Nvidia runtime in a gitlab-ci service - gitlab-ci

I have a service which loads docker daemon docker and I want to utilize nvidia gpus inside this service, I have created an image which has a dockerd as entrypoint and installed nvidia cuda driver but when I am trying to deploy my application for test purposes I am facing errors related with nvidia driver inside the service’s image
docker: Error response from daemon: OCI runtime create failed: container_linux.go:346: starting container process caused “process_linux.go:449: container init caused “process_linux.go:432: running prestart hook 1 caused \“error running hook: exit status 1, stdout: , stderr: nvidia-container-cli: initialization error: driver error: failed to process request\n\”””: unknown.
#image: chatzich/dinvidia
stages:
- test
- package
before_script:
- echo "Before script section"
after_script:
- echo "After script section"
test:
image: chatzich/dinvidia
services:
- name: chatzich/dinvidia
alias: dinvidia
stage: build
variables:
DOCKER_HOST: tcp://dinvidia:2375
DOCKER_DRIVER: overlay2
DOCKER_TLS_CERTDIR: ""
tags:
- gpu
script:
# Build the production image
- echo "Build stage script"
- ldconfig -p | grep nvidia
- docker --version
- docker run --rm --runtime=nvidia nvidia/cuda ldconfig -p | grep nvidia
stage: test
package:
stage: package
script:
# Build the production image
- echo "Package stage script"

Related

Docker Tag Error 25 on gitlab-ci.yml trying to start GitLab Pipeline

I'm going through the "Scalable FastAPI Application on AWS" course. My gitlab-ci.yml file is below.
stages:
- docker
variables:
DOCKER_DRIVER: overlay2
DOCKER_TLS_CERTDIR: "/certs"
cache:
key: ${CI_JOB_NAME}
paths:
- ${CI_PROJECT_DIR}/services/talk_booking/.venv/
build-python-ci-image:
image: docker:19.03.0
services:
- docker:19.03.0-dind
stage: docker
before_script:
- cd ci_cd/python/
script:
- docker login -u "$CI_REGISTRY_USER" -p "$CI_REGISTRY_PASSWORD" $CI_REGISTRY
- docker build -t registry.gitlab.com/chris_/talk-booking:cicd-python3.9-slim .
- docker push registry.gitlab.com/chris_/talk-booking:cicd-python3.9-slim
My Pipeline fails with this error:
See https://docs.docker.com/engine/reference/commandline/login/#credentials-store
Login Succeeded
$ docker build -t registry.gitlab.com/chris_/talk-booking:cicd-python3.9-slim .
invalid argument "registry.gitlab.com/chris_/talk-booking:cicd-python3.9-slim" for "-t, --tag" flag: invalid reference format
See 'docker build --help'.
Cleaning up project directory and file based variables
ERROR: Job failed: exit code 125
It may or may not be relevant but the Container Registry for the GitLab project says there's a Docker connection error.
Thanks
I created a new GitLab account with a new username and things are working now. The underscore does appear to have been the issue.

Gitlab-runner has 'access denied' when pushing built image

I have a problem with pushing a built image with gitlab-runner to a gitlab repository.
My gitlab-ci.yml:
image: docker:latest
services:
- docker:dind
stages:
- build
- release
variables:
TEST_IMAGE: registry.gitlab.com/myhost/haproxy:$CI_COMMIT_REF_NAME
RELEASE_IMAGE: registry.gitlab.com/myhost/haproxy:latest
before_script:
- docker login -u gitlab-ci-token -p $CI_JOB_TOKEN gitlab.com
build:
stage: build
script:
- docker build --pull -t $TEST_IMAGE .
- docker push $TEST_IMAGE
release:
stage: release
script:
- docker pull $TEST_IMAGE
- docker tag $TEST_IMAGE $RELEASE_IMAGE
- docker push $RELEASE_IMAGE
only:
- master
The docker login works - I got "Login success" - but when it comes to the push operation I get:
$ docker push $TEST_IMAGE
The push refers to repository [registry.gitlab.com/myhost/haproxy]
d77ab2f42dd4: Preparing
c70258f465dd: Preparing
96b45c1aa07c: Preparing
28587e66f3e8: Preparing
21b59fc0e3a3: Preparing
9c46f426bcb7: Preparing
9c46f426bcb7: Waiting
denied: access forbidden
ERROR: Job failed: exit code 1
The runner is on my own server, and I'm pushing to gitlab.com
I have also checked on my local machine, executing in terminal commands like in the script - login, build and push - and everything works, but if I run locally with the runner, register it and get the job, I also get an access forbidden error.
So I think the problem is in runner, bo what.
I compared the behaviour on a few versions of gitlab-runner from 10.6 to newest 11.0
Any ideas?
So the problem was wrong registry address - it should be registry.gitlab.com
It misled me that it shows "Loged in" in terminal even without "registry" prefix, so the best solution is using build in variables during login in gitlab-ci.yml:
docker login -u gitlab-ci-token -p $CI_JOB_TOKEN $CI_REGISTRY

Selenium (Docker, python, behave) takes hours to complete in CI

Tests are written with python behave and executed with Docker. Locally on Mac/Windows/Linux all 110 test steps complete under 4 minutes using the same sites whereas on CI agent (AWS) it takes 120-160 mins. Video showed that browser is spinning for about 4 minutes between steps. Is there a way to debug selenium to find out what resource is loading slow?
Selenium DEBUG logging is not helpful. After submit it took 4 minutes for a POST request to complete:
04-Apr-2018 20:13:01 DEBUG:selenium.webdriver.remote.remote_connection:POST http://127.0.0.1:4444/wd/hub/session/a6975e2107e574693fb48f21420c1850/element {"using": "id", "value": "submit", "sessionId": "a6975e2107e574693fb48f21420c1850"}
04-Apr-2018 20:13:01 DEBUG:selenium.webdriver.remote.remote_connection:Finished Request
04-Apr-2018 20:13:01 DEBUG:selenium.webdriver.remote.remote_connection:POST http://127.0.0.1:4444/wd/hub/session/a6975e2107e574693fb48f21420c1850/element/0.7198908368495902-4/click {"id": "0.7198908368495902-4", "sessionId": "a6975e2107e574693fb48f21420c1850"}
04-Apr-2018 20:17:00 DEBUG:selenium.webdriver.remote.remote_connection:Finished Request
docker-compose.yml:
version: '3.4'
services:
test:
build:
context: .
image: qa-automation
network_mode: host
environment:
LOCAL: "false"
BROWSER: "${TEST_BROWSER:-chrome}"
SERVER: "${TEST_URL:-https://example.com}"
selenium:
image: elgalu/selenium:3.11.0-p5
shm_size: 2g
ports:
- 4444:24444
- 6000:25900
environment:
SCREEN_WIDTH: 1920
SCREEN_HEIGHT: 1080
TZ: "US/Central"
VIDEO_FILE_NAME: "${TEST_BROWSER:-chrome}"
volumes:
- ./target/videos:/videos
Dockerfile:
FROM python:3.6.4-alpine3.7
# Install utilities
RUN apk --update --no-cache add bash curl git && rm -rf /var/cache/apk/*
ENV PYTHONUNBUFFERED 1
WORKDIR /app
COPY requirements.txt .
RUN pip install -r requirements.txt
# see .dockerignore
COPY . .
ENTRYPOINT ["bin/docker-entrypoint.sh"]
CMD ["behave"]
requirements.txt:
appdirs==1.4.3
behave==1.2.6
packaging==16.8
pyparsing==2.2.0
requests==2.13.0
selenium==3.11.0
six==1.11.0
test.sh:
#!/usr/bin/env bash
docker-compose up -d selenium
docker-compose exec selenium wait_all_done 30s
docker-compose exec selenium start-video
# run tests
docker-compose build test
docker-compose run test bash -c "behave"
ret_code=$?
echo "===== Auto Tests COMPLETED ====="
# clean up containers
docker-compose exec selenium stop-video
docker-compose exec selenium stop
docker-compose down
exit $ret_code

Using redis with Gitlab CI

I am currently using serverless framework and setting up gitlab ci using shared runner.
Following is my gitlab-ci.yml:
image: node:latest
services:
- redis
cache:
paths:
- node_modules/
- java/
stages:
- build
- test
- review
- staging
- production
build:
stage: build
script:
- npm install
artifacts:
paths:
- node_modules/
install:java:
stage: build
script:
- apt-get update
- apt-get install -y default-jre default-jdk openjdk-7-jre openjdk-7-jdk
- apt-get update
- sls dynamodb install
artifacts:
paths:
- java/
connect:
image: redis
script:
- redis-cli -h redis PING
unit test:
stage: test
script:
- sls dynamodb start
- babel-node ./aws/createDB.js
- npm run unit
dependencies:
- build
- install:java
unit test job requires redis and is not able to connect. Following error gets thrown, when unit test job starts:
Error while creating redis client: Error: Redis connection to
127.0.0.1:6379 failed - connect ECONNREFUSED 127.0.0.1:6379
Can someone point out what's wrong with current config file, thanks!
The host address of the redis service is redis not 127.0.0.1 or localhost.
So make sure you set the host for the redis service to redis in all of your scripts and configuration files.
Just to make people's life easier, I list an example .gitlab-ci.yml to configure Redis in Gitlab CI.
services:
- redis:latest
stages:
- test
test:
script:
- echo "hello world!"
stage: test
variables:
REDIS_PORT: 6379
REDIS_HOST: redis
REDIS_URL: redis://redis:6379

Use GitLab CI to run tests locally?

If a GitLab project is configured on GitLab CI, is there a way to run the build locally?
I don't want to turn my laptop into a build "runner", I just want to take advantage of Docker and .gitlab-ci.yml to run tests locally (i.e. it's all pre-configured). Another advantage of that is that I'm sure that I'm using the same environment locally and on CI.
Here is an example of how to run Travis builds locally using Docker, I'm looking for something similar with GitLab.
Since a few months ago this is possible using gitlab-runner:
gitlab-runner exec docker my-job-name
Note that you need both docker and gitlab-runner installed on your computer to get this working.
You also need the image key defined in your .gitlab-ci.yml file. Otherwise won't work.
Here's the line I currently use for testing locally using gitlab-runner:
gitlab-runner exec docker test --docker-volumes "/home/elboletaire/.ssh/id_rsa:/root/.ssh/id_rsa:ro"
Note: You can avoid adding a --docker-volumes with your key setting it by default in /etc/gitlab-runner/config.toml. See the official documentation for more details. Also, use gitlab-runner exec docker --help to see all docker-based runner options (like variables, volumes, networks, etc.).
Due to the confusion in the comments, I paste here the gitlab-runner --help result, so you can see that gitlab-runner can make builds locally:
gitlab-runner --help
NAME:
gitlab-runner - a GitLab Runner
USAGE:
gitlab-runner [global options] command [command options] [arguments...]
VERSION:
1.1.0~beta.135.g24365ee (24365ee)
AUTHOR(S):
Kamil Trzciński <ayufan#ayufan.eu>
COMMANDS:
exec execute a build locally
[...]
GLOBAL OPTIONS:
--debug debug mode [$DEBUG]
[...]
As you can see, the exec command is to execute a build locally.
Even though there was an issue to deprecate the current gitlab-runner exec behavior, it ended up being reconsidered and a new version with greater features will replace the current exec functionality.
Note that this process is to use your own machine to run the tests using docker containers. This is not to define custom runners. To do so, just go to your repo's CI/CD settings and read the documentation there. If you wanna ensure your runner is executed instead of one from gitlab.com, add a custom and unique tag to your runner, ensure it only runs tagged jobs and tag all the jobs you want your runner to be responsible of.
I use this docker-based approach:
Edit: 2022-10
docker run --entrypoint bash --rm -w $PWD -v $PWD:$PWD -v /var/run/docker.sock:/var/run/docker.sock gitlab/gitlab-runner:latest -c 'git config --global --add safe.directory "*";gitlab-runner exec docker test'
For all git versions > 2.35.2. You must add safe.directory within the container to avoid fatal: detected dubious ownership in repository at.... This also true for patched git versions < 2.35.2. The old command will not work anymore.
Details
0. Create a git repo to test this answer
mkdir my-git-project
cd my-git-project
git init
git commit --allow-empty -m"Initialize repo to showcase gitlab-runner locally."
1. Go to your git directory
cd my-git-project
2. Create a .gitlab-ci.yml
Example .gitlab-ci.yml
image: alpine
test:
script:
- echo "Hello Gitlab-Runner"
3. Create a docker container with your project dir mounted
docker run -d \
--name gitlab-runner \
--restart always \
-v $PWD:$PWD \
-v /var/run/docker.sock:/var/run/docker.sock \
gitlab/gitlab-runner:latest
(-d) run container in background and print container ID
(--restart always) or not?
(-v $PWD:$PWD) Mount current directory into the current directory of the container - Note: On Windows you could bind your dir to a fixed location, e.g. -v ${PWD}:/opt/myapp. Also $PWD will only work at powershell not at cmd
(-v /var/run/docker.sock:/var/run/docker.sock) This gives the container access to the docker socket of the host so it can start "sibling containers" (e.g. Alpine).
(gitlab/gitlab-runner:latest) Just the latest available image from dockerhub.
4. Execute with
Avoid fatal: detected dubious ownership in repository at... More info
docker exec -it -w $PWD gitlab-runner git config --global --add safe.directory "*"
Actual execution
docker exec -it -w $PWD gitlab-runner gitlab-runner exec docker test
# ^ ^ ^ ^ ^ ^
# | | | | | |
# (a) (b) (c) (d) (e) (f)
(a) Working dir within the container. Note: On Windows you could use a fixed location, e.g. /opt/myapp.
(b) Name of the docker container
(c) Execute the command "gitlab-runner" within the docker container
(d)(e)(f) run gitlab-runner with "docker executer" and run a job named "test"
5. Prints
...
Executing "step_script" stage of the job script
$ echo "Hello Gitlab-Runner"
Hello Gitlab-Runner
Job succeeded
...
Note: The runner will only work on the commited state of your code base. Uncommited changes will be ignored. Exception: The .gitlab-ci.yml itself does not have be commited to be taken into account.
Note: There are some limitations running locally. Have a look at limitations of gitlab runner locally.
I'm currently working on making a gitlab runner that works locally.
Still in the early phases, but eventually it will become very relevant.
It doesn't seem like gitlab want/have time to make this, so here you go.
https://github.com/firecow/gitlab-runner-local
If you are running Gitlab using the docker image there: https://hub.docker.com/r/gitlab/gitlab-ce, it's possible to run pipelines by exposing the local docker.sock with a volume option: -v /var/run/docker.sock:/var/run/docker.sock. Adding this option to the Gitlab container will allow your workers to access to the docker instance on the host.
The GitLab runner appears to not work on Windows yet and there is an open issue to resolve this.
So, in the meantime I am moving my script code out to a bash script, which I can easily map to a docker container running locally and execute.
In this case I want to build a docker container in my job, so I create a script 'build':
#!/bin/bash
docker build --pull -t myimage:myversion .
in my .gitlab-ci.yaml I execute the script:
image: docker:latest
services:
- docker:dind
before_script:
- apk add bash
build:
stage: build
script:
- chmod 755 build
- build
To run the script locally using powershell I can start the required image and map the volume with the source files:
$containerId = docker run --privileged -d -v ${PWD}:/src docker:dind
install bash if not present:
docker exec $containerId apk add bash
Set permissions on the bash script:
docker exec -it $containerId chmod 755 /src/build
Execute the script:
docker exec -it --workdir /src $containerId bash -c 'build'
Then stop the container:
docker stop $containerId
And finally clean up the container:
docker container rm $containerId
Another approach is to have a local build tool that is installed on your pc and your server at the same time.
So basically, your .gitlab-ci.yml will basically call your preferred build tool.
Here an example .gitlab-ci.yml that i use with nuke.build:
stages:
- build
- test
- pack
variables:
TERM: "xterm" # Use Unix ASCII color codes on Nuke
before_script:
- CHCP 65001 # Set correct code page to avoid charset issues
.job_template: &job_definition
except:
- tags
build:
<<: *job_definition
stage: build
script:
- "./build.ps1"
test:
<<: *job_definition
stage: test
script:
- "./build.ps1 test"
variables:
GIT_CHECKOUT: "false"
pack:
<<: *job_definition
stage: pack
script:
- "./build.ps1 pack"
variables:
GIT_CHECKOUT: "false"
only:
- master
artifacts:
paths:
- output/
And in nuke.build i've defined 3 targets named like the 3 stages (build, test, pack)
In this way you have a reproducible setup (all other things are configured with your build tool) and you can test directly the different targets of your build tool.
(i can call .\build.ps1 , .\build.ps1 test and .\build.ps1 pack when i want)
I am on Windows using VSCode with WSL
I didn't want to register my work PC as a runner so instead I'm running my yaml stages locally to test them out before I upload them
$ sudo apt-get install gitlab-runner
$ gitlab-runner exec shell build
yaml
image: node:10.19.0 # https://hub.docker.com/_/node/
# image: node:latest
cache:
# untracked: true
key: project-name
# key: ${CI_COMMIT_REF_SLUG} # per branch
# key:
# files:
# - package-lock.json # only update cache when this file changes (not working) #jkr
paths:
- .npm/
- node_modules
- build
stages:
- prepare # prepares builds, makes build needed for testing
- test # uses test:build specifically #jkr
- build
- deploy
# before_install:
before_script:
- npm ci --cache .npm --prefer-offline
prepare:
stage: prepare
needs: []
script:
- npm install
test:
stage: test
needs: [prepare]
except:
- schedules
tags:
- linux
script:
- npm run build:dev
- npm run test:cicd-deps
- npm run test:cicd # runs puppeteer tests #jkr
artifacts:
reports:
junit: junit.xml
paths:
- coverage/
build-staging:
stage: build
needs: [prepare]
only:
- schedules
before_script:
- apt-get update && apt-get install -y zip
script:
- npm run build:stage
- zip -r build.zip build
# cache:
# paths:
# - build
# <<: *global_cache
# policy: push
artifacts:
paths:
- build.zip
deploy-dev:
stage: deploy
needs: [build-staging]
tags: [linux]
only:
- schedules
# # - branches#gitlab-org/gitlab
before_script:
- apt-get update && apt-get install -y lftp
script:
# temporarily using 'verify-certificate no'
# for more on verify-certificate #jkr: https://www.versatilewebsolutions.com/blog/2014/04/lftp-ftps-and-certificate-verification.html
# variables do not work with 'single quotes' unless they are "'surrounded by doubles'"
- lftp -e "set ssl:verify-certificate no; open mediajackagency.com; user $LFTP_USERNAME $LFTP_PASSWORD; mirror --reverse --verbose build/ /var/www/domains/dev/clients/client/project/build/; bye"
# environment:
# name: staging
# url: http://dev.mediajackagency.com/clients/client/build
# # url: https://stg2.client.co
when: manual
allow_failure: true
build-production:
stage: build
needs: [prepare]
only:
- schedules
before_script:
- apt-get update && apt-get install -y zip
script:
- npm run build
- zip -r build.zip build
# cache:
# paths:
# - build
# <<: *global_cache
# policy: push
artifacts:
paths:
- build.zip
deploy-client:
stage: deploy
needs: [build-production]
tags: [linux]
only:
- schedules
# - master
before_script:
- apt-get update && apt-get install -y lftp
script:
- sh deploy-prod
environment:
name: production
url: http://www.client.co
when: manual
allow_failure: true
The idea is to keep check commands outside of .gitlab-ci.yml. I use Makefile to run something like make check and my .gitlab-ci.yml runs the same make commands that I use locally to check various things before committing.
This way you'll have one place with all/most of your commands (Makefile) and .gitlab-ci.yml will have only CI-related stuff.
I have written a tool to run all GitLab-CI job locally without have to commit or push, simply with the command ci-toolbox my_job_name.
The URL of the project : https://gitlab.com/mbedsys/citbx4gitlab
Years ago I build this simple solution with Makefile and docker-compose to run the gitlab runner in docker, you can use it to execute jobs locally as well and should work on all systems where docker works:
https://gitlab.com/1oglop1/gitlab-runner-docker
There are few things to change in the docker-compose.override.yaml
version: "3"
services:
runner:
working_dir: <your project dir>
environment:
- REGISTRATION_TOKEN=<token if you want to register>
volumes:
- "<your project dir>:<your project dir>"
Then inside your project you can execute it the same way as mentioned in other answers:
docker exec -it -w $PWD runner gitlab-runner exec <commands>..
I recommend using gitlab-ci-local
https://github.com/firecow/gitlab-ci-local
It's able to run specific jobs as well.
It's a very cool project and I have used it to run simple pipelines on my laptop.