Konspekt · Ma'lumot va saqlash
Sharding
Muammo
Chat servisining xabarlar jadvali 6 TB ga yetadi va bitta server uni ko'tara olmaydi. Uni shard'larga bo'lamiz, shard kalitini tanlaymiz, qizib ketgan shard va qizib ketgan kalitga duch kelamiz va hamma narsani ko'chirmasdan to'rtinchi shard qo'shamiz.
Nimani o'rganasiz
- Oraliq (range) va hash bo'yicha sharding farqi, nega vaqt oralig'i qizigan shard yaratadi
- Shard kalitini qanday tanlash: eng ko'p ishlatiladigan so'rov hal qiladi
- Bir nechta shard'ga ketadigan so'rov narxi va qizigan kalitni qanday bo'lish
- Nega
hash % nqayta shard'lashni og'riqli qiladi va qat'iy slotlar buni qanday tuzatadi
Diagramma kod sifatida
Darsdagi yakuniy sxemani matn ko'rinishida oling va Jira, GitLab, GitHub, Notion yoki hujjatga qo'ying: Mermaid u yerlarda o'zi chiziladi. D2 va PlantUML (Confluence, IntelliJ) uchun alohida variantlar bor.
Mermaid: sxema
flowchart LR
n_app(["Chat ilovasi"])
n_api["Chat API"]
n_shard_0[("Shard 0")]
n_shard_1[("Shard 1")]
n_shard_2[("Shard 2")]
n_shard_3[("Shard 3")]
n_mover("Ma'lumot ko'chiruvchi")
n_app --> n_api
n_api --> n_shard_0
n_api --> n_shard_1
n_api --> n_shard_2
n_api --> n_shard_3
n_shard_2 --> n_mover
n_mover --> n_shard_3Mermaid: so'rovlar ketma-ketligi
sequenceDiagram
participant n_api as Chat API
participant n_db as Xabarlar bazasi
participant n_shard_2 as Shard 2
participant n_shard_1 as Shard 1
participant n_shard_0 as Shard 0
participant n_mover as Ma'lumot ko'chiruvchi
participant n_shard_3 as Shard 3
Note over n_api,n_db: Bitta server, 6 TB xabar
n_api->>n_db: INSERT INTO messages (chat_id, sender_id, body)
n_db-->>n_api: INSERT 0 1
Note over n_api,n_shard_2: Sana oralig'i bo'yicha bo'lish
n_api->>n_shard_2: route INSERT messages
n_shard_2-->>n_api: key: created_at = 2026-11-18
Note over n_api,n_shard_2: Har yangi xabar bitta shard'ga uriladi
n_api->>n_shard_2: INSERT INTO messages (chat_id, sender_id, body)
n_shard_2-->>n_api: ERROR: canceling statement due to statement timeout
Note over n_api,n_shard_2: chat_id hash'i bo'yicha shard'lash
n_api->>n_shard_2: route INSERT messages
n_shard_2-->>n_api: key: chat_id = c_8812
Note over n_api,n_shard_2: Shard kalitisiz so'rov hamma shard'dan so'raydi
n_api->>n_shard_2: query messages WHERE sender_id = 'u_42'
n_shard_2-->>n_api: -#62; shard-0: 20 rows, 41 ms
Note over n_api,n_shard_1: Bitta ulkan kanal bitta shard'ni qizdiradi
n_api->>n_shard_1: top keys, last 5 min (shard-1)
n_shard_1-->>n_api: chat_id=c_1 41% of writes
Note over n_api,n_shard_0: Qizigan chatni bucket'larga bo'lish
n_api->>n_shard_0: route INSERT messages chat_id=c_1
n_shard_0-->>n_api: hot chat: 8 buckets
Note over n_mover,n_shard_3: To'rtinchi shard, % 4 esa qatorlarning 75% ini ko'chiradi
n_mover->>n_shard_3: reshard: hash % 3 -#62; hash % 4
n_shard_3-->>n_mover: rows to check: 6.9B
Note over n_mover,n_shard_3: Qat'iy slotlar: faqat yangi shard oladiganini ko'chir
n_mover->>n_shard_3: move slot 517: shard-2 -#62; shard-3
n_shard_3-->>n_mover: copy: 6.8M rowsD2
direction: right
n_app: "Chat ilovasi" {
shape: rectangle
}
n_api: "Chat API" {
shape: rectangle
}
n_shard_0: "Shard 0" {
shape: cylinder
}
n_shard_1: "Shard 1" {
shape: cylinder
}
n_shard_2: "Shard 2" {
shape: cylinder
}
n_shard_3: "Shard 3" {
shape: cylinder
}
n_mover: "Ma'lumot ko'chiruvchi" {
shape: rectangle
style.multiple: true
}
n_app -> n_api
n_api -> n_shard_0
n_api -> n_shard_1
n_api -> n_shard_2
n_api -> n_shard_3
n_shard_2 -> n_mover
n_mover -> n_shard_3PlantUML: so'rovlar ketma-ketligi
@startuml
participant "Chat API" as n_api
database "Xabarlar bazasi" as n_db
database "Shard 2" as n_shard_2
database "Shard 1" as n_shard_1
database "Shard 0" as n_shard_0
participant "Ma'lumot ko'chiruvchi" as n_mover
database "Shard 3" as n_shard_3
note over n_api, n_db : Bitta server, 6 TB xabar
n_api -> n_db : INSERT INTO messages (chat_id, sender_id, body)
n_db --> n_api : INSERT 0 1
note over n_api, n_shard_2 : Sana oralig'i bo'yicha bo'lish
n_api -> n_shard_2 : route INSERT messages
n_shard_2 --> n_api : key: created_at = 2026-11-18
note over n_api, n_shard_2 : Har yangi xabar bitta shard'ga uriladi
n_api -> n_shard_2 : INSERT INTO messages (chat_id, sender_id, body)
n_shard_2 --> n_api : ERROR: canceling statement due to statement timeout
note over n_api, n_shard_2 : chat_id hash'i bo'yicha shard'lash
n_api -> n_shard_2 : route INSERT messages
n_shard_2 --> n_api : key: chat_id = c_8812
note over n_api, n_shard_2 : Shard kalitisiz so'rov hamma shard'dan so'raydi
n_api -> n_shard_2 : [dublikat] query messages WHERE sender_id = 'u_42'
n_shard_2 --> n_api : -> shard-0: 20 rows, 41 ms
note over n_api, n_shard_1 : Bitta ulkan kanal bitta shard'ni qizdiradi
n_api -> n_shard_1 : top keys, last 5 min (shard-1)
n_shard_1 --> n_api : chat_id=c_1 41% of writes
note over n_api, n_shard_0 : Qizigan chatni bucket'larga bo'lish
n_api -> n_shard_0 : route INSERT messages chat_id=c_1
n_shard_0 --> n_api : hot chat: 8 buckets
note over n_mover, n_shard_3 : To'rtinchi shard, % 4 esa qatorlarning 75% ini ko'chiradi
n_mover -> n_shard_3 : reshard: hash % 3 -> hash % 4
n_shard_3 --> n_mover : rows to check: 6.9B
note over n_mover, n_shard_3 : Qat'iy slotlar: faqat yangi shard oladiganini ko'chir
n_mover -> n_shard_3 : move slot 517: shard-2 -> shard-3
n_shard_3 --> n_mover : copy: 6.8M rows
@endumlPlantUML: komponentlar
@startuml
left to right direction
agent "Chat ilovasi" as n_app
component "Chat API" as n_api
database "Shard 0" as n_shard_0
database "Shard 1" as n_shard_1
database "Shard 2" as n_shard_2
database "Shard 3" as n_shard_3
collections "Ma'lumot ko'chiruvchi" as n_mover
n_app --> n_api
n_api --> n_shard_0
n_api --> n_shard_1
n_api --> n_shard_2
n_api --> n_shard_3
n_shard_2 --> n_mover
n_mover --> n_shard_3
@endumlXulosa
Sharding bitta serverga sig'may qolgan jadvalni bir nechta serverga taqsimlaydi. Hammasi shard kalitiga bog'liq.
1) Eng ko'p so'rov filtrlaydigan kalitni tanlang. 2) Vaqt bilan o'sadigan kalitlardan qoching: ularni hash qiling. 3) Kalitsiz so'rov hamma shard'ga tarqaladi; ularni kam saqlang. 4) Qizigan kalitlarni kuzating va bo'ling. 5) Qat'iy slotlar orqali yo'naltiring, shunda shard qo'shish faqat uning ulushini ko'chiradi.
Ko'rsatkichlar: boshida va oxirida
| Ko'rsatkich | Boshida | Oxirida |
|---|---|---|
| So'rov p95 | 1 400 ms | 28 ms |
| Eng qizigan shard yuki | 98% | 42% |
| Bir so'rovdagi shard'lar | 1 | 1 |
| Eng katta shard | 6 TB | 2,0 TB |
Qadamlar
- Muammo Bitta server, 6 TB xabar
Har chat xabari bitta
messagesjadvaliga yoziladi. - Yechim Sana oralig'i bo'yicha bo'lish
Jadvalni uchta shard'ga — har biri alohida baza — bo'lamiz: 2024 shard 0 da, 2025 shard 1 da, 2026 shard 2 da.
- Muammo Har yangi xabar bitta shard'ga uriladi
Yangi xabarlarning hammasi bugungi, shuning uchun har yozuv shard 2 ga tushadi.
- Yechim chat_id hash'i bo'yicha shard'lash
Yangi shard kaliti —
chat_id, shard esahash(chat_id) % 3. - Holat Shard kalitisiz so'rov hamma shard'dan so'raydi
Yangi funksiya: «men yuborgan barcha xabarlar».
- Muammo Bitta ulkan kanal bitta shard'ni qizdiradi
2 million a'zoli ochiq
c_1kanali mashhur bo'lib ketadi. - Yechim Qizigan chatni bucket'larga bo'lish
Qizigan deb belgilangan chatlar uchun kalit
(chat_id, bucket)bo'ladi, bucket esamsg_id % 8. - Muammo To'rtinchi shard,
% 4esa qatorlarning 75% ini ko'chiradiShard'lar 2,6 TB gacha o'sdi, shard 3 ni qo'shamiz.
- Yechim Qat'iy slotlar: faqat yangi shard oladiganini ko'chir
Ikki bosqichda yo'naltiramiz:
hash % 1024slotni tanlaydi, kichik xarita esa har slotni shard'ga biriktiradi. - Xulosa Xulosa
Sharding bitta serverga sig'may qolgan jadvalni bir nechta serverga taqsimlaydi.
Atamalar
- Sharding
- Bitta serverga sig'may qolgan jadvalni qatorlar bo'yicha bir nechta serverga (shard'ga) bo'lish. Har qator qayerda turishini shard kaliti hal qiladi.
- Shard kaliti
- Qator qaysi shard'da turishini hal qiladigan qiymat. Eng ko'p so'rov filtrlaydigan ustun tanlanadi, vaqt bilan o'sadigan kalit esa hash qilinadi, aks holda bitta shard qiziydi.
- Qizigan kalit (hot key)
- Yuki boshqalardan juda katta bo'lgan bitta kalit, masalan ulkan kanal yoki aksiyadagi mahsulot. U bitta shard yoki kesh node'ini qizdiradi, shuning uchun bucket'larga bo'linadi.