关于Flink Slots Metrics（可用/运行中/已注册）的差异咨询

Hey folks, let's get straight to the differences between these three critical slot metrics in Flink—they're key to understanding your cluster's resource health and utilization.

Registered Slots #

Think of this as the total capacity of your cluster's slot pool. Every slot that's been set up on a TaskManager and successfully registered with the JobManager counts here. It doesn't matter if the slot is empty or busy; as long as the TaskManager is online and connected to the JobManager, all its configured slots are included in this number.

For example: If you spin up 3 TaskManagers each with 4 slots, your registered metric will be 12, even if none of those slots are running any tasks yet. This number only changes when you add or remove TaskManagers, or adjust the slot count per TaskManager and restart them.

Running Slots #

This metric tracks how many slots are actively executing tasks. When the JobManager assigns a task (like a map, filter, or sink operator) to a slot, and that task is up and running, the slot gets counted here. It's a direct measure of your cluster's current workload—higher numbers mean more of your capacity is being used for actual work.

If that same 12-slot cluster is running a job that uses 7 slots, your running metric will be 7. This number fluctuates as jobs start, finish, scale up, or fail (and tasks restart).

Available Slots #

This is the number of idle slots ready to take on new tasks. It's essentially the remaining capacity in your cluster, calculated as registered slots - running slots (minus any slots temporarily unavailable due to maintenance or TaskManager issues, though those are rare in healthy clusters).

Using the previous example: 12 registered slots minus 7 running slots gives you 5 available slots. This metric tells you if you have room to submit new jobs or scale existing ones without needing to add more TaskManagers.

Quick Recap of Their Relationship #

• registered = total cluster slot capacity (fixed until you adjust your TaskManager setup)
• running = slots currently doing work (varies with job activity)
• available = slots waiting to be assigned (varies inversely with running slots, assuming registered stays constant)

内容的提问来源于stack exchange，提问作者Slava Shpitalny