{"id":36484,"date":"2019-10-31T22:11:54","date_gmt":"2019-10-31T19:11:54","guid":{"rendered":"https:\/\/prohoster.info\/blog\/kubernetes-priklyuchenie-dailymotion-sozdanie-infrastruktury-v-oblakah-on-premises\/"},"modified":"2019-10-31T22:11:54","modified_gmt":"2019-10-31T19:11:54","slug":"kubernetes-priklyuchenie-dailymotion-sozdanie-infrastruktury-v-oblakah-on-premises","status":"publish","type":"post","link":"https:\/\/prohoster.info\/sq\/blog\/administrirovanie\/kubernetes-priklyuchenie-dailymotion-sozdanie-infrastruktury-v-oblakah-on-premises","title":{"rendered":"Aventura Kubernetes e Dailymotion: krijimi i infrastruktur\u00ebs n\u00eb cloud + on-premises","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"

\"Aventura
\n
\nSh\u00ebn. p\u00ebrk.<\/b>: Dailymotion \u2014 nj\u00eb nga sh\u00ebrbimet m\u00eb t\u00eb m\u00ebdha n\u00eb bot\u00eb p\u00ebr hostimin e videove dhe prandaj nj\u00eb p\u00ebrdorues i njohur i Kubernetes. N\u00eb k\u00ebt\u00eb material, arkitekti i sistemeve David Donchez ndan rezultatet e krijimit t\u00eb platform\u00ebs prodhim t\u00eb kompanis\u00eb bazuar n\u00eb K8s, e cila filloi me nj\u00eb instalim n\u00eb cloud n\u00eb GKE dhe p\u00ebrfundoi si nj\u00eb zgjidhje hibride, e cila lejoi arritjen e koh\u00ebzgjatjes m\u00eb t\u00eb mir\u00eb dhe kursimin e kostove t\u00eb infrastruktur\u00ebs.<\/i><\/p>\n

Kur mor\u00ebm vendimin p\u00ebr t\u00eb ristrukturuar API-n\u00eb kryesore Dailymotion<\/a><\/noindex> tre vjet m\u00eb par\u00eb, ne donim t\u00eb zhvillonim nj\u00eb m\u00ebnyr\u00eb m\u00eb efikase p\u00ebr t\u00eb vendosur aplikacione dhe p\u00ebr t\u00eb leht\u00ebsuar proceset n\u00eb zhvillim dhe prodhim<\/a><\/noindex>. P\u00ebr k\u00ebt\u00eb q\u00ebllim, ne vendos\u00ebm t\u00eb p\u00ebrdorim platform\u00ebn e orkestrimit t\u00eb kontejner\u00ebve dhe natyrsh\u00ebm zgjedh\u00ebm Kubernetes.<\/p>\n

Pse ia vlen t\u00eb krijosh nj\u00eb platform\u00eb t\u00eb vet\u00ebn mbi Kubernetes?<\/a><\/noindex><\/p>\n

API n\u00eb nivel prodhimi n\u00eb nj\u00eb koh\u00eb t\u00eb shkurt\u00ebr me ndihm\u00ebn e Google Cloud<\/h2>\n

<\/p>\n

Vera e vitit 2016<\/p><\/blockquote>\n

\nTre vjet m\u00eb par\u00eb, menj\u00ebher\u00eb pas blerjes s\u00eb Dailymotion nga Vivendi<\/a><\/noindex>, ekipet tona inxhinierike u p\u00ebrqendruan n\u00eb nj\u00eb q\u00ebllim global: t\u00eb krijojm\u00eb nj\u00eb produkt krejt t\u00eb ri Dailymotion.<\/p>\n

Pas pas analiz\u00ebs s\u00eb kontejner\u00ebve, zgjidhjeve p\u00ebr orkestrimin dhe p\u00ebrvoj\u00ebs ton\u00eb t\u00eb kaluar, arrit\u00ebm n\u00eb p\u00ebrfundimin se Kubernetes \u00ebsht\u00eb zgjedhja e duhur. Pjesa e zhvilluesve tashm\u00eb kishte njohuri mbi konceptet baz\u00eb dhe dinin si ta p\u00ebrdorin, gj\u00eb q\u00eb ishte nj\u00eb avantazh i madh p\u00ebr transformimin e infrastruktur\u00ebs.<\/p>\n

Nga perspektiva e infrastruktur\u00ebs, na duhej nj\u00eb sistem i fuqish\u00ebm dhe fleksib\u00ebl p\u00ebr t\u00eb akomoduar lloje t\u00eb reja aplikacionesh cloud-native. Ne preferuam t\u00eb q\u00ebndronim n\u00eb cloud n\u00eb fillim t\u00eb udh\u00ebtimit ton\u00eb, p\u00ebr t\u00eb nd\u00ebrtuar pa stress nj\u00eb platform\u00eb lokale sa m\u00eb t\u00eb besueshme. Aplikacionet ton\u00eb i zgjodh\u00ebm t'i vendosnim p\u00ebrmes Google Kubernetes Engine, megjith\u00ebse e dinim se her\u00ebt a von\u00eb do t\u00eb kalonim n\u00eb Qendrat tona t\u00eb t\u00eb Dh\u00ebnave dhe do t\u00eb aplikonim nj\u00eb strategji hibrid.<\/p>\n

P\u00ebrse zgjodh\u00ebm GKE?<\/h3>\n

\nKy vendim u mor kryesisht p\u00ebr arsye teknike. P\u00ebr m\u00eb tep\u00ebr, ishte e nevojshme t\u00eb ofronim shpejt infrastruktur\u00eb q\u00eb i p\u00ebrgjigjej nevojave t\u00eb biznesit. Kishim disa k\u00ebrkesa p\u00ebr vendosjen e aplikacioneve, si\u00e7 jan\u00eb shp\u00ebrndarja gjeografike, shkall\u00ebzueshm\u00ebria dhe q\u00ebndrueshm\u00ebria ndaj d\u00ebshtimeve.<\/p>\n

\"Aventura
\nKlaster\u00ebt GKE n\u00eb Dailymotion<\/i><\/p>\n

Dailymotion \u00ebsht\u00eb nj\u00eb platform\u00eb video globale, dhe ne doja t\u00eb p\u00ebrmir\u00ebsonim cil\u00ebsin\u00eb e sh\u00ebrbimit duke reduktuar koh\u00ebn e pritjes (latenc\u00eb)<\/i>. M\u00eb par\u00eb API yn\u00eb<\/a><\/noindex> ishte n\u00eb dispozicion vet\u00ebm n\u00eb Paris, q\u00eb ishte e pap\u00ebrshtatshme. Donim t\u00eb kishim mund\u00ebsin\u00eb p\u00ebr t\u00eb vendosur aplikacionet jo vet\u00ebm n\u00eb Evrop\u00eb, por edhe n\u00eb Azi dhe SHBA.<\/p>\n

Kjo ndjeshm\u00ebri ndaj vonesave n\u00ebnkuptonte se do t\u00eb duhej t\u00eb punonim shum\u00eb n\u00eb arkitektur\u00ebn rrjetit t\u00eb platform\u00ebs. Nd\u00ebrsa shumica e sh\u00ebrbimeve cloud k\u00ebrkonin krijimin e rrjetit t\u00eb tyre n\u00eb \u00e7do rajon dhe pastaj t'i lidhin ato p\u00ebrmes VPN-s\u00eb ose ndonj\u00eb sh\u00ebrbimi t\u00eb menaxhuar, Google Cloud lejonte t\u00eb krijohej nj\u00eb rrjet t\u00ebr\u00ebsor i menaxhuesh\u00ebm q\u00eb mbulonte t\u00eb gjitha rajonet e Google. Kjo \u00ebsht\u00eb nj\u00eb avantazh i madh p\u00ebr operimin dhe efikasitetin e sistemit.<\/p>\n

P\u00ebr m\u00eb tep\u00ebr, sh\u00ebrbimet rrjet dhe balancuesit e ngarkes\u00ebs nga Google Cloud p\u00ebrmir\u00ebsojn\u00eb jasht\u00ebzakonisht mir\u00eb pun\u00ebn e tyre. Ato thjesht lejojn\u00eb p\u00ebrdorimin e IP-ve publike nga \u00e7do rajon, dhe protokolli i mrekulluesh\u00ebm BGP do t\u00eb kujdeset p\u00ebr gjith\u00e7ka tjet\u00ebr (dometh\u00ebn\u00eb, do t'i drejtoj\u00eb p\u00ebrdoruesit n\u00eb grupin m\u00eb t\u00eb af\u00ebrt). \u00cbsht\u00eb e qart\u00eb se n\u00eb rast t\u00eb nj\u00eb d\u00ebshtimi, trafiku do t\u00eb kaloj\u00eb automatikisht n\u00eb nj\u00eb rajon tjet\u00ebr pa ndonj\u00eb nd\u00ebrhyrje t\u00eb njeriut.<\/p>\n

\"Aventura
\nMonitorimi i balancimit t\u00eb ngarkesave n\u00eb Google<\/i><\/p>\n

Platforma jon\u00eb gjithashtu p\u00ebrfshin aktive grafik\u00ebt e procesor\u00ebve. Google Cloud lejon p\u00ebrdorimin e tyre mjaft efikas direkt n\u00eb klasteret Kubernetes.<\/p>\n

N\u00eb t\u00eb nj\u00ebjt\u00ebn koh\u00eb, ekipi yn\u00eb i infrastruktur\u00ebs u p\u00ebrqendrua kryesisht n\u00eb stakun e vjet\u00ebr, t\u00eb vendosur n\u00eb serverat fizik\u00eb. Kjo \u00ebsht\u00eb arsyeja pse p\u00ebrdorimi i sh\u00ebrbimeve t\u00eb menaxhuara (p\u00ebrfshir\u00eb komponent\u00ebt kryesor\u00eb t\u00eb Kubernetes) p\u00ebrmbushi k\u00ebrkesat tona dhe ua mund\u00ebsoi ekipeve t\u00eb m\u00ebsojn\u00eb se si t\u00eb punojn\u00eb me klasteret lokale.<\/p>\n

Si rezultat, ne arrit\u00ebm t\u00eb fillojm\u00eb t\u00eb pranojm\u00eb trafik production n\u00eb infrastruktur\u00ebn Google Cloud vet\u00ebm 6 muaj pas fillimit t\u00eb pun\u00ebs.<\/p>\n

Megjithat\u00eb, pavar\u00ebsisht nj\u00eb s\u00ebr\u00eb p\u00ebrfitimesh, puna me ofruesin e sh\u00ebrbimeve n\u00eb cloud ka kosto t\u00eb caktuara, t\u00eb cilat mund t\u00eb rriten n\u00eb var\u00ebsi t\u00eb ngarkes\u00ebs. Kjo \u00ebsht\u00eb arsyeja pse ne e analizuam me kujdes \u00e7do sh\u00ebrbim t\u00eb menaxhuar q\u00eb p\u00ebrdorim, me synimin p\u00ebr ta implementuar at\u00eb vet\u00eb n\u00eb t\u00eb ardhmen. N\u00eb t\u00eb v\u00ebrtet\u00eb, implementimi i klaster\u00ebve lokal\u00eb filloi n\u00eb fund t\u00eb vitit 2016 dhe po at\u00ebher\u00eb u nis strategjia hibride.<\/p>\n

Lan\u00e7imi i platform\u00ebs lokale t\u00eb orkestrimit t\u00eb kontejner\u00ebve Dailymotion<\/h2>\n

<\/p>\n

Autumn 2016<\/p><\/blockquote>\n

\nN\u00eb kushtet kur e gjith\u00eb staku ishte gati p\u00ebr prodhim, dhe puna mbi API vazhdoi<\/a><\/noindex>, ishte koha p\u00ebr t'u p\u00ebrqendruar n\u00eb klaster\u00ebt rajonal\u00eb.<\/p>\n

N\u00eb at\u00eb koh\u00eb, p\u00ebrdoruesit shihnin m\u00eb shum\u00eb se 3 miliard video \u00e7do muaj. Natyrisht, ne kishim nj\u00eb rrjet t\u00eb gjer\u00eb t\u00eb Content Delivery Network q\u00eb funksiononte prej vitesh. Ne donim t\u00eb p\u00ebrfitojm\u00eb nga kjo mund\u00ebsi dhe t\u00eb vendosnim klaster\u00eb Kubernetes n\u00eb qendrat ekzistuese t\u00eb t\u00eb dh\u00ebnave.<\/p>\n

Infrastruktur\u00eb Dailymotion ka m\u00eb shum\u00eb se 2,5 mij\u00eb servera n\u00eb gjasht\u00eb qendra t\u00eb t\u00eb dh\u00ebnave. T\u00eb gjith\u00eb ata konfigurohen me Saltstack. Ne filluam t\u00eb p\u00ebrgatisim t\u00eb gjitha recetat e nevojshme p\u00ebr krijimin e master- dhe worker-nod\u00ebve, si dhe t\u00eb klasterit etcd.<\/p>\n

\"Aventura
\n<\/p>\n

Pjesa rrjet\u00eb<\/h3>\n

\nRrjeti yn\u00eb \u00ebsht\u00eb plot\u00ebsisht i \u043c\u0430\u0440\u0448\u0440\u0443tuesh\u00ebm. \u00c7do server shpall IP-n\u00eb e tij n\u00eb rrjet me ndihm\u00ebn e Exabgp. Ne krahasuam disa shtesa t\u00eb rrjetit dhe i vetmi q\u00eb p\u00ebrmbushte t\u00eb gjitha nevojat (p\u00ebr shkak t\u00eb qasjes s\u00eb p\u00ebrdorur n\u00eb nivelin L3) ishte Calico<\/a><\/noindex>. Ai u p\u00ebrshtat mrekullisht n\u00eb modelin ekzistues t\u00eb rrjetit t\u00eb infrastruktur\u00ebs.<\/p>\n

Duke d\u00ebshironim t\u00eb shfryt\u00ebzonim t\u00eb gjitha elementet e infrastruktur\u00ebs, s\u00eb pari duhej t\u00eb kuptonim mjetin ton\u00eb t\u00eb rrjetit (t\u00eb p\u00ebrdorur n\u00eb t\u00eb gjitha server\u00ebt): ta p\u00ebrdornim p\u00ebr t\u00eb njoftuar diapazonet e adreseve IP n\u00eb rrjetin me nyjet Kubernetes. Lejuam q\u00eb Calico t\u00eb caktoj\u00eb adresat IP p\u00ebr pod’at, por nuk e p\u00ebrdorim ende p\u00ebr sesionet BGP n\u00eb pajisjet rrjet\u00ebsore. N\u00eb fakt, ruterimi b\u00ebhet nga Exabgp, i cili njofton subnetet q\u00eb p\u00ebrdoren nga Calico. Kjo na lejon t\u00eb arrijm\u00eb \u00e7do pod nga rrjeti i brendsh\u00ebm (dhe ve\u00e7an\u00ebrisht nga balancuesit e ngarkes\u00ebs).<\/p>\n

Si e menaxhojm\u00eb trafikun ingress<\/h3>\n

\nP\u00ebr t\u00eb ridrejtuar k\u00ebrkesat e ardhshme n\u00eb sh\u00ebrbimin e duhur, vendos\u00ebm t\u00eb p\u00ebrdorim Ingress Controller p\u00ebr shkak t\u00eb integrimit t\u00eb tij me burimet ingress t\u00eb Kubernetes.<\/p>\n

Tre vjet m\u00eb par\u00eb, nginx-ingress-controller ishte kontrolluesi m\u00eb i pjekur: Nginx ishte p\u00ebrdorur prej koh\u00ebsh dhe ishte i njohur p\u00ebr stabilitetin dhe performanc\u00ebn e tij.<\/p>\n

N\u00eb sistemin ton\u00eb, vendos\u00ebm q\u00eb t\u00eb vendosnim kontrollor\u00ebt n\u00eb blade-server\u00eb t\u00eb dedikuar 10-gigabit. \u00c7do kontrollor lidhej me endpoint-in kube-apiserver t\u00eb klashtit p\u00ebrkat\u00ebs. N\u00eb k\u00ebto server\u00eb, gjithashtu p\u00ebrdorej Exabgp p\u00ebr njoftimin e adresave IP publike ose private. Topologjia e rrjetit ton\u00eb lejon t\u00eb p\u00ebrdoret BGP nga k\u00ebta kontrollor\u00eb p\u00ebr t\u00eb rritur t\u00eb gjith\u00eb trafikun direkt n\u00eb pod’at pa p\u00ebrdorur sh\u00ebrbime si NodePort. Ky qasje ndihmon n\u00eb shmangien e trafikut horizontal midis nyjeve dhe rrit efikasitetin.<\/p>\n

\"Aventura
\nTrafiku i internetit drejt pod’ave<\/i><\/p>\n

Tani q\u00eb jemi p\u00ebrgatitur me platform\u00ebn ton\u00eb hibride, mund t\u00eb thellohemi n\u00eb procesin e migrimit t\u00eb trafikut.<\/p>\n

Migrimi i trafikut nga Google Cloud n\u00eb infrastruktur\u00ebn e Dailymotion<\/h2>\n

<\/p>\n

Vjeshta 2018<\/p><\/blockquote>\n

\nPas pothuajse dy vitesh nd\u00ebrtimi, testimi dhe konfigurimi, ne p\u00ebrfunduam me nj\u00eb stak t\u00eb plot\u00eb Kubernetes, t\u00eb gatsh\u00ebm p\u00ebr t\u00eb pranuar nj\u00eb pjes\u00eb t\u00eb trafikut.<\/p>\n

\"Aventura
\n
\nStrategjia aktuale e marrjes s\u00eb vendimeve \u00ebsht\u00eb mjaft e thjesht\u00eb, por plot\u00ebson nevojat. P\u00ebrshtat\u00ebset IP publike (n\u00eb Google Cloud dhe Dailymotion) p\u00ebrdorin AWS Route 53 p\u00ebr t\u00eb vendosur politika dhe p\u00ebr t\u00eb drejtuar p\u00ebrdoruesit n\u00eb klasterin sipas zgjedhjes son\u00eb.<\/p>\n

\"Aventura
\nShembulli i politik\u00ebs s\u00eb routing-ut duke p\u00ebrdorur Route 53<\/i><\/p>\n

Me Google Cloud, kjo \u00ebsht\u00eb e thjesht\u00eb, pasi ne p\u00ebrdorim nj\u00eb IP t\u00eb vetme p\u00ebr t\u00eb gjith\u00eb klaster\u00ebt dhe p\u00ebrdoruesi drejtohet n\u00eb klasterin m\u00eb t\u00eb af\u00ebrt GKE. P\u00ebr klaster\u00ebt tan\u00eb, teknologjia \u00ebsht\u00eb ndryshe, pasi IP-t\u00eb e tyre ndryshojn\u00eb.<\/p>\n

Gjat\u00eb migrimit, ne synuam t\u00eb drejtonim k\u00ebrkesat rajonale n\u00eb klaster\u00ebt p\u00ebrkat\u00ebs dhe vler\u00ebsuam p\u00ebrfitimet e k\u00ebtij qasje.<\/p>\n

Sepse klaster\u00ebt tan\u00eb GKE jan\u00eb konfiguruar p\u00ebr shkall\u00ebzim automatik duke p\u00ebrdorur Metrika t\u00eb Personalizuara, ata rrisin\/reduktin kapacitetet n\u00eb var\u00ebsi t\u00eb trafikut q\u00eb hyn.<\/p>\n

N\u00eb kushtet normale, gjith\u00eb trafiku lokal drejtohet n\u00eb klashtin lokal, nd\u00ebrsa GKE sh\u00ebrben si rezerv\u00eb n\u00eb rast se ndodhin probleme (kontrolli i sh\u00ebndetit kryhet nga Route 53).<\/p>\n

\u2026<\/h2>\n

\nN\u00eb t\u00eb ardhmen, ne duam t\u00eb automatizojm\u00eb plot\u00ebsisht politikat e routing-ut p\u00ebr t\u00eb krijuar nj\u00eb strategji hibride autonome, e cila p\u00ebrmir\u00ebson vazhdimisht aksesin p\u00ebr p\u00ebrdoruesit. N\u00eb lidhje me p\u00ebrfitimet: shpenzimet p\u00ebr cloud jan\u00eb reduktuar ndjesh\u00ebm dhe gjithashtu kemi arritur t\u00eb zvog\u00eblojm\u00eb koh\u00ebn e reagimit t\u00eb API. Ne besojm\u00eb n\u00eb platform\u00ebn e cloud q\u00eb kemi krijuar dhe jemi t\u00eb gatsh\u00ebm t\u00eb drejtojm\u00eb m\u00eb shum\u00eb trafik n\u00eb t\u00eb n\u00ebse \u00ebsht\u00eb e nevojshme.<\/p>\n

P.S. nga p\u00ebrkthyesi<\/h2>\n

\nNdoshta do t'ju interesoj\u00eb nj\u00eb publikim tjet\u00ebr t\u00eb fundit nga Dailymotion n\u00eb lidhje me Kubernetes. Ai \u00ebsht\u00eb dedikuar deploy-it t\u00eb aplikacioneve me Helm n\u00eb shum\u00eb klastera Kubernetes dhe u publikua<\/a><\/noindex> rreth nj\u00eb muaji m\u00eb par\u00eb.<\/p>\n

Lexoni gjithashtu n\u00eb blogun ton\u00eb:<\/p>\n