Skip to content

ARCHITECTURE.md

Latest commit

 

History

History
189 lines (134 loc) · 4.75 KB

ARCHITECTURE.md

File metadata and controls

189 lines (134 loc) · 4.75 KB
Raw

Architecture — Galaxy Media Pro

Language: English · فارسی

English

Layers

galaxy_ultimate.py          Entry (PyInstaller)
    └── galaxy.app.main     QApplication bootstrap
        └── GalaxyWindow    Shell + footer + navigation
            └── *Page       Image / Video / Audio / Settings / About
                └── *Worker ImageWorker | MediaWorker
                    └── services.*     FFmpeg · pipeline · scheduler

High-level diagram (same as README):

flowchart TB
    subgraph UI["UI Layer"]
        MW[MainWindow]
        P1[ImagePage]
        P2[VideoPage]
        P3[AudioPage]
    end

    subgraph Workers["Workers"]
        IW[ImageWorker · ThreadPool]
        MWk[MediaWorker · ThreadPool]
    end

    subgraph Engine["Engine"]
        FP[ffmpeg_pipeline]
        HP[hw_probe]
        JS[job_scheduler]
        SEM[GlobalGpuSemaphore]
    end

    MW --> P1 & P2 & P3
    P1 --> IW
    P2 & P3 --> MWk
    MWk --> FP & JS & SEM
    IW --> JS
    FP --> HP
Loading

Pipeline tiers (video)

Tier Condition Path
COPY copy_streams / fast_copy -c copy
CPU VP9/AV1/GIF · force_cpu · no GPU libx264/libx265
TIER1 GPU + full path (scale_cuda/…) hw decode + gpu filter + hw encode
TIER2 GPU encode · CPU filters hybrid

File: galaxy/services/ffmpeg_pipeline.py
Selection: pick_pipeline_tier() · build_video_command()

Hardware probe

ffmpeg -encoders / -hwaccels / -filters
        ↓
SystemProfile { backends, gpu_devices, recommended_parallel }
        ↓
cache ~/.galaxy_converter/hw_cache.json  (TTL 24h)

File: galaxy/services/hw_probe.py

Job scheduler

compute_slots(app_settings, JobDescriptor) → int
GlobalGpuSemaphore.configure(gpu_slots)    # startup
GlobalGpuSemaphore.acquire/release()       # per GPU encode
per_job_cpu_threads(...)                   # FFmpeg -threads
Job kind Smart slots (typical)
Image min(cpu-1, 6) · ThreadPool
Audio min(cpu-1, 4) · ThreadPool
Video GPU gpu_slots cap
Video CPU min(cpu-1, 3)

File: galaxy/services/job_scheduler.py

GPU acquire is interruptible by Stop (short acquire loop + is_running check) so parallel cancel does not block for minutes.

Workers

MediaWorker (QThread)

  • Sequential or ThreadPoolExecutor
  • GPU sem → encode → release before CPU fallback
  • _run_ffmpeg progress from stderr time=

ImageWorker (QThread)

  • Sequential or ThreadPoolExecutor (no new process spawn — avoids multi-window on Windows)
  • Top-level convert_image_task (Windows spawn-safe)
  • Module-level output lock

Shutdown

UI idle flags and QThread.isRunning() can diverge briefly after finish. Window close / force-reset join workers (stop → wait → terminate-on-timeout) before dropping references.

Settings schema

{
  "language": "fa",
  "theme": "dark_cyan",
  "ui_mode": "simple|professional",
  "parallel_mode": "smart|sequential|parallel",
  "accel_mode": "auto|cpu|nvidia|amd|intel",
  "nvenc_preset": "p4",
  "max_parallel_jobs": 0,
  "cpu_threads": 0
}

File: galaxy/services/settings.py

i18n

  • galaxy_i18n.py — STRINGS['fa'|'en']
  • UI: I18n.t(key)
  • Workers: WorkerI18n (no Qt)

Build / bundle

Spec Output
GalaxyConverter.spec onedir
GalaxyConverter.portable.spec onefile + UPX

Hidden imports: hw_probe, ffmpeg_pipeline, job_scheduler, worker_i18n, …

فارسی

لایه‌ها

همان درخت لایه‌ها و دیاگرام Mermaid بخش انگلیسی — موتور در galaxy/ با entry در galaxy_ultimate.py.

Pipeline tiers (ویدیو)

Tier شرط مسیر
COPY copy_streams / fast_copy -c copy
CPU VP9/AV1/GIF · force_cpu · بدون GPU libx264/libx265
TIER1 GPU + مسیر کامل hw decode + gpu filter + hw encode
TIER2 encode روی GPU · فیلتر CPU hybrid

Hardware probe / Job scheduler / Workers

  • probe → SystemProfile + cache ۲۴ ساعته
  • GlobalGpuSemaphore برای encode موازی GPU؛ acquire با Stop قطع‌پذیر است
  • ImageWorker بدون spawn پردازش جدید (جلوگیری از چند پنجره در Windows)
  • closeEvent حتی وقتی UI idle است worker در حال اجرا را join می‌کند

i18n

کلیدها در galaxy_i18n.py برای fa و en؛ Workers با WorkerI18n بدون Qt.

← Docs index · Development · Testing