{"id":36822,"date":"2019-10-31T22:14:01","date_gmt":"2019-10-31T19:14:01","guid":{"rendered":"https:\/\/prohoster.info\/blog\/kafka-na-kubernetes-eto-horosho\/"},"modified":"2019-10-31T22:14:01","modified_gmt":"2019-10-31T19:14:01","slug":"kafka-na-kubernetes-eto-horosho","status":"publish","type":"post","link":"https:\/\/prohoster.info\/sq\/blog\/administrirovanie\/kafka-na-kubernetes-eto-horosho","title":{"rendered":"Kafka n\u00eb Kubernetes \u2014 a \u00ebsht\u00eb e mir\u00eb?","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"

Mir\u00ebsevini, Habr!<\/p>\n

N\u00eb koh\u00ebn e saj, ne ishim t\u00eb par\u00ebt q\u00eb soll\u00ebm n\u00eb tregun rus tem\u00ebn Kafka<\/a><\/noindex> dhe vazhdojm\u00eb ndjek\u00eb<\/a><\/noindex> ta zhvillojm\u00eb. N\u00eb ve\u00e7anti, na duket interesante tema e bashk\u00ebpunimit mes Kafka dhe Kubernetes<\/a><\/noindex>. Nj\u00eb p\u00ebrmbledhje (dhe mjaft e kujdesshme) artikull<\/a><\/noindex> n\u00eb k\u00ebt\u00eb tem\u00eb u publikua n\u00eb blogun e kompanis\u00eb Confluent n\u00eb tetor t\u00eb vitit t\u00eb kaluar, nga autorja Gwen Shapiro. Sot, d\u00ebshirojm\u00eb t\u00eb t\u00ebrheqim v\u00ebmendjen tuaj n\u00eb nj\u00eb artikull m\u00eb t\u00eb ri, t\u00eb shkruar n\u00eb prill nga Johann Gyger, i cili, megjith\u00ebse nuk shmangu nj\u00eb pik\u00ebpyetje n\u00eb titull, e shqyrton tem\u00ebn n\u00eb nj\u00eb m\u00ebnyr\u00eb m\u00eb t\u00eb qart\u00eb, duke shoq\u00ebruar tekstin me lidhje interesante. Ju lutem, na falni p\u00ebr p\u00ebrkthimin e lirisht t\u00eb \"chaos monkey\", n\u00ebse e mundeni!<\/p>\n

\"Kafka<\/a><\/noindex>
\n<\/a><\/noindex> <\/p>\n

Hyrje<\/h4>\n

\nKubernetes \u00ebsht\u00eb i destinuar p\u00ebr t\u00eb punuar me ngarkesa q\u00eb nuk mbajn\u00eb gjendje. N\u00eb p\u00ebrgjith\u00ebsi, k\u00ebto ngarkesa paraqiten n\u00eb form\u00ebn e nj\u00eb arkitekture mikrosh\u00ebrbimesh, jan\u00eb t\u00eb lehta, mir\u00ebmbahen mir\u00eb p\u00ebr shkall\u00ebzimin horizontal, p\u00ebrputhen me parimet e aplikacioneve 12-faktor\u00ebshe, lejojn\u00eb t\u00eb punosh me qarkullues automatik\u00eb (circuit breaker) dhe majmun\u00ebt e chaos (chaos monkeys).<\/p>\n

Kafka, nga ana tjet\u00ebr, n\u00eb thelb vepron si nj\u00eb baz\u00eb t\u00eb dh\u00ebnash t\u00eb shp\u00ebrndara. K\u00ebshtu, gjat\u00eb pun\u00ebs duhet t\u00eb merresh me gjendjen, dhe ajo \u00ebsht\u00eb shum\u00eb m\u00eb e r\u00ebnd\u00eb se nj\u00eb mikrosh\u00ebrbim. Kubernetes mb\u00ebshtet ngarkesa q\u00eb mbajn\u00eb gjendje, por, si\u00e7 thekson Kelsey Hightower n\u00eb dy tweet-et e tij, ato duhet trajtuar me kujdes:<\/p>\n

Disa mendojn\u00eb se, n\u00ebse aplikon Kubernetes n\u00eb nj\u00eb ngarkes\u00eb q\u00eb mban gjendje, ajo shnd\u00ebrrohet n\u00eb nj\u00eb baz\u00eb t\u00eb dh\u00ebnash t\u00eb menaxhuar plot\u00ebsisht, e cila \u00ebsht\u00eb n\u00eb gjendje t\u00eb konkurroj\u00eb me RDS. Kjo nuk \u00ebsht\u00eb e v\u00ebrtet\u00eb. Ndoshta, n\u00ebse punoni mjaftuesh\u00ebm, urdh\u00ebroni komponente shtes\u00eb dhe angazhoni nj\u00eb ekip inxhinier\u00ebsh SRE, do t\u00eb arrini t\u00eb vendosni nj\u00eb RDS mbi Kubernetes.<\/p><\/blockquote>\n

Gjithmon\u00eb rekomandoj t\u00eb tregoni kujdes t\u00eb jasht\u00ebzakonsh\u00ebm kur lan\u00e7oni ngarkesa q\u00eb mbajn\u00eb gjendje n\u00eb Kubernetes. Shumica e atyre q\u00eb jan\u00eb t\u00eb interesuar, \"a mund t\u00eb ekzekutoj ngarkesa q\u00eb mbajn\u00eb gjendje n\u00eb Kubernetes\" nuk kan\u00eb p\u00ebrvoj\u00eb t\u00eb mjaftueshme n\u00eb pun\u00ebn me Kubernetes, dhe shpesh as me ngarkes\u00ebn p\u00ebr t\u00eb cil\u00ebn pyesin.<\/p><\/blockquote>\n

Pra, a duhet t\u00eb ekzekutohet Kafka n\u00eb Kubernetes? Nj\u00eb pyetje tjet\u00ebr: a do t\u00eb funksionoj\u00eb m\u00eb mir\u00eb Kafka pa Kubernetes? K\u00ebshtu q\u00eb d\u00ebshiroj t\u00eb theksoj n\u00eb k\u00ebt\u00eb artikull se si Kafka dhe Kubernetes plot\u00ebsojn\u00eb nj\u00ebri-tjetrin, dhe cilat jan\u00eb pengesat q\u00eb mund t\u00eb hasen kur i kombinoj\u00eb ato.<\/p>\n

Koha e ekzekutimit<\/h4>\n

\nLe t\u00eb flasim p\u00ebr nj\u00eb gj\u00eb bazike \u2014 mjedisi i ekzekutimit si nj\u00eb i till\u00eb.<\/p>\n

Procesi<\/b><\/i><\/p>\n

Brokerat Kafka jan\u00eb t\u00eb p\u00ebrshtatsh\u00ebm p\u00ebr pun\u00ebn me CPU. TLS mund t\u00eb sjell\u00eb disa kosto. Megjithat\u00eb, klient\u00ebt e Kafka mund t\u00eb ngarkojn\u00eb m\u00eb shum\u00eb CPU n\u00ebse p\u00ebrdorin enkriptim, por kjo nuk ndikon n\u00eb brokerat.<\/p>\n

Memoria<\/b><\/i><\/p>\n

Brokerat Kafka konsumojn\u00eb memorie. Madh\u00ebsia e grumbullit JVM zakonisht rekomandohet t\u00eb kufizohet n\u00eb 4\u20135 GB, por ju gjithashtu do t'ju nevojitet shum\u00eb memorie sistemike, pasi Kafka e p\u00ebrdor shum\u00eb aktivisht cache-in e faqeve. N\u00eb Kubernetes, caktoni n\u00eb m\u00ebnyr\u00eb t\u00eb p\u00ebrshtatshme kufizimet e burimeve t\u00eb konteiner\u00ebve dhe k\u00ebrkesat.<\/p>\n

Ruajtja e t\u00eb dh\u00ebnave<\/b><\/i><\/p>\n

Ruajtja e t\u00eb dh\u00ebnave n\u00eb konteiner\u00eb \u00ebsht\u00eb efemere \u2013 t\u00eb dh\u00ebnat humbin gjat\u00eb rinisjes. P\u00ebr t\u00eb dh\u00ebnat e Kafka, mund t\u00eb p\u00ebrdorni volumin emptyDir<\/code>, dhe efekti do t\u00eb jet\u00eb i ngjash\u00ebm: t\u00eb dh\u00ebnat e brokerit tuaj do t\u00eb humbasin pas p\u00ebrfundimit. Mesazhet tuaja mund t\u00eb ruajn\u00eb ende n\u00eb broker\u00eb t\u00eb tjer\u00eb si replika. Prandaj, pas rinisjes, brokeri i d\u00ebmtuar duhet t\u00eb riprodhoj\u00eb t\u00eb gjitha t\u00eb dh\u00ebnat e para, dhe ky proces mund t\u00eb marr\u00eb nj\u00eb koh\u00eb t\u00eb konsiderueshme. <\/p>\n

Kjo \u00ebsht\u00eb arsyeja pse duhet p\u00ebrdorur ruajtja e t\u00eb dh\u00ebnave afatgjat\u00eb. Le t\u00eb jet\u00eb kjo nj\u00eb ruajtje e jashtme afatgjat\u00eb me sistemin e skedar\u00ebve XFS ose, m\u00eb sakt\u00ebsisht, ext4. Mos p\u00ebrdorni NFS. T\u00eb kam paralajm\u00ebruar. NFS versionet v3 ose v4 nuk do t\u00eb funksionojn\u00eb. N\u00eb p\u00ebrmbledhje, brokeri Kafka do t\u00eb p\u00ebrfundoj\u00eb, n\u00ebse nuk arrin t\u00eb fshij\u00eb katalogun e t\u00eb dh\u00ebnave p\u00ebr shkak t\u00eb nj\u00eb problemi me \"em\u00ebrtimet e budallaq\u00ebve\", q\u00eb \u00ebsht\u00eb aktual n\u00eb NFS. N\u00ebse deri tani nuk jam bindur, lexoni me kujdes k\u00ebt\u00eb artikull<\/a><\/noindex>. Ruajtja e t\u00eb dh\u00ebnave duhet t\u00eb jet\u00eb e jashtme, n\u00eb m\u00ebnyr\u00eb q\u00eb Kubernetes t\u00eb mund t\u00eb zgjedh\u00eb m\u00eb fleksib\u00ebl nj\u00eb nyje t\u00eb re pas rinisjes ose relokimit.<\/p>\n

RRjeti<\/b><\/i><\/p>\n

Si dhe n\u00eb shumic\u00ebn e sistemeve t\u00eb shp\u00ebrndara, performanca e Kafka varet shum\u00eb nga minimizimi i vonesave n\u00eb rrjet dhe maksimizimi i gjer\u00ebsis\u00eb s\u00eb brezit. Mos provoni t\u00eb vendosni t\u00eb gjith\u00eb broker\u00ebt n\u00eb t\u00eb nj\u00ebjt\u00ebn nyje, pasi ky do t\u00eb ul\u00eb disponueshm\u00ebrin\u00eb. N\u00ebse ndalon nj\u00eb nyje Kubernetes, do t\u00eb ndaloj\u00eb gjithashtu t\u00ebr\u00eb grumbullin Kafka. Gjithashtu, mos shp\u00ebrndani grumbullin Kafka p\u00ebr qendra t\u00eb plota t\u00eb t\u00eb dh\u00ebnave. E nj\u00ebjta gj\u00eb vlen p\u00ebr grumbullin Kubernetes. Nj\u00eb kompromis i mir\u00eb n\u00eb k\u00ebt\u00eb rast \u00ebsht\u00eb t\u00eb zgjidhni zona t\u00eb ndryshme t\u00eb disponueshm\u00ebris\u00eb.<\/p>\n

Konfigurimi<\/h4>\n

\nManifestet e zakonshme<\/b><\/i><\/p>\n

N\u00eb faqen e Kubernetes ka nj\u00eb udh\u00ebzues t\u00eb shum\u00eb t\u00eb mir\u00eb<\/a><\/noindex> rreth m\u00ebnyr\u00ebs se si t\u00ebKonfiguroni ZooKeeper me ndihm\u00ebn e manifest\u00ebve. Duke qen\u00eb se ZooKeeper \u00ebsht\u00eb pjes\u00eb e Kafka, kjo \u00ebsht\u00eb nj\u00eb pik\u00eb e mir\u00eb p\u00ebr t\u00eb filluar t\u00eb njoh\u00ebsh konceptet e Kubernetes q\u00eb jan\u00eb t\u00eb aplikueshme k\u00ebtu. Pas k\u00ebsaj, do t\u00eb jeni n\u00eb gjendje t\u00eb aplikoni t\u00eb nj\u00ebjtat koncepte edhe p\u00ebr klasterin Kafka.<\/p>\n