與 Nginx 同行，Pipy究竟有何能耐？

{"type":"doc","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"embedcomp","attrs":{"type":"video","data":{"id":"378021","name":"Flomesh創始人&CEO 專訪","poster":"","url":"https:\/\/media001.geekbang.org\/b090cddd49674409990060c021d2a5d1\/874fd88eed38407986454c6d13b56d7d-dbe5194cf055bb770a226d151371b861-sd.m3u8"}}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"作者 | 施堯"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"blockquote","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如何精細化管理企業內網流量？來看看 Flomesh 創始人怎麼說"}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"互聯網包羅萬象的今天，流量管理無處不在。這一普遍需求，催生了很多流量管理軟件。除了Nginx外，具有代表性的既有Apache httpd、Tomcat、untertow、Lighttpd、Microsoft IIS、IBM WebSphere、Oracle Weblogic等老牌web和應用服務器，也有OpenResty，Envoy等專注於流量管理的新力量。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"隨着內網微服務等大型分佈式系統的發展，技術棧從傳統的“內容編程”向“流量編程”演化。這些流量管理軟件都有自身擅長的領域，在互聯網發展、終端形態不斷變化的過程中發揮着重要的作用。與此同時，流量管理從互聯網領域快速進入雲上和企業內網領域，“內(網)外(網)結合、(雲)上(雲)下結合、東西(向流量)南北(向流量)結合”的流量管理能力成爲雲時代的新熱點。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Flomesh 成立於 2018年，於今年初開源了Pipy。Pipy 最初爲企業內網提供流量管理服務，經過這幾年的打磨，Pipy如今已成爲“流量編程”領域的新力量，在衆多業務場景中可與Nginx和Envoy一較高低，同時在企業內網這塊空地上獨樹一幟。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"爲探索 Pipy 的成長之路以及背後的設計思想，以及對“流量編程”領域的前沿思考。近期，小編趁 InfoQ Arch Summit 上海站專訪了 Pipy 開發者蔡書。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.infoq.cn\/resource\/image\/a9\/ef\/a9b340736a749d1c696f2b2eef8934ef.jpg","alt":null,"title":"","style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":"","fromPaste":false,"pastePass":false}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"蔡書，Flomesh聯合創始人。開源技術愛好者，先後在IBM、RedHat供職多年。2019年開始創建Flomesh，2021年開源其核心Proxy組件Pipy，靜態內容服務和HTTP代理性能全面超越Nginx。目前全職開發和推廣Pipy及其商業化產品，在多家股份制商業銀行成功上線運行，支撐單一客戶日DAU超過一億。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"以下是訪談內容整理，分享給大家，Enjoy～"}]},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"定位於流量編程引擎，Pipy的誕生是爲了高效解決流量處理需求"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"主持人：您之前提到過開源的Pipy對標Nginx，而且在性能、擴展性等方面都比Nginx更有優勢，那是不是可以認爲Pipy就是Nginx的替代品？"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"**蔡書：**這個問題非常好，身邊好多朋友都問過我這個問題。首先肯定的說我們做Pipy並不是爲了替代Nginx，這裏面有兩層含義，第一是說Nginx本身作爲一個非常優秀的堆棧的Web Server，包含反向代理，負載均衡等經典功能，我們沒有必要重複造輪子去替代他，無論從市場角度還是從技術角度來講，這都沒有必要。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"第二，Pipy嘗試解決的一些問題我們稱之爲“流量編程”，實際上是Nginx甚至整個行業都還沒有解決的，從技術角度看，整個行業都還沒有一個明確的解決方案，市場也是一片空白。那麼什麼是“流量編程”呢？對於有編程經驗，尤其是瞭解Java編程的人，我們可以認爲Pipy類似netty，是一個高性能、易擴展的框架，基於這個框架可以快速開發出高性能網絡服務程序，如web server。就像vertx基於netty，cassandra也基於netty一樣。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"除了像Netty封裝了底層網絡API操作外，pipy提供了一個簡化的JS runtime，開發者可以使用JS語法開發上層邏輯。作爲和netty對比，netty是Java編寫的，包含了GC，多線程等特徵；而pipy是單線程事件驅動的，同時pipy有自己的內存管理，基於C++的pipy，規避了類似Java和Go等編程語言常見的GC等“深坑”。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"回到nginx的話題，我們在開發過程中需要做功能和性能測試，利用HTTP堆棧的各種工具，測出的數據更有參考意義，更容易對比。舉個例子，採用Pipy跑1000萬請求測試的平均延遲是多少？很多基於HTTP的工具可以使用（wrk, ab, gatling等），也有很多可以對比參考的server實現（nginx, envoy, undertow等），因此我們選擇HTTP堆棧進行基準測試。但是實際上除了HTTP，pipy還支持Redis，Dubbo，socks等協議，並且還在繼續開發支持其他協議。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"再回到Pipy這個話題，Pipy實際上更像一個底層的發動機模版，在這個模版之上我們可以快速開發適配摩托車的發動機，適配汽車的發動機，甚至適配大卡車、越野車，以及適配輪船的發動機。所以從軟件定位上來說，Pipy其實比Nginx要更加底層，是一個全新的領域。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"主持人：您剛剛提到的流量編程概念很新穎，能詳細說說她的來龍去脈嗎？"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"**蔡書：**流量編程是一個非常有趣的話題，其目的是滿足流量處理的需求。早期流量處理需求是通過專用的網絡硬件設備來解決。而隨着雲計算的快速發展，很多硬件設施被軟件化，我們看到越來越多的底層資源比如網絡、存儲等被軟件定義，軟件定義的特徵之一就是“可編程”。有一個詞叫做InfrastructureAs Code，說的就是基礎設施可編程。相應地，我們的流量編程是這個細分領域裏的一個子集，爲的是更高效地解決流量處理的需求。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"以此爲基礎我們可以將\"HTTP流量編程\"這個模式泛化一下，比如我們是不是可以對MySQL的流量做編程，對一個kafka的請求做編程，對一個Redis的請求做編程，對Dubbo的請求做編程？目前這樣的需求很旺盛。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Nginx在最開始是一個Web Server，它預留了一些擴展接口藉口，可以擴展新的功能，對於HTTP流量編程，Nginx提供了成熟的模型；但是對於非HTTP流量，nginx預留的接口就很簡單了，並不適合擴展開發。一方面接口比較晦澀，不那麼容易使用；另一方面C語言擴展的編程難度還是挺大的，不利於普及。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"今天在Session裏我會分享一些Pipy的落地應用。舉個例子，我們攔截了Redis請求之後，他訪問的是哪一個Redis的實例，或者哪個集羣？這實際上可以由中間Proxy的環節來決定，需要有一些If else之類的條件進行分流。這個就是“redis流量可編程“的需求。我們把這種能力叫做流量編程，尤其泛指面向應用層的流量處理能力，包括應用程序本身，比如rest請求；當然，也有一部分面向中間件的。"}]},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"源起於定製化修改報文，Pipy要做流量通路上的智能交通系統"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"主持人：您開發Pipy的時候，看到了哪些當時還沒有被滿足的市場需求，或者說客戶的具體需求？"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"**蔡書：**這個話題非常好，回到本源來說，這個問題就是所謂的“Pipy從哪裏來”這樣的經典話題。實際上最開始的時候，我們在編程過程中發現客戶有一些需求具有行業共性。舉個例子，銀行客戶常用的定長報紋，需要固定前6個字節作爲報頭。醫療、電信和泛金融行業也希望類似信息能夠被路由，被取樣。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"一開始，我們去找相應工具比如openresty去解決這個問題，但發現它缺少一個抽象層。於是我們打算從下往上寫，自己寫一個socket程序，但後來也發現這樣開發的應用維護和擴展難度很大。最後我們嘗試在應用協議和底層socket之前加入一個抽象層，這個抽象層解決了底層編程的複雜度，同時又提高了現場開發的效率。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"那我們就發現存在這樣的一些行業性需求未被滿足，我們沿着這條路走下去不僅可以解決專屬的行業業務問題，還可以解決泛化的技術領域的問題，比如Redis的問題，比如Kafka的問題。於是我們不斷地迭代，到現在三年時間，Pipy已經越來越通用，能解決更多需求了。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"主持人：Pipy開發的整個藍圖是什麼樣的，現在進行到哪一步了？接下來有什麼樣的計劃呢？"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"**蔡書：**我們的藍圖可以類比交通系統。城市中的公路網絡一樣，有環路，有小衚衕，各處的交通設備可以管車流量，管traffic。對於我們來說，在一個企業內部網絡中，如果規則明確並且遍佈足夠多的proxy，就像城市裏每一個路口設置了足夠多的標識符，那麼交通流量就可以自動地做智能化調度，類似自動駕駛。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"繼續想象道路系統，首先道路上有一些可識別的東西，那麼無人駕駛系統就可以自動根據設定的路線行駛。對於我們來說，我們的第一步是鋪設所有控制點，這是類似紅綠燈的基礎設施。Pipy在整個藍圖中是一個開始，它只是一個小組件，負責爲流量導航。接下來我們會開發控制體系，也就是基於這些“標識”的“智能交通”體系。這樣企業內網的流量就可以高效、可控、安全、智能的自動流動，並且需要更少的人工干預。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"下一階段我們會佈置一些控制器（Controller），用來適配K8s，阿里雲，AWS等，甚至可以和傳統網絡設備耦合。最終構建出一個完整的可自動調度流量的控制體系。這是我們的整個藍圖。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"主持人：您覺得未來在流量管理這個領域中，行業集中度會如何發展？"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"**蔡書：**首先我們看到這種細粒度的智能流量管理是最近幾年才演化出來的，也就是說變革纔剛剛開始。很早之前大家靠手工去配置網絡，設定流量傳輸規則；2017年左右隨着微服務體系的普及，單一集羣節點變多，智能流量管理的需求越來越旺盛，於是很多團隊開始嘗試解決這類需求，典型的有Nginx，envoy，Linkerd等等。除此外還有客戶自己編寫工具，典型的有traefik，百度的BFE等。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"我們認爲這個領域在未來幾年會是百花齊放的階段。然後慢慢地，整個領域會收斂；由於產品性能，資源佔用率等原因，最後可能只剩2～3個成熟產品。那從終局來看，能同時做到高性能，低延遲，可擴展，可編程而且學習曲線容易被大衆接受，這件事並不容易。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"對於我們而言，我們希望在普及度，易用性方面多下功夫，希望這條路能一直走下去，到最後留下的成熟產品中有我們一個。"}]}]}