# code tpl

{%hackmd theme-dark %}

E4
do things right

E5 trade-offs (

SQL / NOSQL,

core puzzle,

push pull,

REST/GraphQL/GRPC,

sync or async ( message queue→ also decoupe

bandwidth → use websockey instead

E6 go deep ( offline support, multi-language support, battrey optimal

E7
System Design: Log exchange event for on-premise analysis or ML

purpose:
1. us eu trading log -> us or eu server -> TW

usecase:
1. write log to local server
2. send back to main server (TW)

metric:
every day log -> 100 milion

RPS:
1157.40740741 request / second
- LB server number -> scale out
- scale up -> CPU / MEM-> base on statistics

storage:
1 log 1KB -> 100 million -> 100GB / day
- database / plan for storage

bandwidth : upload -> 100GB / 86400 -> 1MB / s
- ISP network

database
- log
- write perfer
- time series
- cassandra DB

user produce log -> [LB] -> (log service) -> [cassandra DB]
|
relica to main server -> TW [cassandra]
|
ML analyisis

usecase:
1. write log to local server
- user log -> hit LB
- LB -> dispatch log service
- log service wirte to cassandra

2. send back to main server (TW)
- regular replica to main server

1. cassandra DB by location
- single point failure
- N (log service) -> dispatch by LB
- cassandra -> availibilty
- write freqency -> ? cassandra LSM(mem) -> full? -> disk ss stable

- producer -> 1 ms. write latency 10 ms
- write mem cache -> hard digest traffic
- 16 hr active
- cassandra 10 ms X witre op
-> data -> LSM(mem) -> full -> sstable , 10 ms

pararell -> memory

1
2 3
4
# 1 -> 4
# 1. shortest path for ans
# 2. sakeholder ask information

# 1. build ralstionship 0:[1], 1:[0], 2:[3] ...
# 2. shortest path -> BFS
# iter1 src 2
# it2 -> 3 -> 2 in seen
# it3 -> 4 -> dst -> [2, 3, 4]

def sol(weight, edges, src, dst):
# weight [1, 2, 3, 4, 5]
# edges: [[0, 1], [2, 3], [3, 4]]

###### tags: `mock interview` `面試`