{"id":11,"date":"2026-09-23T20:36:59","date_gmt":"2026-09-23T20:36:59","guid":{"rendered":"https:\/\/www.akifsen.com.tr\/blog\/?p=11"},"modified":"2026-09-23T20:36:59","modified_gmt":"2026-09-23T20:36:59","slug":"redis-cache-aside-stampede-invalidation","status":"publish","type":"post","link":"https:\/\/www.akifsen.com.tr\/blog\/2026\/09\/23\/redis-cache-aside-stampede-invalidation\/","title":{"rendered":"Redis Eklemek Performans Problemini \u00c7\u00f6zmez: Cache Tasar\u0131m\u0131nda As\u0131l Zor K\u0131s\u0131m"},"content":{"rendered":"\n

Cache h\u0131zl\u0131d\u0131r. Cache tasar\u0131m\u0131 ise zordur.<\/h2>\n\n\n\n

Bir endpoint yava\u015flad\u0131\u011f\u0131nda en kolay \u00f6nerilerden biri \u015fudur:<\/p>\n\n\n\n

\n

Redis koyal\u0131m.<\/p>\n<\/blockquote>\n\n\n\n

Bazen ger\u00e7ekten do\u011fru \u00e7\u00f6z\u00fcmd\u00fcr. Ancak cache eklemek problemi \u00e7\u00f6zmek yerine sisteme yeni bir consistency problemi de ekleyebilir.<\/p>\n\n\n\n

\u00c7\u00fcnk\u00fc as\u0131l soru veriyi Redis’e nas\u0131l koyaca\u011f\u0131n\u0131z de\u011fil, hangi veriyi ne kadar s\u00fcreyle do\u011fru kabul edece\u011finizdir.<\/strong><\/p>\n\n\n\n

En s\u0131k kullan\u0131lan modellerden biri cache-aside’d\u0131r. Uygulama \u00f6nce Redis’e bakar. Veri yoksa ana veritaban\u0131ndan okur ve sonucu cache’e yazar. Redis’in resmi dok\u00fcmantasyonu da cache-aside modelini \u00f6zellikle tekrar eden read operasyonlar\u0131nda database y\u00fck\u00fcn\u00fc azaltmak i\u00e7in konumland\u0131r\u0131yor.<\/p>\n\n\n\n

Basit haliyle:<\/p>\n\n\n\n

GET \/products\/42\n\n1. Redis GET product:42\n2. Cache hit  \u2192 response\n3. Cache miss \u2192 PostgreSQL SELECT\n4. Redis SET product:42 EX 300\n5. response<\/code><\/pre>\n\n\n\n
Buraya kadar kolay.<\/p>\n\n\n\n
Sorun 300 saniye sonra ba\u015fl\u0131yor.<\/p>\n\n\n\n
Cache stampede<\/h2>\n\n\n\n
product:42<\/code> \u00e7ok pop\u00fcler bir kay\u0131t olsun.<\/p>\n\n\n\n
TTL doldu\u011fu anda ayn\u0131 milisaniyede 500 request gelirse 500 request de cache miss g\u00f6rebilir. Bu durumda tamam\u0131 database’e gider ve ayn\u0131 sorguyu \u00e7al\u0131\u015ft\u0131r\u0131r.<\/p>\n\n\n\n
Cache’in database’i korumas\u0131 gerekirken tam tersine y\u00fck patlamas\u0131n\u0131n sebebi olur.<\/p>\n\n\n\n
Redis dok\u00fcmantasyonu da pop\u00fcler bir key expire oldu\u011funda ayn\u0131 kayd\u0131 isteyen \u00e7ok say\u0131da process’in ayn\u0131 anda database’e y\u00f6nelmesini cache stampede olarak ele al\u0131yor.<\/p>\n\n\n\n
Bunu azaltmak i\u00e7in TTL jitter, request coalescing, distributed lock veya stale-while-revalidate gibi teknikler kullan\u0131labilir.<\/p>\n\n\n\n
\u00d6rne\u011fin her key’i tam 300 saniyede expire etmek yerine:<\/p>\n\n\n\n
long ttl = 300 + ThreadLocalRandom.current().nextLong(0, 30);<\/code><\/pre>\n\n\n\n
gibi k\u00fc\u00e7\u00fck bir da\u011f\u0131l\u0131m bile toplu expiration dalgalar\u0131n\u0131 azaltabilir.<\/p>\n\n\n\n
En zor problem: invalidation<\/h2>\n\n\n\n
Cache ile ana veri kayna\u011f\u0131 aras\u0131nda iki farkl\u0131 ger\u00e7eklik olu\u015fur.<\/p>\n\n\n\n
Bir kullan\u0131c\u0131 ad\u0131n\u0131 database’de de\u011fi\u015ftirdi\u011finiz halde Redis’teki veri be\u015f dakika daha eski kal\u0131yorsa sisteminiz teknik olarak h\u0131zl\u0131d\u0131r ama yanl\u0131\u015f veri d\u00f6nmektedir.<\/p>\n\n\n\n
Bu nedenle her cache i\u00e7in \u015fu sorunun cevab\u0131 verilmelidir:<\/p>\n\n\n\n
Bu verinin ka\u00e7 saniye eski olmas\u0131 kabul edilebilir?<\/strong><\/p>\n\n\n\n
\u00dcr\u00fcn katalogunda 30 saniye kabul edilebilir olabilir.<\/p>\n\n\n\n
Banka bakiyesinde olmayabilir.<\/p>\n\n\n\n
Authorization bilgisinde ise baz\u0131 durumlarda birka\u00e7 saniyelik stale veri bile g\u00fcvenlik problemi olu\u015fturabilir.<\/p>\n\n\n\n
Bu y\u00fczden TTL teknik bir parametre de\u011fil, asl\u0131nda i\u015f kural\u0131d\u0131r.<\/p>\n\n\n\n
Cache hit ratio tek ba\u015f\u0131na yeterli de\u011fildir<\/h2>\n\n\n\n
Cache sistemini de\u011ferlendirirken yaln\u0131zca hit ratio’ya bakmak yan\u0131lt\u0131c\u0131 olabilir.<\/p>\n\n\n\n
%99 hit ratio \u00e7ok iyi g\u00f6r\u00fcnebilir. Ama kalan %1 miss database’i \u00e7\u00f6kerten birka\u00e7 hot key’den olu\u015fuyorsa sistem yine sorunludur.<\/p>\n\n\n\n
Ben a\u015fa\u011f\u0131daki sinyalleri birlikte izlerdim:<\/p>\n\n\n\n
cache hit\/miss oran\u0131, P95\/P99 cache latency, database query hacmi, hot key da\u011f\u0131l\u0131m\u0131, eviction say\u0131s\u0131 ve expiration sonras\u0131 olu\u015fan y\u00fck.<\/p>\n\n\n\n
Cache optimizasyonunun ba\u015far\u0131s\u0131 \u201cRedis \u00e7al\u0131\u015f\u0131yor\u201d de\u011fildir.<\/p>\n\n\n\n
Ba\u015far\u0131l\u0131 cache tasar\u0131m\u0131, Redis devre d\u0131\u015f\u0131 kald\u0131\u011f\u0131nda veya bir anda y\u00fczlerce key expire oldu\u011funda sistemin nas\u0131l davranaca\u011f\u0131n\u0131 da bilmektir.<\/p>\n\n\n\n
\u00c7\u00fcnk\u00fc cache performans katman\u0131d\u0131r.<\/p>\n\n\n\n
Ana veri kayna\u011f\u0131 de\u011fildir.<\/strong><\/p>\n","protected":false},"excerpt":{"rendered":"
Cache h\u0131zl\u0131d\u0131r. Cache tasar\u0131m\u0131 ise zordur. Bir endpoint yava\u015flad\u0131\u011f\u0131nda en kolay \u00f6nerilerden biri \u015fudur: Redis koyal\u0131m. Bazen ger\u00e7ekten do\u011fru \u00e7\u00f6z\u00fcmd\u00fcr. Ancak cache eklemek problemi \u00e7\u00f6zmek yerine sisteme yeni bir consistency problemi de ekleyebilir. \u00c7\u00fcnk\u00fc as\u0131l soru veriyi Redis’e nas\u0131l koyaca\u011f\u0131n\u0131z de\u011fil, hangi veriyi ne kadar s\u00fcreyle do\u011fru kabul edece\u011finizdir. En s\u0131k kullan\u0131lan modellerden biri cache-aside’d\u0131r. […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[6],"tags":[],"class_list":["post-11","post","type-post","status-publish","format-standard","hentry","category-mimari"],"_links":{"self":[{"href":"https:\/\/www.akifsen.com.tr\/blog\/wp-json\/wp\/v2\/posts\/11","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.akifsen.com.tr\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.akifsen.com.tr\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.akifsen.com.tr\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.akifsen.com.tr\/blog\/wp-json\/wp\/v2\/comments?post=11"}],"version-history":[{"count":1,"href":"https:\/\/www.akifsen.com.tr\/blog\/wp-json\/wp\/v2\/posts\/11\/revisions"}],"predecessor-version":[{"id":12,"href":"https:\/\/www.akifsen.com.tr\/blog\/wp-json\/wp\/v2\/posts\/11\/revisions\/12"}],"wp:attachment":[{"href":"https:\/\/www.akifsen.com.tr\/blog\/wp-json\/wp\/v2\/media?parent=11"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.akifsen.com.tr\/blog\/wp-json\/wp\/v2\/categories?post=11"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.akifsen.com.tr\/blog\/wp-json\/wp\/v2\/tags?post=11"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}