Yaya ake gyara karkatar bayanai da partitions masu nauyi wajen sharding?

Question

IT Interview Practice · Accepted Answer

**Da farko gano abin da bai daidaita ba: girman bayanan da aka adana, yawan requests, ko updates ga record ɗaya.** Ƙara servers yana taimako ne kawai idan ana iya mayar da aikin da ya yi yawa zuwa gare su.

A ce an raba oda bisa merchant ID. Yawancin merchants ƙanana ne, amma ɗaya yana gudanar da babban campaign. Shard nasa ya cika yayin da sauran ba su da aiki sosai. Wannan hot partition ne: partition da yake samun aikin da ya fi ƙarfinsa.

## 1. Auna rashin daidaito kafin sauya key

Kwatanta data size, request rate, latency da matsin CPU ko disk tsakanin shards. Sannan rarraba overloaded shard bisa merchant da operation. Shard mai tsofaffin records zai iya zama babba amma marar aiki; ƙaramin shard zai iya cika saboda updates akai-akai ga kaya ɗaya mai shahara.

Ga campaign ɗin, gano ko load yana fitowa daga karanta product details, ƙirƙirar independent orders da yawa, ko maimaita canjin inventory record ɗaya. Waɗannan uku suna buƙatar hanyoyin magancewa daban.

## 2. Fahimci dalilin da rarrabawar yanzu take ci gaba da karkata

Shard key yana yanke inda records za su je. Hashing merchant IDs zai iya rarraba merchants daban tsakanin shards, amma duk odar merchant key ɗaya tana bin placement rule ɗaya. Hashing ba ya raba load na merchant ɗaya kai tsaye.

Keys masu distinct values kaɗan suna takaita damar raba data. Time range na iya tattara sabbin writes a partition na baya-bayan nan. Key mai ƙarin warwatsewa na iya inganta distribution amma ya sa queries na yau da kullum su tuntuɓi shards da yawa. [Jagorar shard key ta MongoDB](https://www.mongodb.com/docs/manual/core/sharding-choose-a-shard-key/) ta bayyana waɗannan batutuwa.

## 3. Haɗa mafita da aikin da ya yi yawa

| Hujja | Hanyar da za a tantance | Muhimmiyar iyaka |
|---|---|---|
| Movable partitions da yawa a shard ɗaya | Rebalance su zuwa capacity da ake da ita | Motsa data yana amfani ne idan aikin da ya shafe ta ya motsa ma |
| Maimaita karanta kaya ɗaya | Cache ko read replicas idan freshness ta halatta | Wannan ba ya raba writes zuwa wannan kayan |
| Merchant ɗaya yana ƙirƙirar independent orders da yawa | Raba shi da ƙarin order-based key ko bucket | Queries ga merchant gaba ɗaya na iya buƙatar reads da yawa da merging |
| Writes da yawa suna gasa ga inventory record ɗaya | Rage contention ko sake tsara stock allocation | Ƙarin shards ba ya cire buƙatar daidaita correctness |

Ware babban merchant zai iya kare sauran tenants kuma ya ba da lokaci. Ba ya ba merchant ɗin capacity mara iyaka, saboda haka auna bottleneck nasa har yanzu.

## 4. Bincika farashin raba merchant

Ka yi tunanin routing oda da merchant ID tare da bucket da aka samo daga order ID. Independent orders na campaign za su iya zuwa partitions da yawa.

```text
Merchant A + bucket 0 -> Partition 1
Merchant A + bucket 1 -> Partition 2
Merchant A + bucket 2 -> Partition 3
```

Samun oda ɗaya ya fi sauƙi idan ana iya lissafa bucket ɗinta. Samun duk recent orders na Merchant A yanzu yana buƙatar karanta partitions da yawa da haɗa su cikin tsari. Pagination, totals da transactions a tsakaninsu suna ƙara wahala. Zaɓi wannan rabo bayan duba cewa ƙarin kuɗin reads da coordination ya halatta.

## 5. Yi migration da hanyar tabbatar da correctness

Sauya routing rule yayin da writes suke gudana yana buƙatar migration plan. Kwafi records, kama canje-canje yayin kwafi, sannan reconcile adadin records da business totals kafin sauya traffic. Counts kawai za su iya rasa order values marasa kyau.

Bayyana yadda za a kula da writes da suke gudana lokacin cutover da yadda za a dawo da routing ba tare da rasa writes da sabon tsarin ya karɓa ba. Dual writes suna buƙatar failure handling; rubutawa zuwa wurare biyu kawai ba ya tabbatar sun kasance iri ɗaya.

A ƙarshe, gwada da workload mai karkatar gaske, ciki har da babban merchant, ba traffic da aka rarraba daidai kawai ba. Nasara tana nufin rage matsin overloaded shard, query costs da suka halatta da business results marasa canji. **Key mai kyau yana daidaita workload distribution da yadda application yake karantawa da sabunta data.**

Hujja	Hanyar da za a tantance	Muhimmiyar iyaka
Movable partitions da yawa a shard ɗaya	Rebalance su zuwa capacity da ake da ita	Motsa data yana amfani ne idan aikin da ya shafe ta ya motsa ma
Maimaita karanta kaya ɗaya	Cache ko read replicas idan freshness ta halatta	Wannan ba ya raba writes zuwa wannan kayan
Merchant ɗaya yana ƙirƙirar independent orders da yawa	Raba shi da ƙarin order-based key ko bucket	Queries ga merchant gaba ɗaya na iya buƙatar reads da yawa da merging
Writes da yawa suna gasa ga inventory record ɗaya	Rage contention ko sake tsara stock allocation	Ƙarin shards ba ya cire buƙatar daidaita correctness

1. Auna rashin daidaito kafin sauya key

2. Fahimci dalilin da rarrabawar yanzu take ci gaba da karkata

3. Haɗa mafita da aikin da ya yi yawa

4. Bincika farashin raba merchant

5. Yi migration da hanyar tabbatar da correctness

Tambayoyi masu alaƙa