serveru<\/a>, tak do HDFS.<\/p>\nV ka\u017ed\u00e9 transformaci na z\u00e1klad\u011b \u0161ablony lze defaultn\u011b vytvo\u0159it vlastn\u00ed skript v jazyce PDL, kter\u00fd pon\u011bkud p\u0159ipom\u00edn\u00e1 shell. <\/p>\n
Pomoc\u00ed jazyka PDL m\u016f\u017eete roz\u0161i\u0159ovat funk\u010dnost transformac\u00ed a zejm\u00e9na m\u016f\u017eete dynamicky (b\u011bhem vykon\u00e1v\u00e1n\u00ed) generovat libovoln\u00e9 fragmenty k\u00f3du v z\u00e1vislosti na parametrech vykon\u00e1n\u00ed.<\/p>\n
Tak\u00e9 v Ab Initio je siln\u011b rozvinuta integrace s opera\u010dn\u00edmi syst\u00e9my prost\u0159ednictv\u00edm shellu. Konkr\u00e9tn\u011b v Sberbank se pou\u017e\u00edv\u00e1 linux ksh. M\u016f\u017eete si vym\u011b\u0148ovat s shell prom\u011bnn\u00e9 a pou\u017e\u00edvat je jako parametry graf\u016f. M\u016f\u017eete p\u0159es shell spou\u0161t\u011bt vykon\u00e1n\u00ed graf\u016f Ab Initio a spravovat Ab Initio.<\/p>\n
Krom\u011b Ab Initio GDE se dod\u00e1v\u00e1 mnoho dal\u0161\u00edch produkt\u016f. Existuje vlastn\u00ed Co>Operation System, kter\u00e1 si klade za c\u00edl b\u00fdt opera\u010dn\u00edm syst\u00e9mem. Existuje Control>Center, ve kter\u00e9m lze napl\u00e1novat a monitorovat toky na\u010d\u00edt\u00e1n\u00ed. Existuj\u00ed produkty pro v\u00fdvoj na primitivn\u011bj\u0161\u00ed \u00farovni, ne\u017e jakou umo\u017e\u0148uje Ab Initio GDE.<\/p>\n
Popis r\u00e1mce MDW a pr\u00e1ce na jeho dolad\u011bn\u00ed pro GreenPlum<\/h3>\n \nSpole\u010dn\u011b se sv\u00fdmi produkty dod\u00e1v\u00e1 dodavatel i produkt MDW (Metadata Driven Warehouse), kter\u00fd p\u0159edstavuje konfigur\u00e1tor graf\u016f ur\u010den\u00fd k podpo\u0159e typick\u00fdch \u00faloh pro napl\u0148ov\u00e1n\u00ed datov\u00fdch sklad\u016f nebo data vaults.<\/p>\n
Obsahuje u\u017eivatelsk\u00e9 (specifick\u00e9 pro projekt) parsery metadat a hotov\u00e9 gener\u00e1tory k\u00f3du \u201ez krabice\u201c.<\/p>\n
V p\u0159\u00edpadech Hive a RDBMS se generuj\u00ed r\u016fznorod\u00e9 grafy pro inicia\u010dn\u00ed a inkrement\u00e1ln\u00ed aktualizaci dat.<\/p>\n
V p\u0159\u00edpad\u011b Hive jsou p\u0159ijat\u00e1 data delta spojena prost\u0159ednictv\u00edm Ab Initio Join s daty, kter\u00e1 byla v tabulce p\u0159ed aktualizac\u00ed. Na\u010d\u00edtac\u00ed aplikace v MDW (jak v Hive, tak v RDBMS) nejen vkl\u00e1daj\u00ed nov\u00e1 data z delty, ale tak\u00e9 uzav\u00edraj\u00ed obdob\u00ed platnosti dat podle prim\u00e1rn\u00edch kl\u00ed\u010d\u016f, ke kter\u00fdm byla delta p\u0159ijata. Krom\u011b toho je nutn\u00e9 znovu p\u0159epsat nezm\u011bn\u011bnou \u010d\u00e1st dat. Ale to je nezbytn\u00e9, proto\u017ee v Hive neexistuj\u00ed operace delete nebo update.<\/p>\n
Takto jsme dosp\u011bli k z\u00e1v\u011bru, \u017ee v p\u0159\u00edpad\u011b Hive mus\u00ed MDW p\u0159istoupit k p\u0159eps\u00e1n\u00ed cel\u00e9 tabulky, proto\u017ee Hive nem\u00e1 funkci aktualizace. A nebylo vymysleno nic lep\u0161\u00edho ne\u017e \u00fapln\u00e9 p\u0159eps\u00e1n\u00ed dat p\u0159i aktualizaci. V p\u0159\u00edpad\u011b RDBMS v\u0161ak tv\u016frci produktu pova\u017eovali za pot\u0159ebn\u00e9 sv\u011b\u0159it spojen\u00ed a aktualizaci tabulek pou\u017e\u00edv\u00e1n\u00ed SQL.<\/p>\n
Pro projekt ve Sberbanku jsme vytvo\u0159ili novou opakovan\u011b pou\u017eitelnou implementaci datab\u00e1zov\u00e9ho loaderu pro GreenPlum. To bylo provedeno na z\u00e1klad\u011b verze, kterou MDW generuje pro Teradata. Pr\u00e1v\u011b Teradata, a ne Oracle, byla pro toto lep\u0161\u00ed a nejbl\u00ed\u017ee, proto\u017ee je tak\u00e9 MPP syst\u00e9mem. Zp\u016fsoby pr\u00e1ce a tak\u00e9 syntaxe Teradata a GreenPlum se uk\u00e1zaly jako podobn\u00e9.<\/p>\n
P\u0159\u00edklady kritick\u00fdch rozd\u00edl\u016f mezi r\u016fzn\u00fdmi RDBMS pro MDW jsou n\u00e1sleduj\u00edc\u00ed. V GreenPlum, na rozd\u00edl od Teradata, p\u0159i vytv\u00e1\u0159en\u00ed tabulek je t\u0159eba napsat klauzuli<\/p>\n
distributed by<\/code><\/pre>\n\nV Teradata se p\u00ed\u0161e<\/p>\n
odstranit <table> all<\/code><\/pre>\n\n, a v GreenPlum se p\u00ed\u0161e<\/p>\n
smazat z <table><\/code><\/pre>\n\nV Oracle se z d\u016fvod\u016f optimalizace p\u00ed\u0161e<\/p>\n
delete from t where rowid in ()<\/code><\/pre>\n
\n, a v Teradata a GreenPlum se p\u00ed\u0161e<\/p>\n
delete from t where exists (select * from delta where delta.pk=t.pk)<\/code><\/pre>\n\nJe\u0161t\u011b bychom m\u011bli poznamenat, \u017ee pro pr\u00e1ci Ab Initio s GreenPlum bylo nutn\u00e9 nainstalovat klienta GreenPlum na v\u0161echny uzly clusteru Ab Initio. To je proto, \u017ee jsme se p\u0159ipojili k GreenPlum sou\u010dasn\u011b ze v\u0161ech uzl\u016f na\u0161eho clusteru. A aby \u010dten\u00ed z GreenPlum bylo paraleln\u00ed a ka\u017ed\u00fd paraleln\u00ed tok Ab Initio \u010detl svou \u010d\u00e1st dat z GreenPlum, bylo t\u0159eba do sekce \u201ewhere\u201c SQL dotaz\u016f um\u00edstit pro Ab Initio srozumiteln\u00e9 konstrukce<\/p>\n
where ABLOCAL()<\/code><\/pre>\n\na definovat hodnotu t\u00e9to konstrukce, p\u0159i\u010dem\u017e transformaci \u010dtouc\u00ed z DB se zad\u00e1v\u00e1 parametr<\/p>\n
ablocal_expr=\u00abstring_concat("mod(t.", string_filter_out("{$TABLE_KEY}","{}"), ",", (decimal(3))(number_of_partitions()),")=", (decimal(3))(this_partition()))\u00bb<\/code><\/pre>\n\n, kter\u00e1 se zkompiluje na n\u011bco jako<\/p>\n
mod(sk,10)=3<\/code><\/pre>\n\n, tedy je t\u0159eba poskytnout GreenPlum explicitn\u00ed filtr pro ka\u017edou partition. Pro jin\u00e9 datab\u00e1ze (Teradata, Oracle) m\u016f\u017ee Ab Initio tuto paralelizaci prov\u00e9st automaticky.<\/p>\n
Srovn\u00e1vac\u00ed charakteristiky v\u00fdkonu Ab Initio p\u0159i pr\u00e1ci s Hive a GreenPlum<\/h3>\n \nVe Sberbank byl proveden experiment porovn\u00e1n\u00ed v\u00fdkonu generovan\u00fdch MDW graf\u016f aplikovateln\u00fdch na Hive a GreenPlum. V r\u00e1mci experimentu m\u011bl Hive 5 uzl\u016f na tom sam\u00e9m clusteru jako Ab Initio, zat\u00edmco GreenPlum m\u011bl 4 uzly na samostatn\u00e9m clusteru. Tedy Hive m\u011bl ur\u010ditou v\u00fdhodu nad GreenPlum \u201ez hlediska hardwaru\u201c.<\/p>\n
Byly zv\u00e1\u017eeny dva p\u00e1ry graf\u016f, kter\u00e9 vykon\u00e1valy stejn\u00fd \u00fakol aktualizace dat v Hive a GreenPlum. Spou\u0161t\u011bly se grafy generovan\u00e9 konfigur\u00e1torem MDW:<\/p>\n
\ninicializa\u010dn\u00ed na\u010d\u00edt\u00e1n\u00ed + inkrement\u00e1ln\u00ed na\u010d\u00edt\u00e1n\u00ed n\u00e1hodn\u011b generovan\u00fdch dat do tabulky Hive<\/li>\n inicializa\u010dn\u00ed na\u010d\u00edt\u00e1n\u00ed + inkrement\u00e1ln\u00ed na\u010d\u00edt\u00e1n\u00ed n\u00e1hodn\u011b generovan\u00fdch dat do stejn\u00e9 tabulky GreenPlum<\/li>\n<\/ul>\n\nVe v\u0161ech p\u0159\u00edpadech (Hive a GreenPlum) byly spu\u0161t\u011bny na\u010d\u00edt\u00e1n\u00ed ve 10 paraleln\u00edch vl\u00e1knech na stejn\u00e9 klastru Ab Initio. Meziprodukty pro v\u00fdpo\u010dty byly ulo\u017eeny v HDFS (v term\u00ednech Ab Initio byl pou\u017eit MFS layout using HDFS). Jedna \u0159\u00e1dka n\u00e1hodn\u011b generovan\u00fdch dat zab\u00edrala v obou p\u0159\u00edpadech 200 byt\u016f.<\/p>\n
V\u00fdsledek byl n\u00e1sleduj\u00edc\u00ed:<\/p>\n
Hive:<\/b><\/p>\n
Inicia\u010dn\u00ed na\u010d\u00edt\u00e1n\u00ed v Hive<\/b><\/p>\n
Vlo\u017eeno \u0159\u00e1dk\u016f
D\u00e9lka inicia\u010dn\u00edho
Inkrement\u00e1ln\u00ed na\u010d\u00edt\u00e1n\u00ed v Hive<\/b><\/p>\n
Po\u010det \u0159\u00e1dk\u016f, kter\u00e9 byly v
Po\u010det \u0159\u00e1dk\u016f delta, kter\u00e9 byly pou\u017eity v
D\u00e9lka inkrement\u00e1ln\u00edho2 541<\/b><\/p>\n
\nGreenPlum:<\/b><\/p>\n
Inicia\u010dn\u00ed na\u010d\u00edt\u00e1n\u00ed v GreenPlum<\/b><\/p>\n
Vlo\u017eeno \u0159\u00e1dk\u016f
D\u00e9lka inicia\u010dn\u00edho
Inkrement\u00e1ln\u00ed na\u010d\u00edt\u00e1n\u00ed v GreenPlum<\/b><\/p>\n
Po\u010det \u0159\u00e1dk\u016f, kter\u00e9 byly v
Po\u010det \u0159\u00e1dk\u016f delta, kter\u00e9 byly pou\u017eity v
D\u00e9lka inkrement\u00e1ln\u00edho321<\/b><\/p>\n
\nVid\u00edme, \u017ee rychlost inicia\u010dn\u00edho na\u010d\u00edt\u00e1n\u00ed jak v Hive, tak v GreenPlum line\u00e1rn\u011b z\u00e1vis\u00ed na objemu dat, a z d\u016fvod\u016f lep\u0161\u00edho hardwaru je m\u00edrn\u011b rychlej\u0161\u00ed pro Hive ne\u017e pro GreenPlum.<\/p>\n
Inkrement\u00e1ln\u00ed na\u010d\u00edt\u00e1n\u00ed v Hive tak\u00e9 line\u00e1rn\u011b z\u00e1vis\u00ed na objemu p\u0159edem nahran\u00fdch dat v c\u00edlov\u00e9 tabulce a prob\u00edh\u00e1 pom\u011brn\u011b pomalu se zvy\u0161uj\u00edc\u00edmi se objemy. To je zp\u016fsobeno nutnost\u00ed kompletn\u011b p\u0159eepisovat c\u00edlovou tabulku. To znamen\u00e1, \u017ee uplatn\u011bn\u00ed mal\u00fdch zm\u011bn na obrovsk\u00fdch tabulk\u00e1ch nen\u00ed pro Hive p\u0159\u00edli\u0161 dobr\u00e1 praxe.<\/p>\n
Inkrement\u00e1ln\u00ed na\u010d\u00edt\u00e1n\u00ed v GreenPlum m\u00e1 slabou z\u00e1vislost na objemu p\u0159edem nahran\u00fdch dat v c\u00edlov\u00e9 tabulce a prob\u00edh\u00e1 dostate\u010dn\u011b rychle. K tomu do\u0161lo d\u00edky SQL Joins a architektu\u0159e GreenPlum, kter\u00e1 umo\u017e\u0148uje operaci delete.<\/p>\n
Tak\u017ee GreenPlum vkl\u00e1d\u00e1 delta metodou delete+insert, a v Hive nejsou operace delete ani update, a proto byl cel\u00fd datov\u00fd blok p\u0159i inkrement\u00e1ln\u00ed aktualizaci nucen b\u00fdt p\u0159eps\u00e1n. Nejv\u00fdznamn\u011bj\u0161\u00ed je srovn\u00e1n\u00ed tu\u010dn\u011b vyzna\u010den\u00fdch bun\u011bk, proto\u017ee odpov\u00eddaj\u00ed nej\u010dast\u011bj\u0161\u00edmu zp\u016fsobu vyu\u017eit\u00ed n\u00e1ro\u010dn\u00fdch na\u010d\u00edt\u00e1n\u00ed. Vid\u00edme, \u017ee GreenPlum v tomto testu vyhr\u00e1l nad Hive 8x.<\/p>\n
Pr\u00e1ce Ab Initio s GreenPlum v re\u017eimu Near Real Time<\/h3>\n \nV tomto experimentu ov\u011b\u0159\u00edme, zda Ab Initio m\u016f\u017ee prov\u00e1d\u011bt aktualizaci tabulky GreenPlum n\u00e1hodn\u011b generovan\u00fdmi d\u00e1vkami dat v re\u017eimu bl\u00edzk\u00e9m k re\u00e1ln\u00e9mu \u010dasu. Pod\u00edv\u00e1me se na tabulku GreenPlum dev42_1_db_usl.TESTING_SUBJ_org_finval, se kterou se bude pracovat.<\/p>\n
Budeme pou\u017e\u00edvat t\u0159i grafy Ab Initio k pr\u00e1ci s n\u00ed:<\/p>\n
1) Graf Create_test_data.mp \u2013 vytv\u00e1\u0159\u00ed v 10 paraleln\u00edch vl\u00e1knech soubory s daty v HDFS s 6 000 000 \u0159\u00e1dky. Data jsou n\u00e1hodn\u00e1, jejich struktura je organizov\u00e1na pro vlo\u017een\u00ed do na\u0161\u00ed tabulky.<\/p>\n
\nvygenerovat 6 000 000 testovac\u00edch \u0159\u00e1dk\u016f.<\/li>\n prov\u00e9st inicializa\u010dn\u00ed na\u010dten\u00ed vlo\u017een\u00edm 6 000 000 testovac\u00edch \u0159\u00e1dk\u016f do pr\u00e1zdn\u00e9 tabulky.<\/li>\n opakovat 5kr\u00e1t inkrement\u00e1ln\u00ed na\u010dten\u00ed.\n\nvygenerovat 6 000 000 testovac\u00edch \u0159\u00e1dk\u016f.<\/li>\n prov\u00e9st inkrement\u00e1ln\u00ed vlo\u017een\u00ed 6 000 000 testovac\u00edch \u0159\u00e1dk\u016f do tabulky (star\u00fdm dat\u016fm se nastav\u00ed \u010das vypr\u0161en\u00ed platnosti valid_to_ts a vkl\u00e1daj\u00ed se nov\u011bj\u0161\u00ed data se stejn\u00fdm prim\u00e1rn\u00edm kl\u00ed\u010dem).<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n\nTakov\u00fd sc\u00e9n\u00e1\u0159 emuluje re\u017eim skute\u010dn\u00e9 pr\u00e1ce n\u011bjak\u00e9ho podnikov\u00e9 syst\u00e9mu \u2013 v re\u00e1ln\u00e9m \u010dase se objev\u00ed dostate\u010dn\u011b objemn\u00e1 porce nov\u00fdch dat a okam\u017eit\u011b se integruje do GreenPlum.<\/p>\n
Nyn\u00ed se pod\u00edvejme na log pr\u00e1ce sc\u00e9n\u00e1\u0159e:<\/p>\n
Start Create_test_data.input.pset at 2020-06-04 11:49:11
Vypad\u00e1 to takto:<\/p>\n
Graf
Vytvo\u0159te_test_data.input.pset
mdw_load.day_one.current.
Vytvo\u0159te_test_data.input.pset
mdw_load.regular.current.
Vytvo\u0159te_test_data.input.pset
mdw_load.regular.current.
Vytvo\u0159te_test_data.input.pset
mdw_load.regular.current.
Vytvo\u0159te_test_data.input.pset
mdw_load.regular.current.
Vytvo\u0159te_test_data.input.pset
mdw_load.regular.current.
\nVid\u00edme, \u017ee 6 000 000 \u0159\u00e1dk\u016f inkrementu je zpracov\u00e1no za 3 minuty, co\u017e je dost rychl\u00e9.
select valid_from_ts, valid_to_ts, count(1), min(sk), max(sk) from dev42_1_db_usl.TESTING_SUBJ_org_finval group by valid_from_ts, valid_to_ts order by 1,2;<\/code><\/pre>\n\n
Z\u00e1v\u011br<\/h2>\n \nNyn\u00ed se Ab Initio pou\u017e\u00edv\u00e1 ve Sberbank pro vytvo\u0159en\u00ed Jednotn\u00e9ho s\u00e9mantick\u00e9ho datov\u00e9ho vrstvy (ESS). Tento projekt zahrnuje vybudov\u00e1n\u00ed jedin\u00e9 verze stavu r\u016fzn\u00fdch bankovn\u00edch obchodn\u00edch entit. Informace p\u0159ich\u00e1zej\u00ed z r\u016fzn\u00fdch zdroj\u016f, jejich\u017e repliky jsou p\u0159ipravov\u00e1ny na Hadoop. Na z\u00e1klad\u011b pot\u0159eb podnik\u00e1n\u00ed se p\u0159iprav\u00ed datov\u00fd model a popisuj\u00ed se transformace dat. Ab Initio nahr\u00e1v\u00e1 informace do ESS a nahran\u00e1 data nejen\u017ee sama o sob\u011b p\u0159edstavuj\u00ed z\u00e1jem pro podnik\u00e1n\u00ed, ale tak\u00e9 slou\u017e\u00ed jako zdroj pro vytv\u00e1\u0159en\u00ed datov\u00fdch vitr\u00edn. P\u0159itom funkce produktu umo\u017e\u0148uje pou\u017e\u00edvat jako p\u0159\u00edjemce r\u016fzn\u00e9 syst\u00e9my (Hive, Greenplum, Teradata, Oracle), co\u017e umo\u017e\u0148uje bez v\u011bt\u0161\u00edch obt\u00ed\u017e\u00ed p\u0159ipravit data pro podnik\u00e1n\u00ed v r\u016fzn\u00fdch form\u00e1tech, kter\u00e9 pot\u0159ebuje.<\/p>\n
Mo\u017enosti Ab Initio jsou \u0161irok\u00e9, nap\u0159\u00edklad p\u0159ilo\u017een\u00fd r\u00e1mec MDW umo\u017e\u0148uje vytv\u00e1\u0159et technickou a obchodn\u00ed historii dat \u201ez krabice\u201c. Pro v\u00fdvoj\u00e1\u0159e Ab Initio umo\u017e\u0148uje \u201eneobjevovat kolo\u201c a vyu\u017e\u00edvat mno\u017estv\u00ed existuj\u00edc\u00edch funk\u010dn\u00edch komponent\u016f, kter\u00e9 v podstat\u011b p\u0159edstavuj\u00ed knihovny pot\u0159ebn\u00e9 p\u0159i pr\u00e1ci s daty.<\/p>\n
Autor \u2014 expert profesion\u00e1ln\u00ed komunity Sberbank SberProfi DWH\/BigData. Profesion\u00e1ln\u00ed komunita SberProfi DWH\/BigData je odpov\u011bdn\u00e1 za rozvoj kompetenc\u00ed v oblastech jako ekosyst\u00e9m Hadoop, Teradata, Oracle DB, GreenPlum, a tak\u00e9 BI n\u00e1stroje Qlik, SAP BO, Tableau a dal\u0161\u00ed.<\/p><\/blockquote>\n
Zdroj: habr.com<\/a> <\/p>","protected":false,"gt_translate_keys":[{"key":"rendered","format":"html"}]},"excerpt":{"rendered":"\u041d\u0435\u043a\u043e\u0442\u043e\u0440\u043e\u0435 \u0432\u0440\u0435\u043c\u044f \u043d\u0430\u0437\u0430\u0434 \u043f\u0435\u0440\u0435\u0434 \u043d\u0430\u043c\u0438 \u0432\u0441\u0442\u0430\u043b \u0432\u043e\u043f\u0440\u043e\u0441 \u0432\u044b\u0431\u043e\u0440\u0430 ETL-\u0441\u0440\u0435\u0434\u0441\u0442\u0432\u0430 \u0434\u043b\u044f \u0440\u0430\u0431\u043e\u0442\u044b \u0441 BigData. \u0420\u0430\u043d\u0435\u0435 \u0438\u0441\u043f\u043e\u043b\u044c\u0437\u043e\u0432\u0430\u0432\u0448\u0435\u0435\u0441\u044f \u0440\u0435\u0448\u0435\u043d\u0438\u0435 Informatica BDM \u043d\u0435 \u0443\u0441\u0442\u0440\u0430\u0438\u0432\u0430\u043b\u043e \u043d\u0430\u0441 \u0438\u0437-\u0437\u0430 \u043e\u0433\u0440\u0430\u043d\u0438\u0447\u0435\u043d\u043d\u043e\u0439 \u0444\u0443\u043d\u043a\u0446\u0438\u043e\u043d\u0430\u043b\u044c\u043d\u043e\u0441\u0442\u0438. \u0415\u0451 \u0438\u0441\u043f\u043e\u043b\u044c\u0437\u043e\u0432\u0430\u043d\u0438\u0435 \u0441\u0432\u0435\u043b\u043e\u0441\u044c \u043a \u0444\u0440\u0435\u0439\u043c\u0432\u043e\u0440\u043a\u0443 \u043f\u043e \u0437\u0430\u043f\u0443\u0441\u043a\u0443 \u043a\u043e\u043c\u0430\u043d\u0434 spark-submit. \u041d\u0430 \u0440\u044b\u043d\u043a\u0435 \u0438\u043c\u0435\u043b\u043e\u0441\u044c \u043d\u0435 \u0442\u0430\u043a \u043c\u043d\u043e\u0433\u043e \u0430\u043d\u0430\u043b\u043e\u0433\u043e\u0432, \u0432 \u043f\u0440\u0438\u043d\u0446\u0438\u043f\u0435 \u0441\u043f\u043e\u0441\u043e\u0431\u043d\u044b\u0445 \u0440\u0430\u0431\u043e\u0442\u0430\u0442\u044c \u0441 \u0442\u0435\u043c \u043e\u0431\u044a\u0451\u043c\u043e\u043c \u0434\u0430\u043d\u043d\u044b\u0445, \u0441 \u043a\u043e\u0442\u043e\u0440\u044b\u043c \u043c\u044b \u0438\u043c\u0435\u0435\u043c \u0434\u0435\u043b\u043e \u043a\u0430\u0436\u0434\u044b\u0439 \u0434\u0435\u043d\u044c. […]<\/p>\n","protected":false,"gt_translate_keys":[{"key":"rendered","format":"html"}]},"author":1,"featured_media":87610,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[688],"tags":[],"class_list":["post-87609","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-administrirovanie"],"aioseo_notices":[],"aioseo_head":"\n\t\t\n\t