Redis

Question 1

Memcache

Answer 1

like Hash table

support only simple data type

not support of data persistence

not support master-slave

not support sharding

Question 2

Redis

Answer 2

support set, list etc, complex data type

support data persistence

support master-slave

support sharding

Question 3

Why Redis is so fast

Answer 3

support 100,000 QPS

1. solely build on memory, not limited by harddisk I/O. one process ,one thread
2. data structure is simple, so does the operation, like hashmap, key-value pair O(1)
3. single thread,support high concurrency, multi-core => multi instance; do not need lock, less conflicts, no complex locking
4. non-blocking IO, I/O多路复用技术（multiplexing）

Question 4

Redis’s multiplex mechanism

Answer 4

Redis support: epoll/kqueue/evport/select

automatically choose base on use scenario

choose on time complexity O(1)

select as an O(n) method keeps as a backup plan

base on react design pattern to monitor I/O event

Question 5

Redis‘s data type

Answer 5

• String (support jpg, serialized sequence)
• Hash (both are String)

Create Hashmap : hmset lilei name “lilei” age 36

Get value : hget lilei age // “36”

Set value: hset lilei name “LL”

• List (like stack, LIFO)

lpush mylist aaa

lpush mylist bbb

lpush mylist ccc

lrange mylist 0 10 //print 0_10 elements in mylist

• Other types: Geo HyperLogLog

//result 1. ‘ccc’, 2. ‘bbb’, 3. ‘aaa’

• Set

sadd myset 111

members myset

• Sorted Set

zadd myzset 3 abc

zadd myzset 1 abd

zrangebyscore myzset 0 10 // 1. abd 3. abc

Question 6

Query for keys with certain prefix in a big set of keys

Answer 6

My Answer: how big is the set?

Use KEYS pattern to query for keys with a certain pattern

“keys k1*” //start with ‘k1’

Interviewer Q: The service is online, Will use KEYS interfere with the service?

My Answer: Yes, it will use a lot of computing resources. Because KEYS will return all the matching keys. if the amount is huge, it will slow down the server.

Use SCAN instead: SCAN cursor [MATCH pattern] [COUNT count]

1. iterator based on the cursor (last time result)
2. it starts with cursor 0, util it returns another 0 (reach the end), called a full iteration
3. does not guarantee the fixed amount of keys a time, around the count parameter; Support fuzzy query

Example:

scan 0 match k1* count 10 // from cursor 0 scan for prefix k1

result:

1) ‘123’
2) 1) ‘k15435’
2) ‘k1644’
3) ‘k14667’

scan 123 match k1* count 10

……

Warning: does not guarantee non-duplication, need to write script to filter out dups.

Question 7

what are the tasks for distributed lock

Answer 7

1. mutex: blocking other
2. security: only the lock owner could release the lock
3. deadlock resolving
4. fault tolerance

Question 8

How to implement distributed lock using Redis？

Answer 8

Option 1 using SETNX and EXPIRE，not an optimal option which disobeys atomicity

SETNX locknx test, only when the key is existing, it will be created and initialized with value; else, it will decline the update

people use its feature to test whether there are other transactions using this part of data. if there is, wait for its release.

expire locknx 2 //2 seconds it will expire

Option 2 SET key value [EX seconds] [PX milliseconds] [NX|XX]

EX PX: expire time

NX: set the key when it is non-exist

XX: set the key when it exist

return value : OK or nil

example:

set locktarget 12345 ex 10 nx

Follow-up

if many key expires at the same time, it will slow down service, how to solve it?

add random number to expire time when it is created