Phawula. transl.: Injineli ye-American DevOps uSid Palas, usebenzisa Njengesikhokelo solwazi, ndithelekise iindleko zenkonzo ye-Kubernetes eLawulwayo (kwizicwangciso ezahlukeneyo) kubaboneleli belifu abahamba phambili behlabathi. Inzuzo eyongezelelweyo yomsebenzi wakhe yayikupapashwa kweJupyter Notebook ehambelanayo, evumela (ngolwazi oluncinci lwePython) ukulungelelanisa izibalo ezenziwe ukuze zihambelane neemfuno zakho.
TL; DR: I-Azure kunye ne-Digital Ocean ayihlawulisi kwi-compute resources ezisetyenziselwa i-airplane yokulawula, ibenze ibe yinto efanelekileyo yokusasaza amaqela amaninzi amancinci. Ukuqhuba inani elincinci lamaqela amakhulu, i-GKE ifaneleka kakhulu. Ukongeza, unokunciphisa kakhulu iindleko ngokusebenzisa i-spot / preemptive / low-priority nodes okanye "ngokubhalisa" ekusebenziseni ixesha elide lee-nodes ezifanayo (oku kusebenza kuzo zonke iiplatifomu).
Ubungakanani beqela (inani labasebenzi)
Ulwazi jikelele
Isibhengezo se-GKE sokuqalisa ukubiza iisenti ezili-10 ngeyure yeqela ngalinye ngeyure yeqela ngalinye landikhuthaza ukuba ndiqalise ukuhlalutya amaxabiso eminikelo elawulwa yiKubernetes.
Esi sibhengezo sibakhathaze kakhulu abanye...
Abalinganiswa abaphambili beli nqaku ngaba:
- Injini kaGoogle Kubernetes (GCP) - ;
- Inkonzo ye-Elastic Kubernetes (AWS) β ;
- Inkonzo yeAzure Kubernetes (Azure) β ;
- Kubernetes kuLwandlekazi lweDijithali- .
Ukucazululwa kweendleko
Ixabiso lilonke lokusebenzisa i-Kubernetes kwiqonga ngalinye kula liqulathe la macandelo alandelayo:
- Umrhumo wolawulo lwamaqela;
- Ukulinganisa umthwalo (ukwenzela i-Ingress);
- Izixhobo zekhompyutha (i-vCPU kunye nememori) yabasebenzi;
- Ukuphuma kwetrafikhi;
- Ukugcinwa okusisigxina;
- Ukulungiswa kwedatha ngokulinganisa umthwalo.
Ukongeza, ababoneleli ngamafu banikezela ngezaphulelo ezibalulekileyo ukuba umxhasi ufuna / unokusebenzisa i-preemptible okanye i-low-priority nodes OKANYE ithembisa ukusebenzisa iindawo ezifanayo iminyaka eyi-1-3.
Kufanelekile ukugxininisa ukuba nangona iindleko zisisiseko esilungileyo sokuthelekisa kunye nokuvavanya ababoneleli ngeenkonzo, ezinye izinto kufuneka zithathelwe ingqalelo:
- Ixesha lokuphumla (iSivumelwano seNqanaba leNkonzo);
- I-ecosystem yamafu ejikelezileyo;
- Iinguqulelo ezifumanekayo zee-K8s;
- Umgangatho wamaxwebhu/izixhobo zokusebenza.
Nangona kunjalo, ezi zinto zingaphaya komda weli nqaku / isifundo. IN Izinto ezingezona ixabiso le-EKS, i-AKS kunye ne-GKE zixoxwa ngokubanzi.
Jupyter Notebook
Ukwenza kube lula ukufumana isisombululo esinenzuzo kakhulu, ndiphuhlise , usebenzisa icebo + ipywidgets kuyo. Ikuvumela ukuba uthelekise unikezelo lomboneleli ngobungakanani beqela elihlukeneyo kunye neeseti zenkonzo.
Unokuziqhelanisa nenguqulelo ephilayo yencwadana yenqaku kwiBinder:
Ndazise ukuba izibalo okanye amaxabiso oqobo awachanekanga (oku kunokwenziwa ngomba okanye ukutsalwa kwesicelo kwiGitHub - ).
ezifunyanisiweyo
Owu, maninzi kakhulu ama-nuances okubonelela ngeengcebiso ezithe ngqo ngakumbi kunezo zibandakanyiweyo kumhlathi we-TL; DR kwasekuqaleni. Nangona kunjalo, ezinye izigqibo zisenokwenziwa:
- Ngokungafaniyo ne-GKE kunye ne-EKS, i-AKS kunye ne-Digital Ocean ayihlawulisi kwimithombo yolawulo yolawulo. I-AKS kunye ne-DO zinenzuzo ngakumbi ukuba ulwakhiwo lubandakanya amaqela amaninzi amancinci (umzekelo, iqoqo elinye wonke umphuhlisi okanye wonke umxhasi).
- Izixhobo zekhompyutha ze-GKE ezixabisa kancinci ziyenza ibenengeniso ngakumbi njengoko ubungakanani beqela* bonyuka.
- Ukusebenzisa i-node engabonakaliyo okanye i-node yexesha elide edibeneyo inokunciphisa iindleko ezingaphezu kwe-50%. Qaphela: I-Digital Ocean ayiboneleli ngezi zaphulelo.
- Imirhumo yeGoogle ephuma ngaphandle iphezulu, kodwa ixabiso lezixhobo zokusebenza zekhompuyutha ngumba omiselayo ekubaleni (ngaphandle kokuba iqela lakho livelisa isixa esibalulekileyo sedatha ephumayo).
- Ukukhetha iintlobo zoomatshini ngokusekwe kwi-CPU kunye neemfuno zememori zomthwalo wakho womsebenzi kuya kukunceda uphephe ukuhlawula ezongezelelweyo kwizixhobo ezingasetyenziswanga.
- I-Digital Ocean ihlawulisa ngaphantsi kwe-vCPU nangaphezulu kwimemori xa kuthelekiswa namanye amaqonga - oku kunokuba yinto ethatha isigqibo kwezinye iintlobo ze-compute workloads.
*Qaphela: Uhlahlelo lusebenzisa idatha ngenjongo eqhelekileyo yokubala iinodi (iinjongo jikelele). Le yimizekelo ye-n1 GCP Compute Engine, i-m5 AWS ec2 imizekelo, iD2v3 Azure virtual machines kunye nee-DO droplets ezine-CPU ezizinikeleyo. Emva koko, kunokwenzeka ukwenza uphando phakathi kwezinye iintlobo zoomatshini ababonakalayo (i-burstable, entry-level). Ukuqala kokujonga, iindleko zoomatshini benyani zixhomekeke ngokuthe ngqo kwinani le-vCPU kunye nobungakanani bememori, kodwa andiqinisekanga ukuba le ngcinga iya kubamba inyani kwimemori engasiyo i-standard / i-CPU ratios.
Inqaku , epapashwe kwi-2018, isebenzise i-reference cluster kunye ne-100 vCPU cores kunye ne-400 GB yememori. Ukuthelekisa, ngokobalo lwam, iqela elifanayo kwinqanaba ngalinye kula maqonga (kwiimeko ezifunekayo) liya kubiza le mali ilandelayo:
- AKS: 51465 USD / ngonyaka
- I-EKS: 43138 USD / ngonyaka
- I-GKE: 30870 USD / ngonyaka
- YENZA: 36131 USD / ngonyaka
Ndiyathemba ukuba eli nqaku kunye nencwadana iya kukunceda uvavanye iminikelo ephambili ye-Kubernetes elawulwayo kunye / okanye ugcine imali kwiziseko zamafu ngokusebenzisa izaphulelo kunye namanye amathuba.
PS evela kumguquleli
Funda nakwibhlog yethu:
- Β«";
- Β«";
- Β«";
- «».
umthombo: www.habr.com