Konspekt · Ma'lumot va saqlash
Bazani replikatsiya qilish
Muammo
Profil servisi bitta bazaga sig'may qoladi. O'qish uchun replika qo'shamiz, replikatsiya kechikishi va «o'z yozganingni o'qi» muammosiga duch kelamiz, liderni yo'qotamiz, failover qilamiz va sinxron replikatsiya bunga arziydimi — hal qilamiz.
Nimani o'rganasiz
- Lider o'z logini follower'larga qanday uzatadi va replikatsiya kechikishi nima
- Nega foydalanuvchi o'z yozganini ko'rmay qoladi va buni tuzatishning ikki yo'li
- Failover nima qiladi va nega asinxron replikatsiya commit'larni yo'qotishi mumkin
- Sinxron replikaning narxi
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(["Profil ilovasi"])
n_api["Profil API"]
n_db_a[("Baza A")]
n_db_b[("Baza B")]
n_db_c[("Baza C")]
n_sentinel["Failover menejeri"]
n_app -->|"chaqiruv"| n_api
n_api -->|"chaqiruv"| n_db_b
n_api -->|"chaqiruv"| n_db_c
n_sentinel -->|"chaqiruv"| n_db_c
n_api -->|"chaqiruv"| n_db_c
n_db_c -->|"chaqiruv"| n_db_b
n_db_c -->|"asinxron"| n_db_aMermaid: so'rovlar ketma-ketligi
sequenceDiagram
participant n_api as Profil API
participant n_db_a as Baza A
participant n_db_b as Baza B
participant n_sentinel as Failover menejeri
participant n_db_c as Baza C
Note over n_api,n_db_a: Bitta baza barcha o'qish va yozishni oladi
n_api->>n_db_a: SELECT name, bio FROM profiles
n_db_a-->>n_api: (1 row)
Note over n_db_a,n_db_b: Ikki follower lider logini nusxalaydi
n_db_a->>n_db_b: START_REPLICATION SLOT db_b PHYSICAL 3/A1000000
n_db_b-->>n_db_a: XLogData lsn=3/A1C04F28 len=8192
Note over n_api,n_db_b: Foydalanuvchi saqlaydi, yangilaydi va eski bio'ni ko'radi
n_api->>n_db_b: SELECT bio, updated_at FROM profiles
n_db_b-->>n_api: bio #124; updated_at
Note over n_api,n_db_a: Replikatsiya kechikishini o'lchash
n_api->>n_db_a: SELECT application_name, replay_lag
n_db_a-->>n_api: application_name #124; replay_lag
Note over n_api,n_db_b: Read-your-writes: foydalanuvchining oxirgi yozuviga qarab yo'naltirish
n_api->>n_db_b: SELECT pg_last_wal_replay_lsn()
n_db_b-->>n_api: caught_up
Note over n_api,n_db_a: Lider ishdan chiqadi
n_api->>n_db_a: UPDATE profiles SET bio = 'Backend + SRE'
n_db_a-->>n_api: ERROR: could not connect to db-a:5432
Note over n_sentinel,n_db_c: Failover: eng yangi replikani ko'tarish
n_sentinel->>n_db_c: failover: db-a down for 30 s
Note over n_db_c,n_db_a: Eski lider follower bo'lib qaytadi
n_db_c->>n_db_a: pg_rewind --source-server=db-c
n_db_a-->>n_db_c: source timeline 8, target timeline 7
Note over n_api,n_db_c: Bitta sinxron replika: tasdiqlangan yozuv yo'qolmaydi
n_api->>n_db_c: ALTER SYSTEM SET synchronous_standby_names
n_db_c-->>n_api: ALTER SYSTEMD2
direction: right
n_app: "Profil ilovasi" {
shape: rectangle
}
n_api: "Profil API" {
shape: rectangle
}
n_db_a: "Baza A" {
shape: cylinder
}
n_db_b: "Baza B" {
shape: cylinder
}
n_db_c: "Baza C" {
shape: cylinder
}
n_sentinel: "Failover menejeri" {
shape: rectangle
}
n_app -> n_api: "chaqiruv"
n_api -> n_db_b: "chaqiruv"
n_api -> n_db_c: "chaqiruv"
n_sentinel -> n_db_c: "chaqiruv"
n_api -> n_db_c: "chaqiruv"
n_db_c -> n_db_b: "chaqiruv"
n_db_c -> n_db_a: "asinxron"PlantUML: so'rovlar ketma-ketligi
@startuml
participant "Profil API" as n_api
database "Baza A" as n_db_a
database "Baza B" as n_db_b
participant "Failover menejeri" as n_sentinel
database "Baza C" as n_db_c
note over n_api, n_db_a : Bitta baza barcha o'qish va yozishni oladi
n_api -> n_db_a : SELECT name, bio FROM profiles
n_db_a --> n_api : (1 row)
note over n_db_a, n_db_b : Ikki follower lider logini nusxalaydi
n_db_a ->> n_db_b : START_REPLICATION SLOT db_b PHYSICAL 3/A1000000
n_db_b -->> n_db_a : XLogData lsn=3/A1C04F28 len=8192
note over n_api, n_db_b : Foydalanuvchi saqlaydi, yangilaydi va eski bio'ni ko'radi
n_api -> n_db_b : SELECT bio, updated_at FROM profiles
n_db_b --> n_api : bio | updated_at
note over n_api, n_db_a : Replikatsiya kechikishini o'lchash
n_api -> n_db_a : SELECT application_name, replay_lag
n_db_a --> n_api : application_name | replay_lag
note over n_api, n_db_b : Read-your-writes: foydalanuvchining oxirgi yozuviga qarab yo'naltirish
n_api -> n_db_b : SELECT pg_last_wal_replay_lsn()
n_db_b --> n_api : caught_up
note over n_api, n_db_a : Lider ishdan chiqadi
n_api -> n_db_a : UPDATE profiles SET bio = 'Backend + SRE'
n_db_a --> n_api : ERROR: could not connect to db-a:5432
note over n_sentinel, n_db_c : Failover: eng yangi replikani ko'tarish
n_sentinel -> n_db_c : failover: db-a down for 30 s
note over n_db_c, n_db_a : Eski lider follower bo'lib qaytadi
n_db_c ->> n_db_a : pg_rewind --source-server=db-c
n_db_a -->> n_db_c : source timeline 8, target timeline 7
note over n_api, n_db_c : Bitta sinxron replika: tasdiqlangan yozuv yo'qolmaydi
n_api -> n_db_c : ALTER SYSTEM SET synchronous_standby_names
n_db_c --> n_api : ALTER SYSTEM
@endumlPlantUML: komponentlar
@startuml
left to right direction
agent "Profil ilovasi" as n_app
component "Profil API" as n_api
database "Baza A" as n_db_a
database "Baza B" as n_db_b
database "Baza C" as n_db_c
component "Failover menejeri" as n_sentinel
n_app --> n_api : chaqiruv
n_api --> n_db_b : chaqiruv
n_api --> n_db_c : chaqiruv
n_sentinel --> n_db_c : chaqiruv
n_api --> n_db_c : chaqiruv
n_db_c --> n_db_b : chaqiruv
n_db_c --> n_db_a : asinxron
@endumlXulosa
Replikalar o'qish quvvatini va failover uchun zaxirani beradi. Yangi ma'lumotni bermaydi: asinxron replikalar kechikadi va kechikish yuk bilan o'sadi.
1) Yozish liderga, o'qish replikalarga. 2) Kechikishni o'lchang va yozuvdan so'ng foydalanuvchining o'qishini liderga yo'naltiring. 3) Failover eng yangi replikani ko'taradi va eski liderni izolyatsiya qilishi shart. 4) Asinxron rejim tasdiqlangan commit'larni yo'qotishi mumkin; bitta sinxron replika buni kechikish evaziga oldini oladi.
Ko'rsatkichlar: boshida va oxirida
| Ko'rsatkich | Boshida | Oxirida |
|---|---|---|
| So'rov p95 | 850 ms | 58 ms |
| Eng katta kechikish | 0 ms | 5 ms |
| Eskirgan o'qishlar | 0% | 0% |
| Failover'da yo'qolgan commit'lar | 0 | 0 |
Qadamlar
- Muammo Bitta baza barcha o'qish va yozishni oladi
Har profilni ko'rish va har tahrir bitta bazaga — Baza A ga boradi.
- Yechim Ikki follower lider logini nusxalaydi
Baza A lider bo'ladi: yozishni faqat u qabul qiladi.
- Muammo Foydalanuvchi saqlaydi, yangilaydi va eski bio'ni ko'radi
u_42foydalanuvchi bio'sini o'zgartiradi. - Holat Replikatsiya kechikishini o'lchash
Lider har follower qayergacha yetganini biladi.
- Yechim Read-your-writes: foydalanuvchining oxirgi yozuviga qarab yo'naltirish
Yozuvdan keyin API shu foydalanuvchi sessiyasi uchun lider logidagi joyni eslab qoladi.
- Muammo Lider ishdan chiqadi
Baza A ning diski buziladi.
- Yechim Failover: eng yangi replikani ko'tarish
Failover menejeri Baza A ni 30 soniya kuzatdi, javob yo'q.
- Holat Eski lider follower bo'lib qaytadi
Baza A ga yangi disk qo'yiladi va u ishga tushadi.
- Yechim Bitta sinxron replika: tasdiqlangan yozuv yo'qolmaydi
Liderni kutishga majbur qilamiz: commit faqat ikki follower'dan kamida bittasi logni diskka yozgandan keyin tasdiqlanadi.
- Xulosa Xulosa
Replikalar o'qish quvvatini va failover uchun zaxirani beradi.
Atamalar
- Replikatsiya (lider va follower)
- Lider barcha yozuvlarni qabul qiladi va o'z logini follower'larga uzatadi, ular uni qayta bajarib nusxa saqlaydi. Replikalar o'qish quvvatini va failover uchun zaxirani beradi.
- Replikatsiya kechikishi
- Asinxron replika liderdan qancha orqada ekani. Yuk ostida u o'sadi va replikadan o'qilgan ma'lumot eskirgan bo'lishi mumkin.
- Read-your-writes
- Foydalanuvchi o'zi yozgan narsani doim ko'rishi kafolati. Yozuvdan keyin uning o'qishlari replika yetib olguncha liderga yo'naltiriladi.