: 1213390.0 i/s - 1.99x slower
Time#iso8601: 1083922.8 i/s - 2.23x slower
write.rb
Compares writing lots of data into a single SQLite database.
single writer 43.9766 seconds
forked writer - same DB 53.5112 seconds
forked writer - multi DB 3.0815 seconds
