專訪科大訊飛吳義平:雲原生趨勢下,如何將“降本”做到極致

{"type":"doc","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"控制成本是企業遷移上雲的主要目標之一,而企業上雲後主要成本就變成了運營上的支出,所以技術人員就需要不斷進行優化,提高資源利用率,從而進一步達到降本的目的。作爲國內AI龍頭企業,科大訊飛在雲原生應用架構轉型的實踐中,實現了高達50%的降本成果。InfoQ採訪了科大訊飛資深架構師吳義平,他講述了一些技術演進中的心得體會。"}]},{"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":"吳義平,科大訊飛資深架構師"},{"type":"text","text":"。2014 年加入科大訊飛,任職雲計算資深架構師,目前負責訊飛 AI Cloud 基礎架構演進工作,主導設計了 AI Cloud 2.0 通用計算平臺、彈性計算平臺等重點項目,個人擅長通用網關、服務治理、彈性計算等方向。他將在"},{"type":"link","attrs":{"href":"https:\/\/qcon.infoq.cn\/2020\/shenzhen\/schedule?utm_source=infoq&utm_medium=arti&utm_campaign=full","title":"xxx","type":null},"content":[{"type":"text","text":"QCon全球軟件開發大會(深圳站)"}]},{"type":"text","text":"上進行演講,主題爲《"},{"type":"link","attrs":{"href":"https:\/\/qcon.infoq.cn\/2020\/shenzhen\/presentation\/2823?utm_source=infoq&utm_medium=arti&utm_campaign=full","title":"","type":null},"content":[{"type":"text","text":"降本 50%——科大訊飛雲原生應用架構轉型實踐"}]},{"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":"InfoQ:有一千個讀者心中有一千個哈姆雷特,對於雲原生定義也有人用這個話來打比喻。那麼在你看來,雲原生該如何定義?"}]},{"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":"吳義平:"},{"type":"text","text":"在我看來,雲原生是一種架構設計範式,是一套面向雲計算領域的技術體系和方法論,擁有開放、可伸縮、低成本等特性。全面應用雲原生架構後,可以實現更少的人力投入、更高的資源利用率,它爲研發人員和企業帶來的直接受益是降本增效。相較於傳統IT架構,雲原生應用支持快速迭代、輕鬆複製、靈活部署,真正實現“Build once, deploy any time and any cloud”。"}]}]},{"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:科大訊飛是什麼時候開始採用雲原生架構的?當時主要想解決的痛點是什麼?(演進過程是什麼樣的?)"}]},{"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":"吳義平:"},{"type":"text","text":"科大訊飛"},{"type":"text","text":"大約在"},{"type":"text","text":"2015年時開始接觸雲原生架構,曾經與Pivotal公司有過深入的技術交流,那時候"},{"type":"text","text":"雲原生"},{"type":"text","text":"技術體系還沒有今天這麼開放。"}]},{"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":"2010年,科大訊飛正式對外發布了語音雲,隨着AI技術和行業的發展,能力上雲需求和開發者、終端用戶均出現爆發式增長。"},{"type":"text","text":"爲了滿足快速發展的節奏,我們不斷在雲端架構中添加新的內容,"},{"type":"text","text":"在經過6年的發展與沉澱後,架構變得非常臃腫。由於缺乏全鏈路的問題分析工具鏈,監控告警延遲高,有時候一個環節出現問題後,會因爲問題解決時間太長而演變成故障"},{"type":"text","text":"。"},{"type":"text","text":"2016年,我們花了大量的時間去處理現網故障和查缺補漏。同時,隨着大量的需求輸入,我們投入到業務開發的人力和服務器數量都出現了較大增長,成本隨之上升,業務方紛紛提出了穩定性和降本訴求。"}]},{"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":"2017年"},{"type":"text","text":"應該算是訊飛AI雲服務的新元年,也是我們全面擁抱雲原生技術的時刻,在研究院副院長龍明康的帶領下,我們完成了基於雲原生技術體系的AICloud2.0架構的設計、研發和上線,在穩定性和降本增效方面均取得了關鍵性成果。"}]},{"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":"相較於架構演進,我們最開始的變化是將技術棧從C++切換到Golang,因爲Go語言在開發效率和服務性能方面的表現更加均衡。"}]},{"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":"提升集羣的穩定性是當時最急迫需要解決的問題,我們選擇基於"},{"type":"text","text":"g"},{"type":"text","text":"RPC"},{"type":"text","text":"開發了一套Go語言版本的高性能微服務框架,從基礎IO框架的性能優化做起,先後開發了配置中心&服務發現、負載均衡、熔斷&降級、流控、全鏈路日誌追蹤技術、運行時指標收集等服務治理特性,基於這套服務框架開發的應用可以獲得一致的性能和服務治理體驗。同時,"},{"type":"text","text":"爲了避免在業務定製需求上投入越來越多的人力,我們設計了一套面向AI領域的通用業務模型,上線新的AI能力通過少量的配置工作即可完成。"}]},{"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":"爲了更快的發現生產環境的問題,我們通過prometheus"},{"type":"text","text":" "},{"type":"text","text":"+"},{"type":"text","text":" "},{"type":"text","text":"zabbix構建了新的監控告警系統,並結合通過自研微服務框架構建的應用實時產生的運行時、請求數據,實現了立體化監控,大幅提升了我們的問題排查和定位效率。但保障現網穩定只靠架構演進是遠遠不夠的,我們借鑑了Google SRE的經驗,分別從SRE團隊、制定SLA&SLO標準、應急預案、應急平臺、故障演練等方面構建了滿足訊飛現網保障需求的SRE體系,已成爲保障生產環境MTTR的不可或缺的一部分。"}]},{"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":"全面落地容器化技術在AICloud2.0架構演進中發揮了至關重要的作用,它解決了我們在異構環境中的環境、資源隔離問題,爲盤活計算資源、彈性調度打下了基礎。同時,容器技術也加速我們CI\/CD的效率,從開發到上線,人的參與度越來越低。"}]},{"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":"在完成新架構導流和持續演進後,我們已經實現全年無故障,部分AI業務的成本相比老架構最高降幅達到50%。"}]}]},{"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:應用雲原生架構時,很多企業都面臨的問題就是“越來越大,越來越慢”,那麼科大訊飛的架構是否也曾有過類似的問題?你們如何在複雜性和穩定性之間做到平衡?"}]},{"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":"吳義平:"},{"type":"text","text":"在做大的技術變革時大家應該都有遇到過"},{"type":"text","text":"“越來越大,越來越慢”的問題,訊飛也不例外,它可能是一個過渡狀態,但也有可能演變成現狀。"}]},{"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":"在我們這邊,“越來越大”主要表現在團隊的規模和系統的架構上,架構演進是個持續的過程,同時兼顧新老架構勢必會帶來人力投入的增長,在這個過程中,我們投入了一小部分人專職做新架構演進,大部分人繼續維持老架構的正常運轉和需求迭代,團隊人員數量在短期內出現了較小規模增長。而在應用架構上,由於當時的訊飛AICloud1.0架構已經迭代了7年,業務邏輯非常臃腫,組件繁多,爲了避免新架構集羣變得更加複雜龐大,我們"},{"type":"text","text":"通過抽象AI領域模型設計出了一套面向AI領域的標準化協議,實現了流量的高度內聚,集羣的複雜度和組件數量得到收斂。"}]},{"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":"作爲國內第一家提供AI雲服務的企業,科大訊飛公有云服務的每日PV高達千億,訊飛開放平臺擁有110w開發者,應用數量數百萬,存量業務成爲了架構演進時“越來越慢”的誘因,同時,業務需求並沒有因爲架構演進而減少"},{"type":"text","text":"。那麼問題就成了是"},{"type":"text","text":"繼續在老架構增量支持,還是待新架構全部上線再響應?"}]},{"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":"因爲穩定性是平臺的立足之本,所以導流比研發新架構更具挑戰。爲了避免或降低遷移對用戶造成的影響,我們採取頭部用戶單獨制定導流方案、長尾用戶分批導流的策略循序漸進導流,精準的流量調度和果斷的決策是決定大規模流量遷移成敗的重要因素。而在架構演進過程中出現的增量需求,我們會按照全新業務基於新架構支撐、存量業務先分級再選型的邏輯去響應,在保證對業務發展不造成明顯影響的前提下逐漸下新架構遷移需求。"}]}]},{"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","marks":[{"type":"strong"}],"text":"InfoQ:有沒有一些判斷標準,來評斷一個企業到底該不該使用雲原生架構?"}]},{"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":"吳義平:"},{"type":"text","marks":[{"type":"strong"}],"text":"我認爲評價一個企業是否滿足使用雲原生架構的條件應該滿足以下三點"},{"type":"text","text":":"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":1,"normalizeStart":1},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"你"},{"type":"text","text":"的"},{"type":"text","text":"系統剛起步,沒有歷史包袱,且團隊有較強"},{"type":"text","text":"的"},{"type":"text","text":"技術能力"},{"type":"text","text":";"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"你的系統趨於穩定,但"},{"type":"text","text":"是"},{"type":"text","text":"還需要迎接未來更大的流量挑戰,或者更快的"},{"type":"text","text":"迭代"},{"type":"text","text":"速度"},{"type":"text","text":";"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":3,"align":null,"origin":null},"content":[{"type":"text","text":"你的系統當前已經不堪重負,轉型雲原生架構已經迫在眉睫;"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":4,"align":null,"origin":null},"content":[{"type":"text","text":"企業能夠接受的投入產出比"},{"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","text":"如果滿足以上條件並且準備實踐雲原生架構時,建議儘量使用公有云、專有云等現成平臺,或者使用社區裏的開源項目構建,一來可以節約成本,二來可以獲得外部技術支持。"}]}]},{"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:應用雲原生架構時,主要成本耗費主要哪些方面導致的?"}]},{"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":"吳義平:"},{"type":"text","text":"應用雲原生架構時的主要成本包括人力和計算成本兩部分,人力成本包括框架研發、應用改造以及流量遷移;而計算成本除了新架構所需基本計算資源外,還包括流量遷移時新老架構共存期間過渡集羣的計算資源。"}]},{"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":"相較於傳統服務架構,應用雲原生架構可以提供更加敏捷的迭代流程、更加靈活的部署方案和更高的資源利用率,這些"},{"type":"text","text":"特性"},{"type":"text","text":"相互疊加後,"},{"type":"text","text":"可以爲企業帶來明顯的降本增效"},{"type":"text","text":"作用。當然,降本增效的前提是企業的IT服務需求和服務量均達到一定規模,足以產生經濟效益。"}]}]},{"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:雲廠商是定價方,可以提供競價實例的方式降低使用方成本,但作爲使用方,你們主要有哪些手段可以用來降低成本?"}]},{"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":"吳義平:"},{"type":"text","text":"訊飛通過技術降本的手段與業內大的雲廠商的方法是基本一致的,其中最直接有效的方法是通過提升軟件的性能來達到節省資源的目的,在業務規模足夠大時,優化性能的收益非常明顯。"}]},{"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":"在訊飛AI業務場景下,我們主要從以下幾個方面進行了實踐:"}]},{"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":"1、高性能計算,我們基於gRPC構建的Go語言版本服務框架中,通過優化PB協議、鏈接池、buffer等細節,"},{"type":"text","text":"性能"},{"type":"text","text":"相對原生"},{"type":"text","text":"gRPC"},{"type":"text","text":"的提升了30%。而在APM日誌採集組件上,我們也進行了深度優化,業內開源APM方案的日誌可用採集性能大概在單機1000QPS左右,我們目前已經做到20w QPS;"}]},{"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":"2、負載均衡調度,面向AI領域的很多計算引擎都需要支持流式處理特性,單次會話持續時間長,這些引擎基本都是部署在GPU服務器之上,負載均衡的調度效率直接決定了這些GPU服務器的有效算力,我們自研的負載均衡調度算法調度效率做到了99.9%,GPU高峯時刻利用率達到90%;"}]},{"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":"3、AI引擎算法優化,持續演進神經網絡模型和並行AI計算框架,部分AI引擎性能獲得數倍提升;"}]},{"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":"4、彈性計算,我們除了有大量的實時計算外,還會存在不小規模的準實時、非實時、離線等對實效性要求低一些的計算需求。在傳統架構下,我們規劃集羣計算資源時需要按照業務的峯值評估配額,造成全局計算水位偏低。爲了突破這種侷限,我們結合公司內外計算需求,研發出了基於雲原生基礎設施的彈性計算平臺,通過配額管理、全局負載管理、彈性調度管理實現了可控的自動化資源調度,大幅提升了全局資源計算水位。 "}]}]},{"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:你怎麼看待雲原生基礎設施的未來發展趨勢?"}]},{"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":"吳義平:"},{"type":"text","text":"雲原生基礎設施的發展離不開需求的驅動和新基建的持續推進,越來越多的開發者在今後將只關注應用層的開發,雲廠商或企業勢必會緊貼市場需求,將企業上雲所依賴的基礎設施做深做厚。"}]},{"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":"一些企業可能會因爲面臨合規性、延遲等方面的需要,又或者是因爲數據安全方面的考慮,業務常常會部署在不同的公有云之上,甚至一部分部署在公有云,一部分部署在私有云,企業混合雲將成爲常態"},{"type":"text","text":",但從目前的實際需求和現狀來看,混合雲的實施過程並沒有想象中那麼順利,企業和雲廠商在基礎設施的兼容性方面還存在較多問題待解決,"},{"type":"text","text":"這也將倒逼雲廠商或企業提供更加標準化基礎設施服務,幫助更多企業通過應用雲原生技術實現多雲無縫切換。"}]},{"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":"伴隨着5G技術的逐漸落地,企業對低延遲、低帶寬的訴求將"},{"type":"text","text":"得以"},{"type":"text","text":"釋放,在未來將會有越來越多的業務場景可以在不通過雲端即可完成計算,在提升用戶體驗的同時也會幫助企業降低運營成本,邊緣計算逐漸會演變成企業混合雲的重要組成部分,企業或雲廠商需要具備將雲上基礎設施延伸到邊緣機房甚至智能設備上的能力,助力業務實現雲邊端一體化方案。"}]}]},{"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:你認爲雲原生架構師需要什麼樣的能力?在雲原生實踐中有哪些重要的原則?"}]},{"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":"吳義平:"},{"type":"text","text":"雲原生架構師除了要具備傳統應用架構師的素養外,還應該掌握虛擬化、網絡、存儲以及網絡安全方面等相關技術,當然,擁抱開源也是必備的素養之一。"}]},{"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":"雲原生應用的12要素已經比較全面的總結了雲原生實踐中應當遵循的重要原則,我從中總結幾條最常用的內容"},{"type":"text","text":":"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":1,"normalizeStart":1},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"堅持"},{"type":"text","text":"不可變基礎設施,保證環境統一"},{"type":"text","text":",無論是應用的版本,還是應用版本所匹配的配置,都應遵循一次構建,多地部署的原則。我們在實踐過程中要求配置與鏡像版本的一致性,而這種約束被固化到了我們的配置中心版本管理中,有效避免了後期因爲轉移部署或回滾時因人爲選擇版本而可能導致的失誤發生。"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"應用需要去操作系統環境依賴並儘量"},{"type":"text","text":"做到"},{"type":"text","text":"無狀態"},{"type":"text","text":"設計"},{"type":"text","text":","},{"type":"text","text":"在拓展了可移植性的同時保證了"},{"type":"text","text":"自由伸縮和調度;"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":3,"align":null,"origin":null},"content":[{"type":"text","text":"應用之間的通信建議"},{"type":"text","text":"使用標準化I\/O接口和協議,"},{"type":"text","text":"例如Http、RPC等,私有化通信協議將會制約應用的兼容性"},{"type":"text","text":";"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":4,"align":null,"origin":null},"content":[{"type":"text","text":"支持優雅啓停特性,雲計算場景下,服務的可用性不強依賴於單個應用實例的絕對穩定性,擴縮容、滾動升級、異常宕機等可能是家常便飯,應用在設計時應當考慮支持優雅上、下線,確保變更時不會帶來可用性抖動,而針對異常宕機場景,則需要在各類負載均衡方案中標配心跳檢查、熔斷等機制。當這些可能會影響到服務穩定性的細枝末節被一一清除,服務的全局穩定性方能獲得保障。"}]}]}]}]},{"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":"2020年12月6-7日,在QCon全球軟件開發大會(深圳站)“雲原生下的應用架構“專題上,吳老師將從訊飛依託雲原生技術構建的全新 AI 2.0 公有云架構切入,向大家講解我們通過技術演進實現降本 50% 的技術路徑和經驗,包括應用 Serverless、彈性計算和運維架構演進的落地案例。此外還有PingCAP、阿里雲、虎牙直播的專家們帶來各自團隊的技術實踐分享。"}]},{"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":"大會日程已上線!感興趣的朋友圈可"},{"type":"link","attrs":{"href":"https:\/\/qcon.infoq.cn\/2020\/shenzhen\/schedule?utm_source=infoq&utm_medium=arti&utm_campaign=full","title":"","type":null},"content":[{"type":"text","text":"點擊鏈接"}]},{"type":"text","text":"查看。"}]},{"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":"會議諮詢:17310043226(同微信)"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}}]}
發表評論
所有評論
還沒有人評論,想成為第一個評論的人麼? 請在上方評論欄輸入並且點擊發布.
相關文章