Details
-
Task
-
Status: Closed (View Workflow)
-
Major
-
Resolution: Fixed
-
N/A
-
None
-
None
Description
For install and upgrade tests we use LibVirt managed KVMs.
We currently have one LibVirt worker that handles these types of workers (bg-bbw5-x64). We need an automatic way of creating VMs for all major operating systems and platforms that we are supporting (basically, for all the OSs and platforms for which we create Docker images - see https://github.com/MariaDB/mariadb.org-tools/blob/master/.github/workflows/bb_containers.yml)
VM creation
Create the image and configure LibVirt to use it. The current naming scheme consists in "buildbot-osname". Probably we need to adjust that.
Setup the base image (see VM installed packages below)
Configure LibVirt hook to always start from a clean VM
VM installed packages
The VMs need to be clean and have nothing installed except:
- buildbot-worker - Python package that needs to be installed similar to the steps in the Dockerfile creation (https://github.com/MariaDB/mariadb.org-tools/blob/bd5c4a6a072e8c9232e203ec985efcf25843fd1d/buildbot.mariadb.org/ci_build_images/common.Dockerfile#L29)
- configure a buildbot.tac that is unique for each worker similar to:
import fnmatch
|
import os
|
import sys
|
|
|
from twisted.application import service
|
from twisted.python.log import FileLogObserver
|
from twisted.python.log import ILogObserver
|
|
|
from buildbot_worker.bot import Worker
|
|
|
# setup worker
|
basedir = os.path.abspath(os.path.dirname(__file__))
|
application = service.Application('buildbot-worker')
|
|
|
|
|
application.setComponent(ILogObserver, FileLogObserver(sys.stdout).emit)
|
# and worker on the same process!
|
buildmaster_host = os.environ.get("BUILDMASTER", 'buildbot.mariadb.org')
|
port = int(os.environ.get("BUILDMASTER_PORT", 9990))
|
workername = os.environ.get("WORKERNAME", 'buildbot-ubuntu1804')
|
passwd = os.environ.get("WORKERPASS", 'PASSWORD')
|
|
|
# delete the password from the environ so that it is not leaked in the log
|
blacklist = os.environ.get("WORKER_ENVIRONMENT_BLACKLIST", "WORKERPASS").split()
|
for name in list(os.environ.keys()):
|
for toremove in blacklist:
|
if fnmatch.fnmatch(name, toremove):
|
del os.environ[name]
|
|
|
keepalive = 600
|
umask = None
|
maxdelay = 300
|
allow_shutdown = None
|
maxretries = 10
|
|
|
s = Worker(buildmaster_host, port, workername, passwd, basedir,
|
keepalive, umask=umask, maxdelay=maxdelay,
|
allow_shutdown=allow_shutdown, maxRetries=maxretries)
|
s.setServiceParent(application)
|
Attention the WORKERNAME must match the name of the LibVirt VM. The password also needs to be set.
- Ideally, include a possibility to easily add packages if needed.
Attachments
Issue Links
- is part of
-
-
- Closed
-
Activity
- POC for generating BB VM, next:
- use ansible to deploy and configure VM;
- improve the bb_worker_vm role;
- test full destroy;
- test deploy from scratch (ok with debian/ubuntu).
- Finish cloud-init POC (test other OS):
- test centos;
- test fedora;
- handle case where VM are created manually (do not destroy them, and keep qemu hook in place);
- create PR (https://gitlab.com/mariadb/sysadmin/-/merge_requests/28);
- activate VM's with Vlad;
- improve installation mechanism (pb with unbuntu18.04 and fedora35);
- still pb with centos-8-stream.
ARM is now correctly handled by the ansible role deployment system.
https://gitlab.com/mariadb/sysadmin/-/commit/c0667b4def7cb5947fb0e9cbab2a11a7e750c6ce
Following VMs are now available to test installation and upgrade on ARM in BB:
- bb-centos-7
- bb-debian-10
- bb-debian-11
- bb-debian-sid
- bb-fedora-34
- bb-fedora-35
- bb-ubuntu-1804
- bb-ubuntu-2004
- bb-ubuntu-2204
Following VM are now also available for activation (ARM and x64)
- fedora-36
- centos-8-stream
- centos-9-stream
TODO:
- rhel-7
- rhel-8
Rhel is now also available:
- rhel 7 (x86 - i can't find rhel7 images for ARM so far.)
- rhel 8 (x86/ARM)
SELinux is now in Enforcing mode for libvirt VM:
https://gitlab.com/mariadb/sysadmin/-/commit/ccb90fccadbaa11bf45ea3f629a034f689b1cffb
First attempt to deploy PPC libvirt workers.
- same problem as for containers, the installation of cryptography module needs rust/libssl-dev and takes ages!;
- try to generate wheel for PPC using manylinux containers;
- this is not as simple as I though (there is some python magic that I do not understand);
- I will try to do as they are in https://github.com/pyca/cryptography/blob/main/.github/workflows/wheel-builder.yml
- this can potentially simplify and drastically speed up quay.io image generations and libvirt deployments.
ppc64le vm have been deployed, here is the list:
- bb-db-p9-bbw1-centos-7
- bb-db-p9-bbw1-centos-8-stream
- bb-db-p9-bbw1-centos-9-stream
- bb-db-p9-bbw1-debian-10
- bb-db-p9-bbw1-debian-11
- bb-db-p9-bbw1-debian-sid
- bb-db-p9-bbw1-rhel-8
- bb-db-p9-bbw1-rhel-9
- bb-db-p9-bbw1-ubuntu-1804
- bb-db-p9-bbw1-ubuntu-2004
- bb-db-p9-bbw1-ubuntu-2204
The automated creation process is working for ppc/amd/arm.
Remaining: deploy s390x machines.
First deployment created by faust https://gitlab.com/mariadb/sysadmin/-/merge_requests/28