Summary of basic concepts of moosefs

I recently learned about a Distributed file system ——MooseFS, Previous pair Distributed Little is known about , Distributed file system 、 Distributed Databases are near and far away , I think it's too complicated. It's still far away from me . With the help of all the teachers, I use 6 This machine has been put into practice moosefs,moosefs The deployment of is still very simple , And configuration NFS It's like , It's a machine with two more roles , It is with them , So that moosefs It is much better in scalability and stability than NFS, In terms of read-write performance , adopt dd A simple test conducted ,moosefs That is, the write speed is slightly better than NFS, No difference in reading . The following is about MFS Some summary of knowledge points .

MFS System consists of 4 Parts make up ：master、metalogger、chunkserver、client.

Master —— mfs Brain , Record management information , such as ： file size , Storage location , Number of copies, etc , and innodb China Space （ibdata） The information stored in is similar to , This information is recorded in metadata.mfs in , When the file is loaded into memory , The file is renamed to metadata.mfs.back, When chunkserver When there is an update on ,master New information will be written back to metadata.mfs.back in , Ensure the reliability of metadata .

Hardware recommendation ： Large memory , Because in memory you need to metadata.mfs Loading in , The size of this file depends on you chunkserver The amount of data stored on , The size of memory will become a problem later , want ECC Can be used for error checking , When the amount of data in memory reaches a certain level , If there is no fault-tolerant mechanism , It's going to be terrible ; Redundant batteries , And disk configuration RAID1/RAID5/RAID10, All to ensure high reliability .

Metalogger —— mfs Backup of , like mysql Medium m-s structure ,metalogger Will be heavy on a regular basis master Admiral's metadata、changelog、session The type of file download is synchronized to the local directory , And suffixes ”_ml” Rename it .

Hardware recommendation ： And master The machine configuration is consistent ,metalogger Itself is master A backup of , When master After downtime , You can directly metalogger Upgrade to master.

Chunkserver —— Where the data is stored , Document to chunk Size storage , Every time chunk The maximum is 64M, Less than 64M Of , The chunk The size of is the file size , exceed 64M The documents will be divided equally , Every one （chunk） The size of is less than 64M For principle ; There can be more than one document copy, That is, in addition to the original documents , The number of copies of the document stored , When goal by 1 when , It means that there is only one copy, This copy Will be randomly stored in one chunkserver On , When goal The number of is greater than 1 when , Every one copy Will be saved separately to each chunkserver On ,goal Don't exceed the size of chunkserver The number of , Otherwise, there will be more copy, There will be no chunkserver To deposit ,goal It doesn't really make sense to have more settings .Copy Number of copies , It is generally set to be greater than 1 Share , So if there is one chukserver When it's broken , At least one more copy, When this one was added again , The share that will be lost copy Make up , Always keep the original copy Count , And if the goal Set to 1copy, So when it comes to storage copy Of chunkserver Broken , And then back in ,copy The number will always be 0, It won't go back to the previous 1 individual copy.

Chunkserver The remaining storage space on is larger than 1GB（Reference Guide Have mentioned ）, New data will be allowed to write , otherwise , You'll see No space left on device A hint of , In the actual , The test found that when the disk utilization reached 95% Left and right , You can no longer write , The available space was 1.9GB.

Hardware recommendations ： Just ordinary machines , Just to save some data , As long as the disk is big enough .

Client —— The client is loaded through the kernel FUSE modular , Through again and master Communication , take chunkserver The shared partition is mounted locally , Then read and write . because FUSE Modules are additional modules , When the system restarts , You need to perform modprobe fuse, Load it into the kernel .