{"id":81888,"date":"2020-05-17T13:42:23","date_gmt":"2020-05-17T11:42:23","guid":{"rendered":"https:\/\/prohoster.info\/blog\/administrirovanie\/thanos-masshtabiruemyj-prometheus"},"modified":"2020-05-17T13:42:23","modified_gmt":"2020-05-17T11:42:23","slug":"thanos-masshtabiruemyj-prometheus","status":"publish","type":"post","link":"https:\/\/prohoster.info\/sq\/blog\/administrirovanie\/thanos-masshtabiruemyj-prometheus","title":{"rendered":"Thanos \u2014 nj\u00eb Prometheus i shkall\u00ebzuesh\u00ebm","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"

P\u00ebrkthimi i artikullit \u00ebsht\u00eb p\u00ebrgatitur ve\u00e7an\u00ebrisht p\u00ebr student\u00ebt e kursit \u00abPraktikat dhe mjetet DevOps\u00bb<\/a><\/noindex>.<\/i><\/b><\/p>\n

\nFabian Reinartz<\/a><\/noindex> \u2014 zhvillues softueri, adhurues i Go dhe i apasionuar pas zgjidhjes s\u00eb problemeve t\u00eb nd\u00ebrlikuara. Ai \u00ebsht\u00eb gjithashtu mjaftues i Prometheus dhe bashkreator i Kubernetes SIG instrumentation. N\u00eb t\u00eb kaluar\u00ebn, ai ka qen\u00eb inxhinier n\u00eb SoundCloud dhe ka drejtuar grupin e monitorimit n\u00eb CoreOS. Aktualisht punon n\u00eb Google.<\/p>\n

Bartek Plotka<\/a><\/noindex> \u2014 inxhinier infrastrukture n\u00eb Improbable. Ai \u00ebsht\u00eb i apasionuar pas teknologjive t\u00eb reja dhe problemeve t\u00eb sistemeve t\u00eb shp\u00ebrndara. Ka p\u00ebrvoj\u00eb n\u00eb programimin n\u00eb nivel t\u00eb ul\u00ebt te Intel, p\u00ebrvoj\u00eb si kontributor n\u00eb Mesos dhe p\u00ebrvoj\u00eb n\u00eb SRE n\u00eb shkall\u00eb globale n\u00eb Improbable. Ai merret me p\u00ebrmir\u00ebsimin e bot\u00ebs s\u00eb mikrosh\u00ebrbimeve. Tre pasionet e tij: Golang, open source dhe volejboll.<\/i><\/p>\n

Duke par\u00eb produktin ton\u00eb m\u00eb t\u00eb njohur, SpatialOS, mund t\u00eb kuptoni se Improbable ka nevoj\u00eb p\u00ebr nj\u00eb infrastruktur\u00eb cloud t\u00eb l\u00ebvizshme n\u00eb shkall\u00eb globale me dhjet\u00ebra klaster\u00eb Kubernetes. Ne ishim disa nga t\u00eb par\u00ebt q\u00eb filluam t\u00eb p\u00ebrdorim sistemin e monitorimit Prometheus<\/a><\/noindex>. Prometheus \u00ebsht\u00eb n\u00eb gjendje t\u00eb ndjek\u00eb miliona metrika n\u00eb koh\u00eb reale dhe vjen me nj\u00eb gjuh\u00eb t\u00eb fuqishme pyetjesh q\u00eb lejon nxjerrjen e informacionit t\u00eb nevojsh\u00ebm.<\/p>\n

Thjesht\u00ebsia dhe besueshm\u00ebria e Prometheus jan\u00eb disa nga p\u00ebrfitimet e tij kryesore. Megjithat\u00eb, pas arritjes s\u00eb nj\u00eb shkalle t\u00eb caktuar, ne u p\u00ebrball\u00ebm me disa disavantazhe. P\u00ebr t\u00eb zgjidhur k\u00ebto probleme, ne zhvilluam Thanos<\/a><\/noindex> \u2014 nj\u00eb projekt me kod t\u00eb hapur, i krijuar nga Improbable, p\u00ebr transformimin e menj\u00ebhersh\u00ebm t\u00eb klaster\u00ebve ekzistues t\u00eb Prometheus n\u00eb nj\u00eb sistem t\u00eb vet\u00ebm monitorimi me ruajtje t\u00eb pafundme t\u00eb t\u00eb dh\u00ebnave historike. Thanos \u00ebsht\u00eb i disponuesh\u00ebm n\u00eb Github k\u00ebtu<\/a><\/noindex>.<\/a><\/noindex><\/p>\n

Q\u00ebndroni t\u00eb informuar mbi lajmet e fundit nga Improbable.<\/a><\/noindex><\/p>\n

Q\u00ebllimet tona me Thanos<\/h2>\n

\nKur arrijm\u00eb nj\u00eb shkall\u00eb t\u00eb caktuar, lindin probleme q\u00eb shk\u00eblqejn\u00eb p\u00ebrtej mund\u00ebsive t\u00eb Prometheus vanilje. Si t\u00eb ruajm\u00eb besuesh\u00ebm dhe me kostot m\u00eb t\u00eb ulta petabajt t\u00eb t\u00eb dh\u00ebnave historike? A mund t\u00eb b\u00ebhet kjo pa e d\u00ebmtuar koh\u00ebn e p\u00ebrgjigjes n\u00eb k\u00ebrkesat? A mund t\u00eb kemi qasje n\u00eb t\u00eb gjitha metrikat q\u00eb ndodhen n\u00eb server\u00ebt e ndrysh\u00ebm Prometheus me nj\u00eb k\u00ebrkes\u00eb API? A mund t\u00eb bashkojm\u00eb ndonj\u00ebher\u00eb t\u00eb dh\u00ebnat e riprodhuara t\u00eb mbledhura nga Prometheus HA?<\/p>\n

P\u00ebr t\u00eb zgjidhur k\u00ebto pyetje, krijuam Thanos. N\u00eb seksionet e ardhshme p\u00ebrshkruhet si i qas\u00ebm k\u00ebtyre pyetjeve dhe shpjegohen q\u00ebllimet q\u00eb kemi ndjekur.<\/p>\n

K\u00ebrkesa e t\u00eb dh\u00ebnave nga disa instanca Prometheus (k\u00ebrkesa globale)<\/h4>\n

\nPrometheus ofron nj\u00eb qasje funksionale ndaj sharding-ut. Edhe nj\u00eb server Prometheus siguron mjaftuesh\u00ebm shkall\u00ebzim p\u00ebr t\u00eb \u00e7liruar p\u00ebrdoruesit nga kompleksitetet e sharding-ut horizontal n\u00eb pothuajse t\u00eb gjitha rastet e p\u00ebrdorimit.<\/p>\n

Megjith\u00ebse kjo \u00ebsht\u00eb nj\u00eb model i shk\u00eblqyer implementimi, shpesh k\u00ebrkohet qasje n\u00eb t\u00eb dh\u00ebna nga servera t\u00eb ndrysh\u00ebm Prometheus p\u00ebrmes nj\u00eb API ose UI t\u00eb vet\u00ebm \u2014 pamja globale. Sigurisht, ekziston mund\u00ebsia p\u00ebr t\u00eb shfaqur disa k\u00ebrkesa n\u00eb nj\u00eb panel Grafana, por secila k\u00ebrkes\u00eb mund t\u00eb ekzekutohet vet\u00ebm n\u00eb nj\u00eb server Prometheus. Nga ana tjet\u00ebr, me Thanos mund t\u00eb k\u00ebrkoni dhe agregoni t\u00eb dh\u00ebna nga disa servera Prometheus, pasi t\u00eb gjitha jan\u00eb t\u00eb aksesueshme nga nj\u00eb pik\u00eb fundore.<\/p>\n

M\u00eb par\u00eb, p\u00ebr t\u00eb marr\u00eb pamjen globale n\u00eb Improbable, ne organizuam instancat tona Prometheus n\u00eb nj\u00eb shum\u00eb-niveli Federata Hierarkike<\/a><\/noindex>. Kjo n\u00ebnkuptonte krijimin e nj\u00eb meta-serveri Prometheus, i cili mbledh disa nga metrikat nga \u00e7do server \"leht\u00ebsues\".<\/p>\n

\"Thanos
\n
\nKy qasje u tregua problematike. Ai \u00e7oi n\u00eb ngarkes\u00eb t\u00eb konfigurimit, shtimin e nj\u00eb pik\u00eb t\u00eb mundshme d\u00ebshtimi dhe zbatimin e rregullave komplekse p\u00ebr t\u00eb ofruar pik\u00ebn e fundore t\u00eb federuar vet\u00ebm t\u00eb dh\u00ebnat e nevojshme. P\u00ebr m\u00eb tep\u00ebr, federata e k\u00ebtij lloji nuk lejon t\u00eb merret nj\u00eb pamje globale reale, pasi jo t\u00eb gjitha t\u00eb dh\u00ebnat jan\u00eb t\u00eb aksesueshme nga nj\u00eb k\u00ebrkes\u00eb API.<\/p>\n

Kjo lidhet ngusht\u00ebsisht me nj\u00eb pamje t\u00eb unifikuar t\u00eb t\u00eb dh\u00ebnave t\u00eb grumbulluara n\u00eb serverat Prometheus me disponueshm\u00ebri t\u00eb lart\u00eb (high-availability, HA). Modeli HA i Prometheus grumbullon t\u00eb dh\u00ebna n\u00eb m\u00ebnyr\u00eb t\u00eb pavarur dy her\u00eb, \u00e7ka \u00ebsht\u00eb aq e thjesht\u00eb sa nuk mund t\u00eb jet\u00eb m\u00eb e thjesht\u00eb. Megjithat\u00eb, do t\u00eb ishte shum\u00eb m\u00eb e p\u00ebrshtatshme t\u00eb p\u00ebrdorej nj\u00eb pamje e bashkuar dhe e deduplicuar e t\u00eb dy rrjedhave.<\/p>\n

Sigurisht, ekziston nj\u00eb nevoj\u00eb p\u00ebr serverat Prometheus me disponueshm\u00ebri t\u00eb lart\u00eb. N\u00eb Improbable ne e marrim seriozisht monitorimin minut\u00eb pas minute t\u00eb t\u00eb dh\u00ebnave, por prania e nj\u00eb instance Prometheus n\u00eb klas\u00ebr \u00ebsht\u00eb nj\u00eb pik\u00eb e vetme d\u00ebshtimi. \u00c7do gabim konfigurimi ose d\u00ebshtim t\u00eb pajisjeve mund potencialisht t\u00eb \u00e7oj\u00eb n\u00eb humbje t\u00eb t\u00eb dh\u00ebnave t\u00eb r\u00ebnd\u00ebsishme. Edhe nj\u00eb implementim i thjesht\u00eb mund t\u00eb shkaktoj\u00eb disa probleme n\u00eb grumbullimin e metrikeve, pasi rinisja mund t\u00eb jet\u00eb shum\u00eb m\u00eb e gjat\u00eb se intervali i skrapimit.<\/p>\n

Ruajtja e besueshme e t\u00eb dh\u00ebnave historike<\/h4>\n

\nNj\u00eb ruajtje e lir\u00eb, e shpejt\u00eb dhe me afat t\u00eb gjat\u00eb p\u00ebr metrikat \u00ebsht\u00eb \u00ebndrra jon\u00eb (t\u00eb ndar\u00eb nga shumica e p\u00ebrdoruesve t\u00eb Prometheus). N\u00eb Improbable, na duhej t\u00eb konfiguronim afatin e ruajtjes s\u00eb metrikave n\u00eb n\u00ebnt\u00eb dit\u00eb (p\u00ebr Prometheus 1.8). Kjo sjell kufizime t\u00eb dukshme n\u00eb sa larg mund t\u00eb shkojm\u00eb mbrapa.<\/p>\n

Prometheus 2.0 u p\u00ebrmir\u00ebsua n\u00eb k\u00ebt\u00eb aspekt, pasi numri i serive temporale nuk ndikon m\u00eb n\u00eb performanc\u00ebn e p\u00ebrgjithshme t\u00eb serverit (shiko. KubeCon keynote about Prometheus 2<\/a><\/noindex>). Megjithat\u00eb, Prometheus ruan t\u00eb dh\u00ebnat n\u00eb diskun lokal. Edhe pse kompresimi i drejtp\u00ebrdrejt\u00eb i t\u00eb dh\u00ebnave mund t\u00eb reduktoj\u00eb ndjesh\u00ebm p\u00ebrdorimin e SSD-s\u00eb lokale, prap\u00ebseprap\u00eb, ekziston nj\u00eb kufizim n\u00eb sasin\u00eb e t\u00eb dh\u00ebnave historike q\u00eb mund t\u00eb ruhen.<\/p>\n

P\u00ebr m\u00eb tep\u00ebr, n\u00eb Improbable ne kujdesemi p\u00ebr besueshm\u00ebrin\u00eb, thjesht\u00ebsin\u00eb dhe costin. Disket e m\u00ebdhenj lokal jan\u00eb m\u00eb t\u00eb v\u00ebshtir\u00eb p\u00ebr t'u operuar dhe p\u00ebr t'u b\u00ebr\u00eb kopje rezerv\u00eb. Ato kushtojn\u00eb m\u00eb shum\u00eb dhe k\u00ebrkojn\u00eb m\u00eb shum\u00eb mjete p\u00ebr kopje rezerv\u00eb, duke sjell\u00eb kompleksitet t\u00eb tepruar.<\/p>\n

Downsampling<\/h4>\n

\nSapo filluam t\u00eb punojm\u00eb me t\u00eb dh\u00ebnat historike, kuptuam se ekzistojn\u00eb komplekse themelore me O-madh\u00ebsi, q\u00eb b\u00ebjn\u00eb q\u00eb k\u00ebrkesat t\u00eb b\u00ebhen gjithnj\u00eb e m\u00eb t\u00eb ngadalta kur punojm\u00eb me t\u00eb dh\u00ebna p\u00ebr jav\u00eb, muaj dhe vite.<\/p>\n

Zgjidhja standarde p\u00ebr k\u00ebt\u00eb problem do t\u00eb jet\u00eb downsampling<\/a><\/noindex> (redukimi i frekuenc\u00ebs s\u00eb mostrave) \u2014 duke ulur frekuenc\u00ebn e mostrave t\u00eb sinjalit. Me downsampling mund t\u00eb 'p\u00ebrmasojm\u00eb' n\u00eb nj\u00eb interval m\u00eb t\u00eb madh kohor dhe t\u00eb mbajm\u00eb t\u00eb nj\u00ebjtin num\u00ebr mostrash, gj\u00eb q\u00eb do t\u00eb ruaj\u00eb p\u00ebrgjigjen e shpejt\u00eb t\u00eb k\u00ebrkesave.<\/p>\n

Downsampling i t\u00eb dh\u00ebnave t\u00eb vjetra \u00ebsht\u00eb nj\u00eb k\u00ebrkes\u00eb e pashmangshme p\u00ebr \u00e7do zgjidhje afatgjat\u00eb ruajtjeje dhe kalon p\u00ebrtej Prometheus-it vanilje.<\/p>\n

Objektiva t\u00eb tjera<\/h4>\n

\nNj\u00eb nga q\u00ebllimet fillestare t\u00eb projektit Thanos ishte integrimi pa nd\u00ebrprerje me \u00e7do instalim ekzistues t\u00eb Prometheus. Q\u00ebllimi i dyt\u00eb ishte operimi i thjesht\u00eb me nj\u00eb penges\u00eb minimale n\u00eb hyrje. \u00c7do var\u00ebsi duhet t\u00eb plot\u00ebsohet leht\u00ebsisht p\u00ebr p\u00ebrdoruesit e vegj\u00ebl dhe ata t\u00eb m\u00ebdhenj, q\u00eb gjithashtu n\u00ebnkupton nj\u00eb kostod t\u00eb ul\u00ebt baz\u00eb.<\/p>\n

Arkitektura e Thanos<\/h2>\n

\nPasi t\u00eb kemi listuar q\u00ebllimet tona n\u00eb seksionin e m\u00ebparsh\u00ebm, le t\u00eb punojm\u00eb mbi to dhe t\u00eb shikojm\u00eb se si Thanos i zgjidh k\u00ebto probleme.<\/p>\n

Pamja globale<\/h4>\n

\nP\u00ebr t\u00eb marr\u00eb nj\u00eb pamje globale mbi ekzistencat ekzistuese t\u00eb Prometheus, na nevojitet lidhja e nj\u00eb pik\u00eb t\u00eb vetme hyrjeje p\u00ebr k\u00ebrkesat me t\u00eb gjitha server\u00ebt. K\u00ebt\u00eb pun\u00eb e kryen komponenti Thanos. Sidecar<\/a><\/noindex>. Ai deploy-het pran\u00eb \u00e7do serveri Prometheus dhe funksionon si nj\u00eb proxy, duke sh\u00ebrbyer t\u00eb dh\u00ebnat lokale t\u00eb Prometheus p\u00ebrmes interfes\u00ebs gRPC t\u00eb Store API, e cila lejon zgjedhjen e t\u00eb dh\u00ebnave time series sipas etiketave dhe intervalit t\u00eb koh\u00ebs.<\/p>\n

Nga ana tjet\u00ebr ndodhet komponenti i shkall\u00ebzuesh\u00ebm horizontalisht Querier pa ruajtje t\u00eb gjendjes, i cili b\u00ebn pak m\u00eb shum\u00eb se thjesht p\u00ebrgjigjet n\u00eb k\u00ebrkesat PromQL p\u00ebrmes standardit HTTP API t\u00eb Prometheus. Komponent\u00ebt Querier, Sidecar dhe Thanos t\u00eb tjer\u00eb komunikojn\u00eb p\u00ebrmes protokollit gossip.<\/a><\/noindex>.<\/p>\n

\"Thanos
\n<\/p>\n

    \n
  1. Kur Querier merr nj\u00eb k\u00ebrkes\u00eb, ai lidhet me serverin p\u00ebrkat\u00ebs t\u00eb Store API, pra me Sidecar-\u00ebt tan\u00eb dhe merr t\u00eb dh\u00ebnat time series nga server\u00ebt p\u00ebrkat\u00ebs t\u00eb Prometheus.<\/li>\n
  2. Pas k\u00ebsaj, ai kombinon p\u00ebrgjigjet dhe realizon k\u00ebrkes\u00ebn n\u00eb PromQL. Querier mund t\u00eb bashkoj\u00eb si t\u00eb dh\u00ebna q\u00eb nuk prahendin ashtu edhe t\u00eb dh\u00ebna t\u00eb p\u00ebrs\u00ebritura nga server\u00ebt HA t\u00eb Prometheus.<\/li>\n<\/ol>\n

    \nKjo zgjidh pjes\u00ebn kryesore t\u00eb misterit ton\u00eb \u2014 bashkimin e t\u00eb dh\u00ebnave nga server\u00ebt e izoluar t\u00eb Prometheus n\u00eb nj\u00eb p\u00ebrfaq\u00ebsim t\u00eb vet\u00ebm. N\u00eb fakt, Thanos mund t\u00eb p\u00ebrdoret thjesht p\u00ebr k\u00ebt\u00eb mund\u00ebsi. Nuk k\u00ebrkohen ndryshime n\u00eb server\u00ebt ekzistues t\u00eb Prometheus!<\/p>\n

    Koh\u00eb ruajtjeje t\u00eb pakufizuar!<\/h4>\n

    \nMegjithat\u00eb, her\u00ebt a von\u00eb do t\u00eb d\u00ebshirojm\u00eb t\u00eb ruajm\u00eb t\u00eb dh\u00ebna q\u00eb kalojn\u00eb p\u00ebrtej koh\u00ebs standarde t\u00eb ruajtjes s\u00eb Prometheus. P\u00ebr ruajtjen e t\u00eb dh\u00ebnave historike, ne kemi zgjedhur nj\u00eb ruajtje objektive. Ajo \u00ebsht\u00eb gjer\u00ebsisht e disponueshme n\u00eb \u00e7do cloud, si dhe n\u00eb qendrat lokale t\u00eb t\u00eb dh\u00ebnave dhe \u00ebsht\u00eb shum\u00eb ekonomike. Gjithashtu, praktikuese \u00e7do ruajtje objektiv \u00ebsht\u00eb n\u00eb dispozicion p\u00ebrmes njohur S3 API.<\/p>\n

    Prometheus shkruan t\u00eb dh\u00ebnat nga memorja n\u00eb disk \u00e7do rreth dy or\u00eb. Blloku i t\u00eb dh\u00ebnave t\u00eb ruajtura p\u00ebrmban t\u00eb gjitha t\u00eb dh\u00ebnat p\u00ebr nj\u00eb periudh\u00eb fikse t\u00eb koh\u00ebs dhe \u00ebsht\u00eb i pandryshuesh\u00ebm. Kjo \u00ebsht\u00eb shum\u00eb praktike, pasi Thanos Sidecar mund t\u00eb shikoj\u00eb thjesht katalogun e t\u00eb dh\u00ebnave t\u00eb Prometheus dhe, nd\u00ebrsa shfaqen blloqe t\u00eb reja, t'i ngarkoj\u00eb ato n\u00eb kapacitetet e ruajtjes objektive.<\/p>\n

    \"Thanos
    \n
    \nNgarkimi n\u00eb ruajtjen objektive menj\u00ebher\u00eb pasi shkruhet n\u00eb disk gjithashtu lejon ruajtjen e thjesht\u00ebsis\u00eb s\u00eb \"skraperit\" (Prometheus dhe Thanos Sidecar). Kjo e b\u00ebn m\u00eb t\u00eb leht\u00eb mb\u00ebshtetje, kosto dhe dizajnin e sistemit.<\/p>\n

    Si e shihni, kopjimi i t\u00eb dh\u00ebnave realizohet shum\u00eb leht\u00eb. Por \u00e7far\u00eb ndodh me k\u00ebrkes\u00ebn p\u00ebr t\u00eb dh\u00ebna n\u00eb ruajtjen objektore?<\/p>\n

    Komponenti Thanos Store vepron si nj\u00eb proxy p\u00ebr marrjen e t\u00eb dh\u00ebnave nga ruajtja objektore. Si\u00e7 \u00ebsht\u00eb Thanos Sidecar, ai merr pjes\u00eb n\u00eb klasterin gossip dhe implementon API-n\u00eb e Store. K\u00ebshtu, Querier-\u00ebt ekzistues mund ta konsiderojn\u00eb at\u00eb si Sidecar, si nj\u00eb burim t\u00eb tjera t\u00eb dh\u00ebnash t\u00eb koh\u00ebs \u2014 nuk k\u00ebrkohet ndonj\u00eb konfigurim t\u00eb ve\u00e7ant\u00eb.<\/p>\n

    \"Thanos
    \n
    \nBlloqet e t\u00eb dh\u00ebnave t\u00eb koh\u00ebs p\u00ebrb\u00ebhen nga disa skedar\u00eb t\u00eb m\u00ebdhenj. Ngarkimi i tyre sipas k\u00ebrkes\u00ebs do t\u00eb ishte mjaft i paefektsh\u00ebm, dhe keqmenaxhimi lokal do t\u00eb k\u00ebrkonte nj\u00eb memorie dhe hap\u00ebsir\u00eb disku t\u00eb madhe.<\/p>\n

    N\u00eb vend t\u00eb k\u00ebsaj, Store Gateway di si t\u00eb trajtoj\u00eb formatin e ruajtjes Prometheus. Fal\u00eb planifikuesit t\u00eb zgjuar t\u00eb k\u00ebrkesave dhe keqmenaxhimit vet\u00ebm t\u00eb pjes\u00ebve t\u00eb nevojshme t\u00eb indekseve t\u00eb blloqeve, \u00ebsht\u00eb b\u00ebr\u00eb e mundur t\u00eb reduktohen k\u00ebrkesat e komplikuara n\u00eb nj\u00eb num\u00ebr minimal HTTP-shkresash p\u00ebr skedar\u00ebt e ruajtjes objektore. K\u00ebshtu, numri i k\u00ebrkesave mund t\u00eb reduktohet nga kat\u00ebr deri n\u00eb gjasht\u00eb rendi dhe t\u00eb arrihet nj\u00eb koh\u00eb p\u00ebrgjigjeje q\u00eb \u00ebsht\u00eb n\u00eb p\u00ebrgjith\u00ebsi e v\u00ebshtir\u00eb t\u00eb ndahen nga k\u00ebrkesat p\u00ebr t\u00eb dh\u00ebna n\u00eb SSD-n\u00eb lokale.<\/p>\n

    \"Thanos
    \n
    \nSi\u00e7 tregohet n\u00eb diagramin e m\u00ebsip\u00ebrm, Thanos Querier e zvog\u00eblon ndjesh\u00ebm koston e k\u00ebrkes\u00ebs p\u00ebr t\u00eb dh\u00ebna n\u00eb ruajtjen objektore, duke p\u00ebrdorur formatin e ruajtjes Prometheus dhe duke vendosur t\u00eb dh\u00ebnat e lidhura af\u00ebr. Duke p\u00ebrdorur k\u00ebt\u00eb qasje, mund t\u00eb kombinojm\u00eb shum\u00eb k\u00ebrkesa individuale n\u00eb nj\u00eb num\u00ebr minimal operacionesh bulk.<\/p>\n

    Kompaktimi dhe ulja e mostrave<\/h4>\n

    \nPasi blloku i ri i t\u00eb dh\u00ebnave t\u00eb koh\u00ebs \u00ebsht\u00eb ngarkuar me sukses n\u00eb ruajtjen objektore, ne e konsiderojm\u00eb at\u00eb si t\u00eb dh\u00ebna \"historike\" q\u00eb menj\u00ebher\u00eb b\u00ebhen t\u00eb disponueshme p\u00ebrmes Store Gateway.<\/p>\n

    Megjithat\u00eb, pas nj\u00eb kohe, blloqet nga nj\u00eb burim (Prometheus me Sidecar) grumbullohen dhe tashm\u00eb nuk p\u00ebrdorin t\u00eb gjith\u00eb potencialin e indekseve. P\u00ebr t\u00eb zgjidhur k\u00ebt\u00eb problem, ne kemi futur nj\u00eb komponent t\u00eb ri t\u00eb quajtur Compactor. Ai thjesht aplikon mekanizmin lokal t\u00eb kompresimit Prometheus n\u00eb t\u00eb dh\u00ebnat historike n\u00eb ruajtjen objektore dhe mund t\u00eb nis\u00eb si nj\u00eb detyr\u00eb periodike t\u00eb thjesht\u00eb.<\/p>\n

    \"Thanos
    \n
    \nFal\u00eb shkrirjes efikase, nj\u00eb k\u00ebrkes\u00eb p\u00ebr ruajtje p\u00ebr nj\u00eb koh\u00eb t\u00eb gjat\u00eb nuk paraqet probleme nga pik\u00ebpamja e madh\u00ebsis\u00eb s\u00eb t\u00eb dh\u00ebnave. Megjithat\u00eb, kostoja e mundshme e shp\u00ebrndarjes s\u00eb nj\u00eb miliardi vlerash dhe kalimit t\u00eb tyre p\u00ebrmes procesorit t\u00eb k\u00ebrkesave do t\u00eb \u00e7oj\u00eb me siguri n\u00eb nj\u00eb rritje drastike t\u00eb koh\u00ebs s\u00eb ekzekutimit t\u00eb k\u00ebrkes\u00ebs. Nga ana tjet\u00ebr, p\u00ebr shkak se p\u00ebr \u00e7do piksel ekrani ka qindra pika t\u00eb dh\u00ebnash, b\u00ebhet e pamundur madje edhe t\u00eb vizualizohen t\u00eb dh\u00ebnat n\u00eb nj\u00eb rezolucion t\u00eb plot\u00eb. Prandaj, downsampling jo vet\u00ebm q\u00eb \u00ebsht\u00eb i mundur, por gjithashtu nuk do t\u00eb \u00e7oj\u00eb n\u00eb nj\u00eb humbje t\u00eb dukshme t\u00eb sakt\u00ebsis\u00eb. <\/p>\n

    \"Thanos
    \n
    \nP\u00ebr downsampling t\u00eb t\u00eb dh\u00ebnave, Compactor grumbullon vazhdimisht t\u00eb dh\u00ebnat me rezolucion pes\u00eb minuta dhe nj\u00eb or\u00eb. P\u00ebr \u00e7do fragment t\u00eb pa p\u00ebrpunuar, t\u00eb koduar me kompresimin TSDB XOR, ruhet nj\u00eb gam\u00eb e llojeve t\u00eb ndryshme t\u00eb t\u00eb dh\u00ebnave t\u00eb agreguar, si min, max apo sum p\u00ebr nj\u00eb bllok. Kjo i lejon Querier-it t\u00eb zgjedh\u00eb automatikisht agregatin q\u00eb \u00ebsht\u00eb i p\u00ebrshtatsh\u00ebm p\u00ebr k\u00ebrkes\u00ebn p\u00ebrkat\u00ebse PromQL. <\/p>\n

    P\u00ebr t\u00eb p\u00ebrdorur t\u00eb dh\u00ebnat me sakt\u00ebsi t\u00eb ulur, p\u00ebrdoruesi nuk ka nevoj\u00eb p\u00ebr asnj\u00eb konfigurim t\u00eb ve\u00e7ant\u00eb. Querier automatikisht kalon midis rezolucioneve t\u00eb ndryshme dhe t\u00eb dh\u00ebnave t\u00eb pa p\u00ebrpunuara nd\u00ebrsa p\u00ebrdoruesi rrit dhe zvog\u00eblon shkall\u00ebn. N\u00ebse d\u00ebshiron, p\u00ebrdoruesi mund ta menaxhoj\u00eb k\u00ebt\u00eb drejtp\u00ebrdrejt p\u00ebrmes parametrave \"step\" n\u00eb k\u00ebrkes\u00eb. <\/p>\n

    Duke qen\u00eb se kostoja e ruajtjes s\u00eb nj\u00eb GB \u00ebsht\u00eb e vog\u00ebl, Thanos p\u00ebr default ruan t\u00eb dh\u00ebnat origjinale, t\u00eb dh\u00ebnat me rezolucion pes\u00eb minuta dhe nj\u00eb or\u00eb. Nuk ka nevoj\u00eb t\u00eb fshihen t\u00eb dh\u00ebnat origjinale.<\/p>\n

    Rregullat e regjistrimit<\/h2>\n

    \nEdhe me Thanos, rregullat e regjistrimit jan\u00eb nj\u00eb pjes\u00eb thelb\u00ebsore e grumbullit t\u00eb monitorimit. Ato reduktojn\u00eb kompleksitetin, vones\u00ebn dhe kostot e k\u00ebrkesave. Ato gjithashtu jan\u00eb t\u00eb dobishme p\u00ebr p\u00ebrdoruesit p\u00ebr t\u00eb marr\u00eb t\u00eb dh\u00ebna t\u00eb agreguara mbi metrikat. Thanos bazohet n\u00eb instancat vanilla t\u00eb Prometheus, prandaj \u00ebsht\u00eb krejt\u00ebsisht e pranueshme t\u00eb ruhet rregullat e regjistrimit dhe rregullat e alarmit n\u00eb serverin ekzistues Prometheus. Sidoqoft\u00eb, n\u00eb disa raste, kjo mund t\u00eb mos jet\u00eb e mjaftueshme:<\/p>\n