redis-0. Introduction to redis and NiO principle (random talk)

reids

redis Introduce

common sense

First of all, common sense , The data in the computer is stored in disk , There are actually two dimensional indicators
1. Addressing ：ms It's on the millisecond level
2. bandwidth ：G/M How many? G How many? M
Memory
1. Addressing ：ns Nanosecond level
2. bandwidth ： It's big
second > millisecond > Microsecond > nanosecond . Disk is slower than memory 10W times , On addressing

I/O buffer： The question of cost
The disk has tracks , A sector , One sector 512Byte Bring a larger index of costs
4K Operating system default , No matter how much you read , It's all the least 4k From disk .

So the file becomes larger and the speed becomes slower

Relational database

And the database appears data page Size is 4k, It just matches the size of the hardware disk . It can be reduced by a certain amount io The operation of , Avoid waste , Because whatever it is 4k. But the database is still slow , So there are indexes in the database , As shown above , The index corresponds to each of these page, Faster than retrieval .B+T What is stored inside is an interval , such as 1,7,10, If you inquire 6 Will fall into the first data page Indexes , Find the corresponding in the index data page data

Create tables for relational databases ： Must give schema type ： Byte width memory ： Prefer row level storage

Interview questions database ： The watch is big , Performance degradation ？ If the table has an index, add, delete, or change Slow queries （ Maintain data index , So it slows down ）
How about the speed ？1,1 One or a few queries are still very fast 2, When the concurrency is large, the speed will be affected by the bandwidth of the hard disk

Non relational database

In fact, there is also an in memory database SAP–HANA . But this kind of family charges , There is a knowledge that ： The volume of data on disk and memory is different .
This kind of situation occurs when people do not want to pay and do not want their own services to be slow redis Non relational row database of cache class , Based on two infrastructures
Basic knowledge of computer ：2 Infrastructure 1, The hardware of von Neumann system 2, Ethernet ,tcp/ip Network of （ The subtext is instability ）
If the above two infrastructures are completed , There will be no more non relational types

https://db-engines.com/en/ This website also sees the ranking of the database

You can also see the current PostgreSQL The hot

redis

redis Chinese official website ：http://redis.cn/

Redis It's open source （BSD The license ） Of , Data structure storage system in memory , It can be used as a database 、 Caching and message middleware . It supports multiple types of data structures , Such as character string （strings）, hash （hashes）, list （lists）, aggregate （sets）, Ordered set （sorted sets） And scope query , bitmaps, hyperloglogs and Geographical space （geospatial） Index radius query . Redis Built in Copy （replication）,LUA Script （Lua scripting）, LRU Driving events （LRU eviction）, Business （transactions） And different levels of Disk persistence （persistence）, And pass Redis sentry （Sentinel） And automatic Partition （Cluster） Provide high availability （high availability）.

Actually redis Type is not very important , It is important to redis Of server There are methods for each type in index()lpop

redis It's a single process , Single thread , Single instance , How requests become fast when concurrent , The operating system has a kernel （kernel） The concept of , The user enters the kernel first , Kernel in epoll ( Multiplexing ) to redis

The client to Linux It's in milliseconds ,Linux Memory interaction is nanoscale

io Technology is evolving

BIO

In the early days, the kernel only had read fd fd( representative Linux File descriptor ), This is the time socket It's blocked , When the thread is processing , The remaining threads can only wait

NIO

yum install man man-pages Install it. man User manual

#  see read  Method 
main 2 read 

linux Everything inside is a document

# see redis
ps -ef|grep redis
#  Enter... Under the change process fd file 
cd /proc/20094/fd

man 2 socket see socket Operation manual , We can see socket It can be non blocking , Before BIO yes block Of

Synchronous nonblocking

NIO Multiplexing

above nio Because the user process needs to poll constantly kernel operation , At this time, the user space can do nothing , The kernel is required to operate the upgrade

 #  see select operation 
 man 2 select 

kernel select The main thing is to monitor more fds File descriptor

In the past, I wrote an endless loop , Now it's time to stop writing dead cycles , If there is 1000, Just put this 1000 Pass it on to you , If there is any change, it will be returned to user space , Then go on read operation , Reduce user status , Kernel state switching

epoll false AIO

window Realized AIO operation

 main 2 mmap

This is the time mmap There is ,mmap It's a way to map files in memory , Mapping a file or other object to the address space of a process , Realize the one-to-one mapping between the file disk address and the virtual address space of a process .

man epoll

see epoll
Actually epoll It doesn't use mmap It's just using Red black tree and linked list Easy to read and find quickly

kafka Of io operation ,nginx gossip