Redis AOF log

Why is the log written after

Because there is no syntax check in the log , Write the log first , Then the wrong command may be written . It's not good for redis Use log recovery
Post writing depends on redis perform , To ensure the correctness of the order

AOF Three kinds of drop mechanism

Drop disk after writing

redis After executing a command , Write a log to disk
Write operations block threads , Influence the follow-up redis Carry out orders
efficiency ： low
Downtime It is possible to lose a record

Write every second

redis After executing a command , Write command to AOF Memory buffer
Write the contents of the buffer to the disk every second （ asynchronous ）
efficiency ： in
Downtime It is possible to lose the log for one second

Operating system controls writeback

redis After executing a command , Write command to AOF Memory buffer
The operating system controls when to write to disk （ asynchronous ）
efficiency ： high
Downtime It is possible to lose a large number of logs

AOF rewrite

AOF The function of rewriting is to ignore the process , Pay attention to the results （ Anyway, I have nothing to do ）
Suppose I have a record A, By 100 The command was modified a hundred times , The end result is 100
here AOF There are records 100 strip , But after rewriting, there is only one
namely A = 100
AOF Rewrite to reduce the log size
Large files are not allowed by the file operating system
The log of large files will affect the writing speed
Logging of large files will affect redis Speed of recovery

AOF Rewrite mechanism

The main thread fork Out bgrewriteaof Subprocesses
bgrewriteaof Middle copy redis Memory data
bgrewriteaof Threads in will reids Memory data + AOF Rewrite buffer data Make up a new AOF To replace the old AOF
stay AOF In the process of rewriting , old AOF Still writing normally , To deal with the downtime problem when the rewrite is not completed

AOF Rewriting will affect the main thread

1. The main thread will be blocked when the child process copies data
2. When the main thread operates on existing data , It will also block
The parent-child process copies data using write time replication
Only the necessary data structures are copied in the subprocess
The child process and the parent process point to the same memory space
When the parent process writes , Will allocate a new memory address , Copy the data in