IOPS have been modelled as a resource since Nomad 0.1 but has never actually been detected and there is no plan in the short term to add detection. This is because IOPS is a bit simplistic of a unit to define the performance requirements from the underlying storage system. In its current state it adds unnecessary confusion and can be removed without impacting any users. This PR leaves IOPS defined at the jobspec parsing level and in the api/ resources since these are the two public uses of the field. These should be considered deprecated and only exist to allow users to stop using them during the Nomad 0.9.x release. In the future, there should be no expectation that the field will exist.
3.1 KiB
layout | page_title | sidebar_current | description |
---|---|---|---|
guides | Resource Utilization - Operating a Job | guides-operating-a-job-resource-utilization | Nomad supports reporting detailed job statistics and resource utilization metrics for most task drivers. This section describes the ways to inspect a job's resource consumption and utilization. |
Resource Utilization
Understanding the resource utilization of an application is important, and Nomad
supports reporting detailed statistics in many of its drivers. The main
interface for seeing resource utilization is the alloc status
command with the
-stats
flag.
This section will utilize the job named "docs" from the previous sections, but these operations and command largely apply to all jobs in Nomad.
As a reminder, here is the output of the run command from the previous example:
$ nomad job run docs.nomad
==> Monitoring evaluation "42d788a3"
Evaluation triggered by job "docs"
Allocation "04d9627d" created: node "a1f934c9", group "example"
Allocation "e7b8d4f5" created: node "012ea79b", group "example"
Allocation "5cbf23a1" modified: node "1e1aa1e0", group "example"
Evaluation status changed: "pending" -> "complete"
==> Evaluation "42d788a3" finished with status "complete"
To see the detailed usage statistics, we can issue the command:
$ nomad alloc status -stats 04d9627d
ID = 04d9627d
Eval ID = 42d788a3
Name = docs.example[2]
Node ID = a1f934c9
Job ID = docs
Client Status = running
Task "server" is "running"
Task Resources
CPU Memory Disk Addresses
75/100 MHz 784 KiB/10 MiB 300 MiB http: 10.1.1.196:5678
Memory Stats
Cache Max Usage RSS Swap
56 KiB 1.3 MiB 784 KiB 0 B
CPU Stats
Percent Throttled Periods Throttled Time
0.00% 0 0
Recent Events:
Time Type Description
<timestamp> Started Task started by client
<timestamp> Received Task received by client
Here we can see that we are near the limit of our configured CPU but we have plenty of memory headroom. We can use this information to alter our job's resources to better reflect is actually needs:
resource {
cpu = 200
memory = 10
}
Adjusting resources is very important for a variety of reasons:
- Ensuring your application does not get OOM killed if it hits its memory limit.
- Ensuring the application performs well by ensuring it has some CPU allowance.
- Optimizing cluster density by reserving what you need and not over-allocating.
While single point in time resource usage measurements are useful, it is often more useful to graph resource usage over time to better understand and estimate resource usage. Nomad supports outputting resource data to statsite and statsd and is the recommended way of monitoring resources. For more information about outputting telemetry see the Telemetry Guide.
For more advanced use cases, the resource usage data is also accessible via the client's HTTP API. See the documentation of the Client's allocation HTTP API.