分佈式計算的8大謬誤

{"type":"doc","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在Ably博客最近的一篇文章中，Alex Diaconu回顧了分佈式計算的8大謬誤，並提供了一些應對建議。InfoQ藉此機會與Diaconu進行了交流，以瞭解更多關於Ably工程師如何處理這些謬誤的信息。"}]},{"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":"在Ably博客最近的一篇文章中，Alex Diaconu"},{"type":"link","attrs":{"href":"https:\/\/ably.com\/blog\/8-fallacies-of-distributed-computing","title":"","type":null},"content":[{"type":"text","text":"回顧"}]},{"type":"text","text":"了分佈式計算的8大謬誤，並提供了一些應對建議。InfoQ藉此機會與Diaconu進行了交流，以瞭解更多關於Ably工程師如何處理這些謬誤的信息。"}]},{"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":"這"},{"type":"link","attrs":{"href":"http:\/\/nighthacks.com\/jag\/res\/Fallacies.html","title":"","type":null},"content":[{"type":"text","text":"8大謬誤"}]},{"type":"text","text":"是關於分佈式計算的一組假設，這些假設可能會導致軟件開發的失敗：網絡是可靠的；延遲爲0；帶寬是無限的；網絡是安全的；拓撲結構是不變的；只有一名管理員；傳輸成本爲0；網絡是同構的。"}]},{"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":"這些謬誤可以視爲分佈式系統設計時需要考慮的架構需求。InfoQ與Diaconu進行了交流，以瞭解更多關於Ably工程師如何處理這些謬誤的信息。"}]},{"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":"InfoQ：關於分佈式計算的謬誤，自從最初提出以來，已經過去了近三十年，但現在它們仍然很有意義。在Ably，它們的作用是什麼？"}]},{"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","marks":[{"type":"strong"}],"text":"Alex Diaconu："},{"type":"text","text":" 所有這些謬誤指向的都是分佈式系統設計的陷阱，它們今天依然很有意義。它們的影響不盡相同，其中有一些更容易處理。對於Ably來說，下面這幾個謬誤的影響最爲普遍："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"網絡是可靠的。"},{"type":"text","text":" 自然，這是所有服務設計和運營都需要考慮的一個方面。不僅網絡本身不可靠，你試圖通過網絡訪問的系統也會失敗。網絡失敗也不是一個二元問題——網絡失敗會有意想不到的形式。預測節點或互連故障是系統設計固有的工作。在Ably的博客上，我們已經對我們的處理方法做了廣泛的探討，比如通過"},{"type":"link","attrs":{"href":"https:\/\/ably.com\/blog\/engineering-dependability-and-fault-tolerance-in-a-distributed-system","title":"","type":null},"content":[{"type":"text","text":"容錯機制"}]},{"type":"text","text":"。"}]},{"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":"拓撲結構是不變的。"},{"type":"text","text":" 這條謬誤對於我們的架構而言也很重要。Ably平臺被設計成具有實時彈性的平臺，因此，其拓撲結構會不斷變化。我們的系統必須把拓撲結構變化作爲一個常規問題來處理，而且，處理這個問題是系統複雜性的一個重要來源。其他系統服務使用發現層來構造系統拓撲的公共視圖。服務間請求路由全都是根據那個拓撲結構圖來完成的：發現層的可擴展性和性能是一項不小的工程挑戰，我們已經在擴展Ably服務的過程中解決了這個問題。"}]},{"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":"帶寬是無限的&傳輸成本爲0。"},{"type":"text","text":" 在實踐中，系統內的網絡成本——在跨多個地區的全球性系統中——在運營成本中佔了相當一部分。因此，我們要知道，在設計系統時，要保證我們的流量可以隨着用戶負載線性擴展，並且要監控流量，確保它在設計參數之內。我們偶爾也會遇到網絡狀況退化的問題，因此，我們需要通過監控來發現這種退化。"}]}]},{"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":"InfoQ：在分佈式系統過去30年的發展歷程中，是否出了其他需要注意的謬誤？"}]},{"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","marks":[{"type":"strong"}],"text":"Diaconu："},{"type":"text","text":" 我覺得，過去30年最大的變化是我們對如何處理這些謬誤的理解更成熟了。這不是說問題更簡單了，而是我們對它們有了更好的理解。我們知道什麼方法好，什麼方法不好，以及某種特定的方法有什麼侷限。關於這些問題，現在已經有了完善的科學理論和工程實踐。計算機科學專業的學生都會學習這些問題及其現狀。"}]},{"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":"當然，重要的是要知道，這些謬誤是長久存在的技術挑戰，而不要把它們想成是可以輕鬆躲開的陷阱。事實上，我覺得你可以說有這樣一種新謬誤——“避免分佈式計算的謬誤很簡單”。"}]}]},{"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":"InfoQ：有些謬誤已經是老生常談，例如，“雲不安全”的觀點已被廣泛接受。不過，這些謬誤中也有一些複雜難解之處，應對起來並不是那麼簡單。"}]},{"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","marks":[{"type":"strong"}],"text":"Diaconu："},{"type":"text","text":" 就像前面提到的那樣，分佈式系統的挑戰，以及與分佈式系統構建技術和機制相關的廣泛的科學領域，已經得到了很好的研究。然而，當你在現實世界中應對這些挑戰時就會發現，學術理解只能帶你走這麼遠。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"構建分佈式系統要遵循工程實用主義進行權衡，最好的解決方案來自於經驗以及試驗。例如，“網絡是可靠的”這一謬誤是你要解決的最基本的問題。已知的解決方案有具有重試機制的協議、共識形成協議或冗餘容錯，使用哪種方案，取決於你所關注的特定的故障模式。不過，工程上的現實情況是，多重故障可能會同時發生。這樣，理想的解決方案應該基於故障分佈分析或是故障預算分析，以及某些故障對具體業務的影響。恢復機制本身可能會因爲系統不可靠而失敗，而此類失敗發生的概率可能會影響解決方案。當然，也有複雜性風險：解決方案理論上很可靠，但也很複雜，與理論上不那麼完善但更簡單的機制相比，當事件發生時，處理或理解的難度都要大很多。"}]}]},{"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":"InfoQ：過去幾年，微服務已經變得非常流行，它們似乎並不把“傳輸成本爲0”當成謬誤。事實上，微服務越小，其數量和隨之而來的傳輸成本就越大。關於這一點，您怎麼看？"}]},{"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","marks":[{"type":"strong"}],"text":"Diaconu："},{"type":"text","text":" 也許這是另一個謬誤“微服務簡化了系統推理”。有時候，將事情分解成表面積較小的組件可以簡化推理。但是，有時候，這些邊界會增加複雜性，會增加故障模式，產生一些新的需要推斷的行爲。"}]},{"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":"和之前的回答一樣，具體的設計選擇，以及何時何地採用已知的理論解，取決於工程判斷和經驗。在Ably，我們運營的系統有多個角色，這些角色可以獨立擴展、相互操作及相互發現。然而，我們很少將功能劃分成不同的角色，而且只在有特定需求時纔會這樣做。例如，如果我們希望某些特定的功能能夠獨立於其他功能進行擴展，這可以證明單獨創建一個角色的合理性，即使它帶來了額外的複雜性。"}]}]},{"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":"Diaconu的文章不僅幫助我們瞭解了謬誤的來源，而且還就當前解決謬誤的技術和方法提供了一些實用的線索，所以如果你對這個問題感興趣，不妨讀一下。"}]},{"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":"查看英文原文："},{"type":"link","attrs":{"href":"https:\/\/www.infoq.com\/news\/2021\/09\/fallacies-distributed-computing\/","title":"","type":null},"content":[{"type":"text","text":"Reviewing the Eight Fallacies of Distributed Computing"}]}]}]}