Files
kubernetes/test/integration/scheduler_perf
Jan Chaloupka 7b5534021c Collect some of scheduling metrics and scheduling throughput
In addition to getting overall performance measurements from golang benchmark,
collect metrics that provides information about insides of the scheduler itself.
This is a first step towards improving what we collect about the scheduler.

Metrics in question:
- scheduler_scheduling_algorithm_predicate_evaluation_seconds
- scheduler_scheduling_algorithm_priority_evaluation_seconds
- scheduler_binding_duration_seconds
- scheduler_e2e_scheduling_duration_seconds

Scheduling throughput is computed on the fly inside perfScheduling.
2020-02-13 13:32:09 +01:00
..
2019-12-19 09:08:11 +08:00
2017-07-12 17:34:55 -07:00
2017-08-22 16:40:16 +08:00
2020-02-08 12:30:21 -05:00

Scheduler Performance Test

Motivation

We already have a performance testing system -- Kubemark. However, Kubemark requires setting up and bootstrapping a whole cluster, which takes a lot of time.

We want to have a standard way to reproduce scheduling latency metrics result and benchmark scheduler as simple and fast as possible. We have the following goals:

  • Save time on testing
    • The test and benchmark can be run in a single box. We only set up components necessary to scheduling without booting up a cluster.
  • Profiling runtime metrics to find out bottleneck
    • Write scheduler integration test but focus on performance measurement. Take advantage of go profiling tools and collect fine-grained metrics, like cpu-profiling, memory-profiling and block-profiling.
  • Reproduce test result easily
    • We want to have a known place to do the performance related test for scheduler. Developers should just run one script to collect all the information they need.

Currently the test suite has the following:

  • density test (by adding a new Go test)
    • schedule 30k pods on 1000 (fake) nodes and 3k pods on 100 (fake) nodes
    • print out scheduling rate every second
    • let you learn the rate changes vs number of scheduled pods
  • benchmark
    • make use of go test -bench and report nanosecond/op.
    • schedule b.N pods when the cluster has N nodes and P scheduled pods. Since it takes relatively long time to finish one round, b.N is small: 10 - 100.

How To Run

# In Kubernetes root path
make generated_files

cd test/integration/scheduler_perf
./test-performance.sh

Analytics

Analytics