Don't buy your own ECS pressure test MySQL! Hard Core Dismantling Four Core Advantages of Alibaba Cloud RDS
If you are still using Docker or source code compilation to install MySQL and PostgreSQL on ECS (cloud server) in order to save a few oceans, and then keep a bunch of Crontab scripts for backup every day, then I can only wish your business never encounter large traffic.
In the era of cloud computing,
RDS(Relational Database Service, Relational Database Service)
It is definitely one of Aliyun's most profitable and hardcore trump card products. Many novices think RDS is more expensive than ECS and do not think it is cost-effective. However, as long as your business has experienced a "master/slave switch, data deletion by mistake, blackmail virus or CPU soaring to 100 in the middle of the night", you will understand:
The money that RDS costs is actually "life-saving insurance" and "24-hour full-time guard for top DBA".
Today, we will not talk about those cold official indicators, but directly use vernacular and actual combat scenarios to talk about the core functions of Aliyun RDS that can make you "sleep peacefully.
1. core function one: high availability architecture and second-level "fault self-healing"
What is the most afraid of self-built database? Afraid that the server hardware will hang up and the network will be cut off. Once the main library goes down, you have to get up in the middle of the night, manually switch the backup library, and change the connection string in the code. The whole process takes at least half an hour, and the whole company is waiting for you.
In Aliyun RDS, high availability (such as three-node enterprise edition and dual-machine high availability edition) is the standard smart chassis:
Low-level logic: RDS will deploy a set of "one master and one standby" (or even one master and two standby) database clusters for you on different data centers (available areas) in the same region or on different racks in the same data center.
Real-time synchronization: All data changes in the primary database are replicated to the secondary database in real time through strong synchronization or semi-synchronization.
Second-level switching: The top layer of RDS has a very smart "HA Daemon". It's like a sentinel, doing heartbeat checks on your main library every second. Once the server where the main library is located is found to be smoking, the sentry will immediately issue an instruction to automatically cut the traffic to the standby library within 30 seconds. The whole process database connection address (Endpoint) is completely unchanged, your back-end code does not need to change any line, the user may just feel a little stuck, the data has been safely transferred.
2. core function 2: data backup and "time machine" function (regret medicine for data deletion by mistake)
"Xiao Wang, go to clear the data of the test library ...... lie trough, what you just deleted is the production library?!"
I believe that every programmer has more or less heard or experienced the tragedy of "deleting the library and running away. A backup of a self-built database is often written.
mysqldump
The script runs once a day in the middle of the night. If the database is deleted by mistake at 4 pm, the data for the first 16 hours of today will be completely wiped out.
Alibaba Cloud RDS
Automated Backup and Log Backup (Binlog)
This is the perfect time machine ":
Snapshot backup + Binlog real-time append: RDS not only helps you do full backup every day, but also automatically uploads every step of the database write operation (Binlog) to the extremely secure OSS (object storage) in real time.
Any point in time recovery (PITR): If you delete data by mistake at 14:05:30 this afternoon. You can directly click "Clone Instance" or "Restore Data" in the Alibaba Cloud background to specify the recovery to 14:05:29. RDS will automatically pull out a new temporary database, pour the full backup into it, and then play the Binlog to the second before you delete the database like a video fast forward. Data perfect resurrection!
3. core function 3: read-write separation and exclusive architecture (antidote to resist large traffic)
When your business is bigger, the number of users increases sharply, and the CPU of the database is stuck at more than 90% every day, what should you do?
If it is a self-built database, you have to do master-slave replication yourself and write logic in the back-end code: write operations go to server a and read operations go to server B. The code structure can become extremely bloated and painful.
The solution to RDS is much more elegant and is called
Built-in read-write separation (exclusive agent)
:
Dumbstart: You only need to buy a few "read-only instances" in the background (the price is much cheaper than the main instance).
Unified entry: RDS will give you a "read-write separation address". You change the link address in the code to this.
AI intelligent offloading: Alibaba Cloud's database proxy layer automatically reviews each SQL statement for you. As soon as you see INSERT, UPDATE, DELETE, it will be automatically sent to the main library. When you see a large number of SELECT queries, they are evenly distributed to those read-only instances according to their weights.
Exclusive resources: By the way, you must select the exclusive type (Dedicated) when purchasing RDS. It means that the CPU core, memory and I/O bandwidth on this physical machine are completely exclusive to you, and you will never squeeze your database and be full of stability just because your next-door neighbor makes a big push.
4. core function four: DAS (database autonomous service) and slow SQL diagnosis
Most of the time, the database is stuck not because the hardware is not strong enough, but because an intern wrote an extremely rubbish SQL statement, such as not building an index, resulting in a "full table scan" of tens of millions of data ".
If you want to check this kind of problem in your self-built database, you have to go through hundreds of megabytes of slow query logs (Slow Log) and reuse them.
Explain
Analyzing one by one, my hair is bald.
Alibaba Cloud RDS comes
DAS(Database Autonomy Service)
It is simply the embodiment of a senior DBA in a large factory:
Slow SQL automatic crawling: it will slow down the database.
The "culprit" is listed directly in your dashboard, and even ranks you according to the proportion of CPU consumed.
One-click optimization suggestion: DAS not only tells you that this SQL is slow, but also clearly points out: "Buddy, if you build an index on user_id field of this table, the performance can be improved by 98%."
Automatic throttling: If you encounter a sudden malicious library brush or a dead loop code, DAS can also trigger automatic throttling to block this toxic SQL and keep the entire database from crashing.
Summary: Calculate an economic account.
Finally, let's settle the score. Buying a basic RDS dual-machine high-availability version may cost 100 or 200 yuan a month more than ECS of the same specification.
But this 100 or 200 dollars, you bought:
A standby server for remote disaster recovery.
An automated backup system that doesn't require you to write scripts and has almost unlimited space.
A failover middleware that automatically wipes your ass in 30 seconds.
A top DBA who doesn't need to be paid and keeps an eye on slow SQL 24 hours a day.
If it is a large and medium-sized enterprise or core business, close your eyes and choose
RDS (especially Enterprise Edition/Three-node Walnut Edition)
; If it's a personal toy project, it doesn't matter if you build it yourself. It is the smartest way for modern developers to leave professional things to professional cloud products and leave their energy to make money and write business logic.
