A distributed, lock-free scheduled platform what provides various ways of scheduling, group isolation, high-performance etc.
Scheduler is a core module in FastJob that supports multiple scheduling strategies, e.g. CRON、Fixed Frequency、Fixed Delayed、API , but we can ignore API because of it will directly startup through interface that client supports, so the remaining scheduling policies are divided into regular tasks and second-level tasks based on execution frequency.
- Regular tasks : Acquire tasks that ready to executed from database,and generate execution record insert into database then push it to delayer timer.
- Second Level tasks : Server will not be scheduled and executed and will be directly lowered into the worker, so worker will schedule and execute.
How to conduct reliable scheduling?
Scheduler can't acquire task due to network latency or other unknown questions, so how do we schedule reliably? WAL(Write-AHead Logging) so server will insert a pre-execution record to database then push task to delayer timer, hence if the server cashed that not execute task, other server will process it, unless all servers crashed.
The traditional practice is to start all services and find a registry to register their information such as NACOS then client lookup service through NACOS then completely connect. Through distribute-lock achieved uniqueness of task scheduling i.e. only be scheduled by one machine at a time, so it can't do high performance, so we need to use the lock-free mind designing it and support group isolation.
How are group defined?
Aimed at providing accurate scheduling for various departments and lines of business,so FastJob has the concept of AppName. An AppName logically corresponds to a set of tasks for an application and physically corresponds to the cluster to which the applications is deployed.
How do you get all the machines in a group to connect to the same server?
If just using a sample hash, there are not going to be able to do that because if you have a network problem or if you have added server, the result of hash will go on offset, so we can borrow ideas from other components, such as the NameServer in RocketMQ. Random chooses a server when client need to connect then server will respond message contains which server the AppName should connect to,so we moved the process of finding a specific server from the client to the server. If most of the applications under the AppName successfully connect to the same server, but the server goes down,the server will move all the application under the AppName to the available server.
How to synchronize the metadata?
The previous answer addresses lock-free scheduling,but introduces new question that how to synchronize metadata between servers? There are two options below.
- Synchronize through database and check the heartbeat periodically.
- Use a consensus algorithm such as Gossip.
I choose the first option,because of this is easier and faster that the database is used to store information such as task execution records.
First, make sure you have the rust env, you can check by：
rustc --versionIf nothing is being output, don't worry, use the following command to download rust.
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | shOk，everything is ready.
1. git clone https://github.com/eliasyaoyc/fastjob
2. cargo build- lock-free scheduler.
- support map-reduce.
- container task.
- task tag.
- support log.
- support metrics, e.g. prometheus etc.
- support alarm, e.g. dingding etc.
- support docker startup.
- support admin.
- java client(internal use).
- improve that spontaneous metadata management.(current version through mysql.)
Licensed under of either of
- MIT license (LICENSE-MIT or http://opensource.org/licenses/MIT)