Versions Compared

Old Version 27

changes.mady.by.user Halil Cakal

Saved on

compared with

New Version Current

changes.mady.by.user Halil Cakal

Saved on

Key

  • This line was added.
  • This line was removed.
  • Formatting was changed.

...

Issue

Notes 

Decision

1

Agree on endurance performance KPI

The aim of the study is to early detection of bugs similar to

Jira Legacy
serverSystem Jira
serverId4733707d-2057-3a0f-ae5e-4fd8aff50176
keyCPS-2430
or any memory leakage in the future.

Thus, we should spot the specific function/s currently used in k6 to be improved. These function/s will be used in the Endurance test suit.

Actions:

  1. Daniel Hanrahan and Lee Anjella Macabuhay can shed light on this item since they were resolving the bug.

  2. After identifying the test function/s, what can we functionally do to improve them?
    The reproduction of the bug includes two more data operations that are not available in the current k6 suit:

    1. Passthrough-write (PATCH)

    2. Passthrough (DELETE) / Delete Resource Data

  3. Regarding parallelism, how many VUs should be assigned to these tests?
    The suggested number of VUs with 'constant-vus' executor:

    1. Passthrough-read (READ): 4

    2. Passthrough-write (POST): 4

    3. Passthrough-write (PATCH) : 4

    4. Passthrough (DELETE) / Delete Resource Data: 4

CPS need to run all the test cases according to FS not according to what was tested in cps-2430 (Seeing they ran test that was outside of FS). Less focus on the KPI for this endurance run. Kolawole Adebisi-Adeolokun Halil Cakal

2

Agree on the test environment

Blocked by

Jira Legacy
serverSystem Jira
serverId4733707d-2057-3a0f-ae5e-4fd8aff50176
keyCPS-2463

  1. if a new test server can be provided, then no need to have a secondary docker-compose deployment descriptor (port and container name changes)

  2. if no new server can be provided, then a secondary docker-compose deployment descriptor will be necessary

    1. There will be 3 different pipelines will be running on the physical test server:

      1. CPS performance tests (groovy-based)

      2. NCMP performance tests (k6-based)

      3. NCMP Endurance tests (k6-based, long-running)

  3. Team Kraken could set up test envi for us

  1. No parallelism. A conversation with Kraken to discuss option #3 Halil Cakal

  2. The discussion was made with Team Kraken and decided to have two docker-compose deployments on the test server. Daniel Hanrahan Halil Cakal Toine Siebelink

    1. The PoC has been done on Nordix.
      PoC using two docker-compose using an environment file

  3. The number of available executors should be sufficient to run a possible three jobs on the physical performance test server. Daniel Hanrahan Toine Siebelink Kolawole Adebisi-Adeolokun Halil Cakal

3

Visualization of memory trend

The most convenient way to represent the memory usage trend in a GUI is Grafana.

  1. Grafana

  2. GNUPlot (3pp library current in the use of k6 and Groovy test Visualization)

    1. If Gnuplot is to visualize memory usage, which parameters, logs, or output should be input?

    2. Issue #4 (external access) will no longer be valid. Since Jenkin’s HTML publisher can do it.

Agreed to use Grafana majorly because it allows storage for longer periods sticking with our A/C Kolawole Adebisi-Adeolokun Halil Cakal

The same decision was made by Daniel Hanrahan and Halil Cakal during the daily stand-up.

Note: A PoC for the GNUPlot option has been made, please see the relevant Jira ticket details. Daniel Hanrahan Halil Cakal Toine Siebelink

Jira Legacy
serverSystem Jira
serverId4733707d-2057-3a0f-ae5e-4fd8aff50176
keyCPS-2466

4

Grafana access externally (If Grafana is selected at #3 then this issue should be discussed, otherwise ignore)

  1. Make local GUI link accessible externally

  2. Give server access to the CPS team

Team Kraken support is needed either way.

Since Grafana is the preferred option, access issues shall be discussed with team Kraken Halil Cakal (consider Jenkins plug-in option as well)7

5

Permanent storage (DB) for Grafana (If Grafana is selected at #3 then this issue should be discussed, otherwise, ignore)

Investigate the solution strategy for the storage e.g. a permanent service 7/24 running …

Halil Cakal to investigate how to tune the volume

...

Suggested Tasks

k6 performance suitAdd new k6 test suit

Description

Jira

Add new

test

profile ‘Endurance’

Jira Legacy
serverSystem Jira
serverId4733707d-2057-3a0f-ae5e-4fd8aff50176
keyCPS-2464

Add new Jenkins job to run endurance test

Jira Legacy
serverSystem Jira
serverId4733707d-2057-3a0f-ae5e-4fd8aff50176
keyCPS-2465

Add Grafana support to visualize memory usage pattern

Jira Legacy
serverSystem Jira
serverId4733707d-2057-3a0f-ae5e-4fd8aff50176
keyCPS-2466

Two docker-compose deployments simultaneously

Jira Legacy
serverSystem Jira
serverId4733707d-2057-3a0f-ae5e-4fd8aff50176
keyCPS-2463

Agree and Define new ‘Suite’ (js)

Jira Legacy
serverSystem Jira
serverId4733707d-2057-3a0f-ae5e-4fd8aff50176
keyCPS-2493

Solution Proposal

...

For the sake of simplicity, the solution would be straightforward if a new test server is provided.

The functional/test-suit improvements in the current k6 performance tests if required

...

Agree and Define new ‘Suite’ (js)

  1. The existing K6 performance tests are enough to detect any memory leakage so that no new K6 test will be added. Daniel Hanrahan Halil Cakal
    There should be a new test suit Toine Siebelink Kolawole Adebisi-Adeolokun Halil Cakal

    Jira Legacy
    serverSystem Jira
    serverId4733707d-2057-3a0f-ae5e-4fd8aff50176
    keyCPS-2493

    All tests should be run in parallel in the Endurance test


    HOW: Daniel and I decided to define a new test suite (endurance). Daniel Hanrahan Halil Cakal
    Have one ncmp-test-runner.js with different configs: kpi.json and endurance.json, moving the scenarios and thresholds settings into the json configs
    The endurance test should run all tests in parallel. Also, the same scenarios, executor types, and VUs should be the same in the Endurance suit except for the legacy_batch_consume_scenario. The executor type for legacy_batch_consume_this scenario should be changed to constant-arrival-rate with 1 req/second. Toine Siebelink Daniel Hanrahan Kolawole Adebisi-Adeolokun Halil Cakal
    As an example:

    Code Block
    languagejson
    {
  "hosts": {
    "ncmpBaseUrl": "http://localhost:8883",
    "dmiStubUrl": "http://ncmp-dmi-plugin-demo-and-csit-stub:8092",
    "kafkaBootstrapServer": "localhost:9092"
  },
  "kafka": {
    "legacyBatchTopic": "legacy_batch_topic"
  },
  "scenarios": {
    "passthrough_read_scenario": {
      "executor": "constant-vus",
      "exec": "passthroughReadScenario",
      "vus": 2,
      "duration": "15m"
    }
  },
  "thresholds": {
    "http_req_failed": ["rate == 0"],
    "cmhandles_created_per_second": ["avg >= 22"],
    "cmhandles_deleted_per_second": ["avg >= 22"],
    "ncmp_overhead_passthrough_read": ["avg <= 40"]
  }
}

The current KPI and ENDURANCE (proposed) test parameters (Last updated )

1

K P I

E N D U R A N C E

2

Test Stages

Scenario Name

Unit

Executor Type

VUs

Duration

Scenario Name

VUS

Executor Type

Duration

3

setup

create_cm_handles

CM-handles/second

N/A

N/A

20m

create_cm_handles

N/A

N/A

20m

4

scenario

passthrough_read_scenario

overhead

constant-vus

2

15m

passthrough_read_scenario

2

constant-vus

1h

5

passthrough_read_alt_id_scenario

overhead

constant-vus

2

passthrough_read_alt_id_scenario

2

constant-vus

6

passthrough_write_scenario

overhead

constant-vus

2

passthrough_write_scenario

2

constant-vus

7

passthrough_write_alt_id_scenario

overhead

constant-vus

2

passthrough_write_alt_id_scenario

2

constant-vus

8

cm_handle_id_search_nofilter_scenario

milliseconds

constant-vus

1

cm_handle_id_search_nofilter_scenario

1

constant-vus

9

cm_handle_id_search_module_scenario

milliseconds

constant-vus

1

cm_handle_id_search_module_scenario

1

constant-vus

10

cm_handle_id_search_property_scenario

milliseconds

constant-vus

1

cm_handle_id_search_property_scenario

1

constant-vus

11

cm_handle_id_search_cpspath_scenario

milliseconds

constant-vus

1

cm_handle_id_search_cpspath_scenario

1

constant-vus

12

cm_handle_id_search_trustlevel_scenario

milliseconds

constant-vus

1

cm_handle_id_search_trustlevel_scenario

1

constant-vus

13

cm_handle_search_nofilter_scenario

milliseconds

constant-vus

1

cm_handle_search_nofilter_scenario

1

constant-vus

14

cm_handle_search_module_scenario

milliseconds

constant-vus

1

cm_handle_search_module_scenario

1

constant-vus

15

cm_handle_search_property_scenario

milliseconds

constant-vus

1

cm_handle_search_property_scenario

1

constant-vus

16

cm_handle_search_cpspath_scenario

milliseconds

constant-vus

1

cm_handle_search_cpspath_scenario

1

constant-vus

17

cm_handle_search_trustlevel_scenario

milliseconds

constant-vus

1

cm_handle_search_trustlevel_scenario

1

constant-vus

18

legacy_batch_produce_scenario

milliseconds

shared-iterations

2

N/A

legacy_batch_produce_scenario

1 (1 req. sec)

constant-arrival-rate

19

legacy_batch_consume_scenario

events/second

per-vu-iterations

1

20

teardown

delete_cm_handles

CM-handles/second

N/A

N/A

20m

delete_cm_handles

N/A

10m

...

Visualizing the ENDURANCE test results

As mentioned in issues/decisions, there are two alternative ways of representing memory trends: Grafana and

...

GnuPlot.

Grafana


Nordix has its own Prometheus and Grafana

...

(externally accessible, no need to install Globalprotect) and it can be configured to show cps-and-ncmp memory trends.
Link to Nordix Grafana: https://monitoring.nordix.org/login

Through prometheus.yml (of Nordix), a new scrape_configs for the cps-and-ncmp microservice can be added.

Code Block
languageyaml
scrape_configs:
- job_name: 'cps-and-ncmp'
  metrics_path: '/actuator/prometheus'
  scrape_interval: 5s
  static_configs:
    - targets:
      - 'cps-and-ncmp:8080' // replace by <physical-server-ip:port>

Also, the dashboard provider and dashboard config (jvm-micrometer-dashboard.json) can be added.

Code Block
languageyaml
providers:
  - name: default
    orgId: 1
    type: file
    options:
      path: /var/lib/grafana/dashboards
      foldersFromFilesStructure: true

Then, the trend of G1 Old Gen space can be observed as seen below:

...

Permanent Storage Alternatives
Prometheus

Configuring the Prometheus with a persistent volume to retain data is possible.

This is an example service config for Prometheus:

Code Block
languagejson
  prometheus:
    container_name: ${PROMETHEUS_CONTAINER_NAME:-prometheus}
    image: prom/prometheus:latest
    ports:
      - ${PROMETHEUS_PORT:-9090}:9090
    restart: always
    volumes:
      - ./config/prometheus.yml:/etc/prometheus/prometheus.yml
      - prometheus_data:/prometheus
    profiles:
      - monitoring
      
volumes:
  prometheus_data:
    driver: local

GNUPlot

GnuPlot can also draw a plot for only G1 Old Gen space

...

Getting support from Team Kraken for external link/server access, if we decide to visualize memory trends by Grafana.

...

.