VictoriaMetrics vs OpenTelemetry 指標採集架構深度評比與選型指南 ¶

flowchart TD
    subgraph VMAgentPhilosophy["VictoriaMetrics / vmagent 設計哲學"]
        VM1["專注時序指標 (Metrics-Centric)"]
        VM2["極簡資源開銷 (Ultra-Low Footprint)"]
        VM3["完全相容 Prometheus 生態 (Drop-in Replacement)"]
        VM4["內建本地磁碟容錯佇列 (Crash-Resilient Disk Queue)"]
    end

    subgraph OTelPhilosophy["OpenTelemetry Collector 設計哲學"]
        OT1["跨領域大一統 (Unified Traces / Metrics / Logs)"]
        OT2["中立標準協議 (Vendor-Neutral OTLP)"]
        OT3["高度可插拔管線 (Receivers ➔ Processors ➔ Exporters)"]
        OT4["強大語意轉換與豐富化 (OTTL / Resource Attributes)"]
    end

flowchart TD
    subgraph ArchitectureComparison["指標採集管線架構對比"]
        direction TB

        subgraph ModeVM["vmagent 採集架構"]
            VM_Target["Prometheus Exporters / Push Endpoints"] --> VM_Scrape["Scrape 模組<br/>(支援 Service Discovery)"]
            VM_Scrape --> VM_Relabel["Stream Relabeling<br/>(標籤重寫 / 指標過濾 / 串流降採樣)"]
            VM_Relabel --> VM_Disk["本地磁碟緩衝<br/>(-remoteWrite.tmpDataPath)"]
            VM_Disk --> VM_Push["Streaming Remote Write<br/>(高效 ZSTD 壓縮傳輸)"]
            VM_Push --> VM_Backend[("VictoriaMetrics / TSDB")]
        end

        subgraph ModeOTel["OTel Collector 採集架構"]
            OT_Target["OTLP SDK / Prometheus Exporters"] --> OT_Recv["Receivers<br/>(otlp, prometheus, hostmetrics)"]
            OT_Recv --> OT_Proc["Processors<br/>(memory_limiter, batch, transform/OTTL)"]
            OT_Proc --> OT_Queue["Sending Queue<br/>(記憶體佇列 / file_storage 擴充)"]
            OT_Queue --> OT_Export["Exporters<br/>(otlp, prometheusremotewrite, kafka)"]
            OT_Export --> OT_Backend[("TSDB / APM / Data Lake")]
        end
    end

flowchart TD
    subgraph EdgeLayer["邊緣節點 / 叢集內部採集層"]
        AppOTel["應用程式 (OTel SDK 注入)"]
        AppProm["微服務 (Prometheus Exporters)"]
        NodeExp["主機 / 節點硬體指標 (Node Exporter)"]
    end

    subgraph CollectionHub["中樞分流與處理層"]
        VMAgent["vmagent (DaemonSet / Sidecar)<br/>專責處理海量高頻率 Prometheus 指標"]
        OTelCol["OTel Collector (Gateway 集群)<br/>專責處理 OTLP Traces、Logs 與業務語意指標"]
    end

    subgraph BackendStorage["中央儲存與分析平台"]
        VMCluster[("VictoriaMetrics Cluster<br/>(時序指標統一大水庫)")]
        TraceStorage[("Jaeger / Tempo<br/>(分散式鏈路追蹤)")]
        LogStorage[("VictoriaLogs / Loki<br/>(日誌全文檢索)")]
    end

    AppProm -->|"Pull Scrape"| VMAgent
    NodeExp -->|"Pull Scrape"| VMAgent
    AppOTel -->|"Push OTLP"| OTelCol

    VMAgent -->|"ZSTD 高效 Remote Write"| VMCluster
    OTelCol -->|"Prometheus Remote Write / OTLP"| VMCluster
    OTelCol -->|"Trace Export"| TraceStorage
    OTelCol -->|"Log Export"| LogStorage

flowchart TD
    Start["評估指標採集需求"] --> Q1{"是否需要同時處理<br/>Trace 或 Log？"}

    Q1 -- "是 (多元數據)" --> Q2{"是否要求嚴格遵從<br/>CNCF OTel 統一標準？"}
    Q2 -- "是" --> ChoiceOTel["優先採用 OpenTelemetry Collector<br/>(統一數據管線)"]
    Q2 -- "否" --> ChoiceHybrid["採用混合架構：<br/>vmagent (指標) + Vector/OTel (日誌與鏈路)"]

    Q1 -- "否 (純指標)" --> Q3{"當前環境是否深度相依<br/>Prometheus 與 Service Discovery？"}
    Q3 -- "是" --> Q4{"邊緣資源是否受限？<br/>(希望佔用記憶體 < 100MB)"}
    Q4 -- "是" --> ChoiceVM["強烈推薦 VictoriaMetrics (vmagent)<br/>(極低資源、內建磁碟快取、極高穩定性)"]
    Q4 -- "否" --> ChoiceVMOptional["推薦 vmagent，亦可評估 OTel Collector<br/>(視未來是否有 Trace 整合需求而定)"]

    Q3 -- "否 (現代微服務原生 OTLP)" --> ChoiceOTelDirect["採用 OTel Collector<br/>或直接以 OTLP 推送至 VictoriaMetrics"]