Project

General

Profile

Actions

Wiki » History » Revision 137

« Previous | Revision 137/309 (diff) | Next »
okurz, 2021-12-17 13:47


Introduction

This is the organisation wiki for the openQA Project.
The source code is hosted in the os-autoinst github project, especially openQA itself and the main backend os-autoinst

If you are interested in the tests for SUSE/openSUSE products take a look into the openqatests project.

If you are looking for entry level issues to contribute to the backend, take a look at this search query

Organisational

ticket workflow

Picture: http://imagebin.suse.de/2127/img

The following ticket statuses are used together and their meaning is explained:

  • New: No one has worked on the ticket (e.g. the ticket has not been properly refined) or no one is feeling responsible for the work on this ticket.
  • Workable: The ticket has been refined and is ready to be picked.
  • In Progress: Assignee is actively working on the ticket.
  • Resolved: The complete work on this issue is done and the according issue is supposed to be fixed as observed (Should be updated together with a link to a merged pull request or also a link to an production openQA showing the effect)
  • Feedback: Further work on the ticket is blocked by open points or is awaiting for the feedback to proceed. Sometimes also used to ask Assignee about progress on inactivity.
  • Blocked: Further work on the ticket is blocked by some external dependency (e.g. bugs, not implemented features). There should be a link to another ticket, bug, trello card, etc. where it can be seen what the ticket is blocked by.
  • Rejected: The issue is considered invalid, should not be done, is considered out of scope.
  • Closed: As this can be set only by administrators it is suggested to not use this status.

It is good practice to update the status together with a comment about it, e.g. a link to a pull request or a reason for reject.

ticket categories

  • Concrete Bugs: Regressions, crashes, error messages
  • Feature requests: Ideas or wishes for extension, enhancement, improvement
  • Organisational: Organisational tasks within the project(s), not directly code related
  • Support: Support of users, usage problems, questions

Please avoid the use of other, deprecated categories

Suggestion by okurz: I recommend to avoid the word "bug" in our categories because of the usual "is it a bug or a feature" struggle. Instead I suggest to strictly define "Regressions & Crashes" to clearly separate "it used to work in before" from "this was never part of requirements" for Features. Any ticket of this category also means that our project processes missed something so we have points for improvements, e.g. extend things to look out for in code review.

Epics and Sagas

[epic]s and [saga]s belong to the "coordination" tracker, project contributors are not required to follow this convention but the tracker may be changed automagically in the future: http://mailman.suse.de/mailman/private/qa-sle/2020-October/002722.html

ticket templates

You can use these templates to fill in tickets and further improve them with more detail over time. Copy the code block, paste it into a new issue, replace every block marked with "<…>" with your content or delete if not appropriate.

Defects

Subject: <Short description, example: "openQA dies when triggering any Windows ME tests">

## Observation
<description of what can be observed and what the symptoms are, provide links to failing test results and/or put short blocks from the log output here to visualize what is happening>

## Steps to reproduce
* <do this>
* <do that>
* <observe result>

## Problem
<problem investigation, can also include different hypotheses, should be labeled as "H1" for first hypothesis, etc.>

## Suggestion
* <what to do as a first step>
* <Fix the actual problem>
* <Consider fixing the design>
* <Consider fixing the team's process>
* <Consider to explore further>

## Workaround
<example: retrigger job>

example ticket: #10526

For tickets referencing "auto_review" see
https://github.com/os-autoinst/scripts/blob/master/README.md#auto-review---automatically-detect-known-issues-in-openqa-jobs-label-openqa-jobs-with-ticket-references-and-optionally-retrigger
for a suggested template snippet.

Feature requests

Subject: <Short description, example: "grub3 btrfs support" (feature)>

## User story
<As a <role>, I want to <do an action>, to <achieve which goal> >

## Acceptance criteria
* <**AC1:** the first acceptance criterion that needs to be fulfilled to do this, example: Clicking "restart button" causes restart of the job>
* <**AC2:** also think about the "not-actions", example: other jobs are not affected>

## Tasks
* <first task to do as an easy starting point>
* <what do do next, all tasks optionally with an effort estimation in hours, e.g. "(0.5-2h)">
* <optional: mark "optional" tasks>

## Further details
<everything that does not fit into above sections>

example ticket: #10212

Further decision steps working on test issues

Test issues could be one of the following sources. Feel free to use the following template in tickets as well

## Problem
* **H1** The product has changed
 * **H1.1** product changed slightly but in an acceptable way without the need for communication with DEV+RM --> adapt test
 * **H1.2** product changed slightly but in an acceptable way found after feedback from RM --> adapt test
 * **H1.3** product changed significantly --> after approval by RM adapt test

* **H2** Fails because of changes in test setup
 * **H2.1** Our test hardware equipment behaves different
 * **H2.2** The network behaves different

* **H3** Fails because of changes in test infrastructure software, e.g. os-autoinst, openQA
* **H4** Fails because of changes in test management configuration, e.g. openQA database settings
* **H5** Fails because of changes in the test software itself (the test plan in source code as well as needles)
* **H6** Sporadic issue, i.e. the root problem is already hidden in the system for a long time but does not show symptoms every time

pull request handling on github

As a reviewer of pull requests on github for all related repositories, e.g. https://github.com/os-autoinst/os-autoinst-distri-opensuse/pulls, apply labels in case PRs are open for a longer time and can not be merged so that we keep our backlog clean and know why PRs are blocked.

  • notready: Triaged as not ready yet for merging, no (immediate) reaction by the reviewee, e.g. when tests are missing, other scenarios break, only tested for one of SLE/TW
  • wip: Marked by the reviewee itself as "[WIP]" or "[DO-NOT-MERGE]" or similar
  • question: Questions to the reviewee, not answered yet

Where to contribute?

If you want to help openQA development you can take a look into the existing issues. There are also some "always valid" tasks to be working on:

  • improve test coverage:
    • user story: As openqa backend as well as test developer I want better test coverage of our projects to reduce technical debt
    • acceptance criteria: test coverage is significantly higher than before
    • suggestions: check current coverage in each individual project (os-autoinst/openQA/os-autoinst-distri-opensuse) and add tests as necessary

Use cases

The following use cases 1-6 have been defined within a SUSE workshop (others have been defined later) to clarify how different actors work with openQA. Some of them are covered already within openQA quite well, some others are stated as motivation for further feature development.

Use case 1

User: QA-Project Managment
primary actor: QA Project Manager, QA Team Leads
stakeholder: Directors, VP
trigger: product milestones, providing a daily status
user story: „As a QA project manager I want to check on a daily basis the „openQA Dashboard“ to get a summary/an overall status of the „reviewers results“ in order to take the right actions and prioritize tasks in QA accordingly.“

Use case 2

User: openQA-Admin
primary actor: Backend-Team
stakeholder: Qa-Prjmgr, QA-TL, openQA Tech-Lead
trigger: Bugs, features, new testcases
user story: „As an openQA admin I constantly check in the web-UI the system health and I manage its configuration to ensure smooth operation of the tool.“

Use case 3

User: QA-Reviewer
primary actor: QA-Team
stakeholder: QA-Prjmgr, Release-Mgmt, openQA-Admin
trigger: every new build
user story: „As an openQA-Reviewer at any point in time I review on the webpage of openQA the overall status of a build in order to track and find bugs, because I want to find bugs as early as possible and report them.“

Use case 4

User: Testcase-Contributor
primary actor: All development teams, Maintenance QA
stakeholder: QA-Reviewer, openQA-Admin, openQA Tech-Lead
trigger: features, new functionality, bugs, new product/package
user story: „As developer when there are new features, new functionality, bugs, new product/package in git I contribute my testcases because I want to ensure good quality submissions and smooth product integration.“

Use case 5

User: Release-Mgmt
primary actor: Release Manager
stakeholder: Directors, VP, PM, TAMs, Partners
trigger: Milestones
user story: „As a Release-Manager on a daily basis I check on a dashboard for the product health/build status in order to act early in case of failures and have concrete and current reports.“

Use case 6

User: Staging-Admin
primary actor: Staging-Manager for the products
stakeholder: Release-Mgmt, Build-Team
trigger: every single submission to projects
user story: „As a Staging-Manager I review the build status of packages with every staged submission to the „staging projects“ in the „staging dashboard“ and the test-status of the pre-integrated fixes, because I want to identify major breakage before integration to the products and provide fast feedback back to the development.“

Use case 7

User: Bug investigator
primary actor: Any bug assignee for openQA observed bugs
stakeholder: Developer
trigger: bugs
user story: „As a developer that has been assigned a bug which has been observed in openQA I can review referenced tests, find a newer and the most recent job in the same scenario, understand what changed since the last successful job, what other jobs show same symptoms to investigate the root cause fast and use openQA for verification of a bug fix.“

Thoughts about categorizing test results, issues, states within openQA

by okurz

When reviewing test results it is important to distinguish between different causes of "failed tests"

Nomenclature

Test status categories

A common definition about the status of a test regarding the product it tests: "false|true positive|negative" as described on https://en.wikipedia.org/wiki/False_positives_and_false_negatives. "positive|negative" describes the outcome of a test ("positive": test signals presence of issue; "negative": no signal) whereas "false|true" describes the conclusion of the test regarding the presence of issues in the SUT or product in our case ("true": correct reporting; "false": incorrect reporting), e.g. "true negative", test successful, no issues detected and there are no issues, product is working as expected by customer. Another example: Think of testing as of a fire alarm. An alarm (event detector) should only go off (be "positive") if there is a fire (event to detect) --> "true positive" whereas if there is no fire there should be no alarm --> "true negative".

Another common but potentially ambiguous categorization:

  • broken: the test is not behaving as expected (Ambiguity: "as expected" by whom?) --> commonly a "false positive", can also be "false negative" but hard to detect
  • failing: the test is behaving as expected, but the test output is a fail --> "true positive"
  • working: the test is behaving as expected (with no comment regarding the result, though some might ambiguously imply 'result is negative')
  • passing: the test is behaving as expected, but the result is a success --> "true negative"

If in doubt declare a test as "broken". We should review the test and examine if it is behaving as expected.

Be careful about "positive/negative" as some might also use "positive" to incorrectly denote a passing test (and "negative" for failing test) as an indicator of "working product" not an indicator about "issue present". If you argue what is "used in common speech" think about how "false positive" is used as in "false alarm" --> "positive" == "alarm raised", also see https://narainko.wordpress.com/2012/08/26/understanding-false-positive-and-false-negative/

Priorization of work regarding categories

In this sense development+QA want to accomplish a "true negative" state whenever possible (no issues present, therefore none detected). As QA and test developers we want to prevent "false positives" ("false alarms" declaring a product as broken when it is not but the test failed for other reasons), also known as "type I error" and "false negatives" (a product issue is not catched by tests and might "slip through" QA and at worst is only found by an external outside customer) also known as "type II error". Also see https://en.wikipedia.org/wiki/Type_I_and_type_II_errors. In the context of openQA and system testing paired with screen matching a "false positive" is much more likely as the tests are very susceptible to subtle variations and changes even if they should be accepted. So when in doubt, create an issue in progress, look at it again, and find that it was a false alarm, rather than wasting more peoples time with INVALID bug reports by believing the product to be broken when it isn't. To quote Richard Brown: "I […] believe this is the route to ongoing improvement - if we have tests which produce such false alarms, then that is a clear indicator that the test needs to be reworked to be less ambiguous, and that IS our job as openQA developers to deal with".

Further categorization of statuses, issues and such in testing, especially automatic tests

By okurz

This categorization scheme is meant to help in communication in either written or spoken discussions being simple, concise, easy to remember while unambiguous in every case.
While used for naming it should also be used as a decision tree and can be followed from the top following each branch.

Categorization scheme

To keep it simple I will try to go in steps of deciding if a potential issue is of one of two categories in every step (maybe three) and go further down from there. The degree of further detailing is not limited, i.e. it can be further extended. Naming scheme should follow arabic number (for two levels just 1 and 2) counting schemes added from the right for every additional level of decision step and detail without any separation between the digits, e.g. "1111" for the first type in every level of detail up to level four. Also, I am thinking of giving the fully written form phonetic name to unambiguously identify each on every level as long as not more individual levels are necessary. The alphabet should be reserved for higher levels and higher priority types.
Every leaf of the tree must have an action assigned to it.

1 failed (ZULU)
11 new (passed->failed) (YANKEE)
111 product issue ("true positive") (WHISKEY)
1111 unfiled issue (SIERRA)
11111 hard issue (openqa fail) (KILO)
111121 critical / potential ship stopper (INDIA) --> immediately file bug report with "ship_stopper?" flag; opt. inform RM directly
111122 non-critical hard issue (HOTEL) --> file bug report
11112 soft issue (openqa softfail on job level, not on module level) (JULIETT) --> file bug report on failing test module
1112 bugzilla bug exists (ROMEO)
11121 bug was known to openqa / openqa developer --> cross-reference (bug->test, test->bug) AND raise review process issue, improve openqa process
11122 bug was filed by other sources (e.g. beta-tester) --> cross-reference (bug->test, test->bug)
112 test issue ("false positive") (VICTOR)
1121 progress issue exists (QUEBEC) --> cross-reference (issue->test, test->issue)
1122 unfiled test issue (PAPA)
11221 easy to do w/o progress issue
112211 need needles update --> re-needle if sure, TODO how to notify?
112212 pot. flaky, timeout
1122121 retrigger yields PASS --> comment in progress about flaky issue fixed
1122122 reproducible on retrigger --> file progress issue
11222 needs progress issue filed --> file progress issue
12 existing / still failing (failed->failed) (XRAY)
121 product issue (UNIFORM)
1211 unfiled issue (OSCAR) --> file bug report AND raise review process issue (why has it not been found and filed?)
1212 bugzilla bug exists (NOVEMBER) --> ensure cross-reference, also see rules for 1112 ROMEO
122 test issue (TANGO)
1221 progress issue exists (MIKE) --> monitor, if persisting reprioritize test development work
1222 needs progress issue filed (LIMA) --> file progress issue AND raise review process issue, see 1211 OSCAR
2 passed (ALFA)
21 stable (passed->passed) (BRAVO)
211 existing "true negative" (DELTA) --> monitor, maybe can be made stricter
212 existing "false negative" (ECHO) --> needs test improvement
22 fixed (failed->passed) (CHARLIE)
222 fixed "true negative" (FOXTROTT) --> TODO split monitor, see 211 DELTA
2221 was test issue --> close progress issue
2222 was product issue
22221 no bug report exists --> raise review process issue (why was it not filed?)
22222 bug report exists
222221 was marked as RESOLVED FIXED
221 fixed but "false negative" (GOLF) --> potentially revert test fix, also see 212 ECHO

Priority from high to low: INDIA->OSCAR->HOTEL->JULIETT->…

Proposals for uses of labels

With Show bug or label icon on overview if labeled (gh#550) it is possible to add custom labels just by writing them. Nevertheless, a convention should be found for a common benefit. Beware that labels are also automatically carried over with (Carry over labels from previous jobs in same scenario if still failing [gh#564])(https://github.com/os-autoinst/openQA/pull/564) which might make consistent test failures less visible when reviewers only look for test results without labels or bugrefs. Labels are not anymore automatically carried over (gh#1071).

List of proposed labels with their meaning and where they could be applied.

  • fixed_<build_ref>: If a test failure is already fixed in a more recent build and no bug reference is known, use this label together with a reference to a more recent passed test run in the same scenario. Useful for reviewing older builds. Example (https://openqa.suse.de/tests/382518#comments):
label:fixed_Build1501

t#382919
  • needles_added: In case needles were missing for test changes or expected product changes caused needle matching to fail, use this label with a reference to the test PR or a proper reasoning why the needles were missing and how you added them. Example (https://openqa.suse.de/tests/388521#comments):
label:needles_added

needles for https://github.com/os-autoinst/os-autoinst-distri-opensuse/pull/1353 were missing, added by jpupava in the meantime.

s390x Test Organisation

See the following picture for a graphical overview of the current s390x test infrastructure at SUSE:

SUSE s390x test infrastructure

Upgrades

on z/VM

special Requirements

Due to the lack of proper use of hdd-images on zVM, we need to workaround this with having a dedicated worker_class aka a dedicated Host where we run two jobs with START_AFTER_TEST,
the first one which installs the basesystem we want to have upgraded and a second one which is doing the actually upgrade (e.g migration_offline_sle12sp2_zVM_preparation and migration_offline_sle12sp2_zVM)

Since we encountered issues with randomly other preparation jobs are started in between there, we need to ensure that we have one complete chain for all migration jobs running on one worker, that means for example:

  1. migration_offline_sle12sp2_zVM_preparation
  2. migration_offline_sle12sp2_zVM (START_AFTER_TEST=#1)
  3. migration_offline_sle12sp2_allpatterns_zVM_preparation (START_AFTER_TEST=#2)
  4. migration_offline_sle12sp2_allpatterns_zVM
  5. ...

This scheme ensures that all actual Upgrade jobs are finding the prepared system and are able to upgrade it

on z/KVM

No special requirements anymore, see details in #18016

Automated z/VM LPAR installation with openQA using qnipl

There is an ongoing effort to automate the LPAR creation and installation on z/VM. A first idea resulted in the creation of qnipl. qnipl enables one to boot a very slim initramfs from a shared medium (e.g. shared SCSI-disks) and supply it with the needed parameters to chainload a "normal SLES installation" using kexec.
This method is required for z/VM because snipl (Simple network initial program loader) can only load/boot LPARs from specific disks, not network resources.

Setup

  1. Get a shared disk for all your LPARs
    • Normally this can easily done by infra/gschlotter
    • Disks needs to be connected to all guests which should be able to network-boot
  2. Boot a fully installed SLES on one of the LPARs to start preparing the shared-disk
  3. Put a DOS partition table on the disk and create one single, large partition on there
  4. Put a FS on there. Our first test was on ext2 and it worked flawlessly in our attempts
  5. Install zipl (The s390x bootloader from IBM) on this partition
    • A simple and sufficient config can be found in poo#33682
  6. clone qnipl to your dracut modules (e.g. /usr/lib/dracut/modules.d/95qnipl)
  7. Include the module named qnipl to your dracut modules for initramfs generation
    • e.g. in /etc/dracut.conf.d/99-qnipl.conf add: add_dracutmodules+=qnipl
  8. Generate your initramfs (e.g. dracut -f -a "url-lib qnipl" --no-hostonly-cmdline /tmp/custom_initramfs)
    • Put the initramfs next to your kernel binary on the partition you want to prepare
  9. From now on you can use snipl to boot any LPAR connected with this shared disk from network
    • example: snipl -f ./snipl.conf -s P0069A27-LP3 -A fa00 --wwpn_scsiload 500507630713d3b3 --lun_scsiload 4001401100000000 --ossparms_scsiload "install=http://openqa.suse.de/assets/repo/SLE-15-Installer-DVD-s390x-Build533.2-Media1 hostip=10.161.159.3/20 gateway=10.161.159.254 Nameserver=10.160.0.1 Domain=suse.de ssh=1 regurl=http://all-533.2.proxy.scc.suse.de"
    • --ossparms_scsiload is then evaluated and used by qnipl to kexec into the installer with the (for the installer) needed parameters

Further details

Further details can also be found in the github repo. Pull requests, questions and ideas always welcome!

Infrastructure setup for o3 (openqa.opensuse.org) and osd (openqa.suse.de)

o3 (openqa.opensuse.org)

o3 consists of a VM running the web UI and physical worker machines. The VM for o3 has netapp backed storage on rotating disk so less performant than SSD but cheaper. So eventually we might have the possibility to use SSD based storage. Currently there are four virtual storage devices provided to o3 totalling to 10 TB.

Automatic update of o3

o3 is automatically deployed on a daily base, that includes both the webUI host as well as the workers.

Automatic update of o3 webUI host

Done with cron job in /etc/cron.d/auto-update

Recurring automatic update of openQA workers

All o3 workers (except power8) apply a daily automatic update and are "Transactional Servers" running openSUSE Leap. power8 is non-transactional with a weekly update every Sunday.

This was for a number of reasons including:

  • Getting all the machines consistent after a few years of drift
  • Making it easier to keep them consistent by leveraging a read only root filesystem
  • Guaranteeing rollbackability by using transactional updates

This was done by rbrown also to fulfill the prerequisite to getting them viable for multi-machine testing

These systems currently patch themselves and reboot automatically in the default maintenance window of 0330-0500 CET/CEST.

On problems this could be changed in the following way:

  • Edit the maintenance window in /etc/rebootmgr.conf
  • Disable the automatic reboot by "systemctl disable rebootmgr.service"
  • Disable the automatic patching by "systemctl disable transactional-update.timer"

SUSE employees have access to the bootmenu for the openQA worker machines, e.g. openqaworker1 and openqaworker4 via openqaworker1- ipmi.suse.de and openqaworker4-ipmi.suse.de which are both connected to the r&d network. For imagetester one would need to go through SUSE-IT in an unlikely event of a boot-preventing update. "snapper rollback" can be executed from a booted, functionally operative machine which one can ssh into.

To execute commands manually on all workers one can do for example the following:

for i in aarch64 openqaworker1 openqaworker4 openqaworker7 power8 imagetester rebel; do echo $i && ssh root@$i "(transactional-update -n dup || zypper -n dup) && reboot" ; done

mind the correct list of machines.

For manual investigation https://github.com/kubic-project/microos-toolbox can be helpful

Rollback of updates

Updates on workers can be rolled back using transactional-update affecting the transactional workers (others are likely not updated that often):

for i in aarch64 openqaworker1 openqaworker4 openqaworker7 power8 imagetester rebel; do echo $i && ssh root@$i "transactional-update rollback last && reboot"; done

Updates on the central webUI host openqa.opensuse.org can be rolled back by using either older variants of packages that receive maintenance updates or using the locally cached packages in e.g. /var/cache/zypp/packages/devel_openQA/noarch using zypper in --oldpackage, similar to https://github.com/os-autoinst/openQA/blob/master/script/openqa-rollback#L39

Accessing o3 infrastructure

The o3 webui host as well as the workers within the o3 infrastructure can be accessed over ssh by using ssh -p 2213 gate.opensuse.org. Ask one of the existing admins to put your ssh key there to be able to login.

To give access for a new user an existing admin can do the following:

sudo useradd -G users,trusted --create-home $user
echo "$ssh_key_from_user" | sudo tee -a /home/$user/.ssh/authorized_keys

SSH configuration

To easily access all hosts behind the jump host you can use the following config for your ssh client (~.ssh/config):

Host ariel
  HostName gate.opensuse.org
  Port 2213

Host *.opensuse.org
  ProxyCommand ssh -q -A -x ariel -W %h:%p

A word of warning: be aware that this enables agent-forwarding to at least the jumphost. Please read up for yourself if and how bad you consider the security implications of doing so.

Debugging qemu SUTs in openqa.opensuse.org

SUT: System Under Test

os-autoinst starts qemu with network type that doesn't allow access from the outside, so ssh is not possible. But, qemu is started with a VNC channel available from the host (the openQA-worker).
Running vncviewer inside a headless server is useless, but it is possible to use gate.opensuse.org as a jump host and SSH port forwarding to start vncviewer client from your desktop environment and connect to the VNC channel of the qemu SUT.

ssh -p 2213 -L LOCAL_PORT:WORKER_HOSTNAME:QEMU_VNC_PORT USERNAME@gate.opensuse.org

For example, if user bernhard, wants to connect to openqaworker7:11, and wants to use local port 43043
Being the IP of openqaworker7 192.168.112.12
And the VNC channel port of openqa-worker@11 6001 (5990 + 11)

1. Create SSH tunnel with port forwarding
  • on laptop shell 1: ssh -p 2213 -L 43043:192.168.112.12:6001 bernhard@gate.opensuse.org
  • Keep shell open to keep the tunnel open and the port forwarding
2. Open vncviewer
  • on laptop shell 2: vncviewer -Shared localhost:43043
  • -shared is needed to not kick the VNC connection of os-autoinst. If it is kicked, the job will terminate and the qemu process will be killed.

AArch64 specific configurations on o3

On o3, the aarch64 workers need additional configuration.

Setup HugePages

You need to setup HugePages support to improve performances with qemu VM and to match current aarch64 MACHINE configuration.
For the D05 machine, the configuration is: 40 pages with a size of 1G.
If there are some permissions issues on /dev/hugepages/, check https://progress.opensuse.org/issues/53234

o3 s390 workers

The s390 workers for openQA are running within podman containers on openqaworker1.
The containers are started using systemd but the unit files are specific to the containers and will end up in a restart-loop if this fact is ignored. Whenever the containers are recreated, the systemd files have to be recreated.

The containers are started like this (for i=101…104):

i=101
podman run -d -h openqaworker1_container --name openqaworker1_container_$i -p $(python3 -c"p=${i}*10+20003;print(f'{p}:{p}')") -e OPENQA_WORKER_INSTANCE=$i -v /opt/s390x_rebel_replacement:/etc/openqa -v /var/lib/openqa/share:/var/lib/openqa/share registry.opensuse.org/devel/openqa/containers15.2/openqa_worker:latest
(cd /etc/systemd/system/; podman generate systemd -f -n openqaworker1_container_$i --restart-policy always)
systemctl daemon-reload
systemctl enable container-openqaworker1_container_$i

As alternative s390x workers can run on the host "rebel" as well. Be aware that openQA workers accessing the same s390x instances must not run in parallel so only enable one worker instance per s390x instance at a time (See https://progress.opensuse.org/issues/97658 for details).

Monitoring

There is an internal munin instance on o3. Anyone wanting to look at the HTML pages, do this:

rsync -a o3:/srv/www/htdocs/munin ~/o3-munin/ 

(where "o3" is configured in your ssh config of course)

Mitigation of boot failure or disk issues

Worker stuck in recovery

Check disk health and consider manual fixup of mount points, e.g.:

test -e /dev/md/openqa || lsblk -n | grep -v nvme | grep "/$" && mdadm --create /dev/md/openqa --level=0 --force --raid-devices=$(ls /dev/nvme?n1 | wc -l) --run /dev/nvme?n1 || mdadm --create /dev/md/openqa --level=0 --force --raid-devices=1 --run /dev/nvme0n1p3

PPC specific configurations

In one case it was necessary to disable snapshots for petitboot with nvram -p default --update-config "petitboot,snapshots?=false" to prevent a race condition between dm_raid and btrfs trying to discover bootable devices (https://progress.opensuse.org/issues/68053#note-25). In another case https://bugzilla.opensuse.org/show_bug.cgi?id=1174166 caused the boot entries to be not properly discovered and it was necessary to prevent grub from trying to update the according sections (https://progress.opensuse.org/issues/68053#note-31).

Moving worker from osd to o3

  • Ensure system management, e.g. over IPMI works. This is untouched by the following steps and can be used during the process for recovery and setup
  • Ensure network is configured for DHCP
  • Instruct SUSE-IT to change VLAN for machine from 2 to 662 (example: https://infra.nue.suse.com/SelfService/Display.html?id=16458)
  • Remove from osd:
salt-key -y -d openqaworker7.suse.de
  • Add entry on o3 to /etc/dnsmasq.d/openqa.conf with MAC address, e.g.
dhcp-host=54:ab:3a:24:34:b8,openqaworker7
  • Add entry to /etc/hosts which dnsmasq picks up to give out a DHCP lease, e.g.
192.168.112.12   openqaworker7.openqanet.opensuse.org openqaworker7
  • Adapt NFS mount point
sed -i '/openqa\.suse\.de/d' /etc/fstab && echo 'openqa1-opensuse:/ /var/lib/openqa/share nfs4 ro,fsc 0 0' >> /etc/fstab
  • Reload dnsmasq with systemctl restart dnsmasq
  • Restart network on machine (over IMPI) using systemctl restart network and monitor in o3:journalctl -f -u dnsmasq until address is assigned, e.g.:
Feb 29 10:48:30 ariel dnsmasq[28105]: read /etc/hosts - 30 addresses
Feb 29 10:48:54 ariel dnsmasq-dhcp[28105]: DHCPREQUEST(eth1) 10.160.1.101 54:ab:3a:24:34:b8
Feb 29 10:48:54 ariel dnsmasq-dhcp[28105]: DHCPNAK(eth1) 10.160.1.101 54:ab:3a:24:34:b8 wrong network
Feb 29 10:49:10 ariel dnsmasq-dhcp[28105]: DHCPDISCOVER(eth1) 54:ab:3a:24:34:b8
Feb 29 10:49:10 ariel dnsmasq-dhcp[28105]: DHCPOFFER(eth1) 192.168.112.12 54:ab:3a:24:34:b8
Feb 29 10:49:10 ariel dnsmasq-dhcp[28105]: DHCPREQUEST(eth1) 192.168.112.12 54:ab:3a:24:34:b8
Feb 29 10:49:10 ariel dnsmasq-dhcp[28105]: DHCPACK(eth1) 192.168.112.12 54:ab:3a:24:34:b8 openqaworker7
  • Ensure all mountpoints up
mount -a
  • Change root password to o3 one
  • Allow ssh password authentication: sed -i 's/^PasswordAuthentication/#&/' /etc/ssh/sshd_config && systemctl restart sshd
  • Add personal ssh key to machine, e.g. openqaworker7:/root/.ssh/authorized_keys
  • Update /etc/openqa/client.conf with the same key as used on other workers for "openqa1-opensuse"
  • Update /etc/openqa/workers.ini with similar config as used on other workers, e.g. based on openqaworker4, example:
# diff -Naur /etc/openqa/workers.ini{.osd,}
--- /etc/openqa/workers.ini.osd 2020-02-29 15:21:47.737998821 +0100
+++ /etc/openqa/workers.ini     2020-02-29 15:22:53.334464958 +0100
@@ -1,17 +1,10 @@
-# This file is generated by salt - don't touch
-# Hosted on https://gitlab.suse.de/openqa/salt-pillars-openqa
-# numofworkers: 10
-
 [global]
-HOST=openqa.suse.de
-CACHEDIRECTORY=/var/lib/openqa/cache
-LOG_LEVEL=debug
-WORKER_CLASS=qemu_x86_64,qemu_x86_64_staging,tap,openqaworker7
-WORKER_HOSTNAME=10.160.1.101
-
-[1]
-WORKER_CLASS=qemu_x86_64,qemu_x86_64_staging,tap,qemu_x86_64_ibft,openqaworker7
+HOST=http://openqa1-opensuse
+WORKER_HOSTNAME=192.168.112.12
+CACHEDIRECTORY = /var/lib/openqa/cache
+CACHELIMIT = 50
+WORKER_CLASS = openqaworker7,qemu_x86_64

-[openqa.suse.de]
-TESTPOOLSERVER = rsync://openqa.suse.de/tests
+[http://openqa1-opensuse]
+TESTPOOLSERVER = rsync://openqa1-opensuse/tests
  • Remove OSD specifics
systemctl disable --now auto-update.timer salt-minion telegraf
for i in  NPI SUSE_CA telegraf-monitoring; do zypper rr $i; done
zypper -n dup --force-resolution --allow-vendor-change
  • If the machine is not a transactional-server one has the following options: Keep as is and handle like power8 (also not transactional), enable transactional updates w/o root being r/o, change to root being r/o on-the-fly, reinstall as transactional. At least option 2 is suggested, enable transactional updates:
zypper -n in transactional-update
systemctl enable --now transactional-update.timer rebootmgr
  • Enable apparmor
zypper -n in apparmor-utils
systemctl unmask apparmor
systemctl enable --now apparmor
  • Switch firewall from SuSEfirewall2 to firewalld
zypper -n in firewalld && zypper -n rm SuSEfirewall2
systemctl enable --now firewalld
firewall-cmd --zone=trusted --add-interface=br1
firewall-cmd --set-default-zone trusted
firewall-cmd --zone=trusted --add-masquerade
  • Copy over special openSUSE UEFI staging images, see #63382
  • Check operation with a single openQA worker instance:
systemctl enable --now openqa-worker.target openqa-worker@1
  • Test with an openQA job cloned from a production job, e.g. for openqaworker7
openqa-clone-job --within-instance https://openqa.opensuse.org/t${id} WORKER_CLASS=openqaworker7
  • After the latest openQA job could successfully finish enable more worker instances
systemctl unmask openqa-worker@{2..14} && systemctl enable --now openqa-worker@{2..14}
  • Monitor if nightly update works, e.g. look for journal entry:
Mar 01 00:08:26 openqaworker7 transactional-update[10933]: Calling zypper up
…
Mar 01 00:08:51 openqaworker7 transactional-update[10933]: transactional-update finished - informed rebootmgr
Mar 01 00:08:51 openqaworker7 systemd[1]: Started Update the system.
…
Mar 01 03:30:00 openqaworker7 rebootmgrd[40760]: rebootmgr: reboot triggered now!
…
Mar 01 03:36:32 openqaworker7 systemd[1]: Reached target openQA Worker.

Distribution upgrades

Note: Performing the upgrade differs slightly depending on the host setup:

  • On hosts with a writeable / you need to enter a root shell i.e. sudo bash
  • Transactional hosts require that you use transactional-update shell thereby creating a snapshot which is applied after a reboot, optionally using --continue if you want to make further changes to an existing snapshot

Consider using https://github.com/okurz/auto-upgrade/blob/master/auto-upgrade or manual:

new_version=15.3 # Specify the target release

# Change the release via the special $releasever
. /etc/os-release
sed -i -e "s/${VERSION_ID}/\$releasever/g" /etc/zypp/repos.d/*
zypper --releasever=$new_version ref
test -f /etc/openqa/openqa.ini && sudo -u geekotest /opt/openqa-scripts/dump-psql
zypper -n --releasever=$new_version dup --auto-agree-with-licenses --replacefiles --download-in-advance

# Check config files for relevant changes
rpmconfigcheck
for i in $(cat /var/adm/rpmconfigcheck) ; do vimdiff ${i%.rpm*} $i ; done
rm $(cat /var/adm/rpmconfigcheck)

reboot
systemctl --failed

Tip: Run this in screen -d -r || screen and use e.g. sudo bash

Remote management with IPMI

o3 and osd worker machines are controllable over IPMI from within the SUSE network, see openqa/workerconf.sls for the commands.
It is recommended to use shell aliases for convenience.

ipmitool can sometimes behave unreliably. It seems (to okurz) as if ipmitool version ipmitool-1.8.18+git20200916.1245aaa387dc from openSUSE Tumbleweed or Factory or the "systemsmanagement" OBS repo is more reliable than the version supplied with openSUSE Leap 15.2 (See #80544#note-14) and given a stable internet connection it is certainly possible to have a consistent serial console experience.

To ensure that remotely controlled machines power on automatically after a power loss ensure to set the power restory policy to "previous", especially for new machines. Using https://gitlab.suse.de/openqa/salt-pillars-openqa/#get-ipmi-definition-aliases :

IFS=$'\n'; for i in $(sed 's/^alias .*="\(.*\)"/\1/' ~/.openqa_ipmi_aliases); do eval "$i" chassis policy previous; done

Accessing imagetester

Imagetester can't output anything over SOL. Therefore it is neccessary to access it over the integrated iKVM console. Unfortunately java-webstart is somewhat broken and requires some extra steps to work:

  1. Access the webinterface of the BMC at http://10.160.65.195 and login
  2. Click on the preview image of the "Remote Console Preview" an download the according "launch.jnlp" webstart script
  3. Grab the required dependencies with curl and place them in a local directory:
mkdir /tmp/ikvm
curl -k https://10.160.65.195:443/liblinux_x86_64__V1.0.3.jar.pack.gz > /tmp/ikvm/liblinux_x86_64__V1.0.3.jar.pack.gz
curl -k https://10.160.65.195:443/iKVM__V1.69.13.0x0.jar.pack.gz > /tmp/ikvm/iKVM__V1.69.13.0x0.jar.pack.gz
  1. Open the previous downloaded "launch.jnlp" and replace the IP in the first line from <jnlp spec="1.0+" codebase="https://10.160.65.195:443/"> to <jnlp spec="1.0+" codebase="http://127.0.0.1:8080/">
  2. Launch some kind of webserver which can serve the previously downloaded dependencies for javaws (from /tmp/ikvm). In this example we use python: python3 -m http.server 8080
  3. Now you can finally launch the webstart application from your modifies "launch.jnlp" file in a second console: javaws -nosecurity -jnlp ~/Downloads/launch.jnlp
    • It will ask you how to run the application. You can run it in a sandbox and everything still works
  4. You should see the monitor output of imagetester now. "Virtual Storage" is also working which allows you to mount an ISO over this remote connection.

Also check https://progress.opensuse.org/issues/96719#note-27 where this was discovered. If you have questions or remarks you can ping @nicksinger

openQA infrastructure needs (o3 + osd)

TL;DR: new OSD ARM workers needed, missing redundancy for o3-ppc, rest is needing replacement as nearly all current hardware is out of vendor provided maintenance (as of 2021-05), SSD storage for o3 would be good

2020-03: SUSE IT (EngInfra) provided us more space for O3 but we have only slow rotating-disk storage. Performance could be improved by providing SSD storage.

The most time and effort we currently struggle with storage space for OSD (openqa.suse.de) both OSD (openqa.suse.de) as well as O3 (openqa.opensuse.org) (2020-03: Situation on o3 resolved with more storage provided by SUSE IT). Both instances (OSD + O3) are using precious netapp-storage but there is currently no better approach to use different, external storage. An increase of the available space would be appreciated, o3 being more important right now than osd, see https://progress.opensuse.org/issues/57494 for details. Graphs like
https://stats.openqa-monitor.qa.suse.de/d/nRDab3Jiz/openqa-jobs-test?orgId=1&from=1578343509900&to=1578653794173&fullscreen&panelId=12 show how usual test backlogs are worked on within OSD by architecture. It can be seen that both the ppc64le and aarch64 backlogs are reduced fast so we do not need more ppc64le or aarch64 machines. However, we have a stability problem with all three aarch64 workers. Potentially new machine(s) could help, see https://progress.opensuse.org/issues/41882 for details.

With number of workers and parallel processed tests as well as with the increased number of products tested on OSD and users using the system the workload on OSD constantly increases. CPU load alerts had been seen recently in #96713 and the higher load is visible in https://monitor.qa.suse.de/d/WebuiDb/webui-summary?viewPanel=25 . From time to time should increase the number of CPU cores on the OSD VM due to the higher usage.

Setup guide for new machines

Take machines out of salt-controlled production

E.g. for investigation or development or manual maintenance work

ssh osd "sudo salt-key -y -d $hostname"
ssh $hostname "sudo systemctl disable --now telegraf openqa-worker-auto-restart@\*"

Bring back machines into production

ssh osd "sudo salt-key -a $hostname && sudo salt --state-output=changes $hostname state.apply"

Depending on your actions further manual cleanup might be necessary, e.g. ssh $hostname "sudo systemctl unmask telegraf salt-minion"

Backup

Both openqa.opensuse.org and openqa.suse.de run on virtual machine clusters that provide redundancy and differential backup using snapshotting of the involved storage. SUSE-IT currently provides backups going back up to 3 days with two daily backups conducted at 23:10Z and 11:00Z. With this it is possible in cases of catastrophic data loss to recover (raise ticket over https://sd.suse.com in that case). Additionally automatic backup for the o3 webui host introduced with https://gitlab.suse.de/okurz/backup-server-salt/tree/master/rsnapshot covering so far /etc and the SQL database dumps. Fixed assets and testresults are backed up on storage.qa.suse.de (see https://gitlab.suse.de/openqa/salt-states-openqa/-/merge_requests/612)

Best practices for infrastructure work

  • Same as in OSD deployment we should look for failed grafana alerts if users report something suspicious
  • Collect all the information between "last good" and "first bad" and then also find the git diff in openqa/salt-states-openqa
  • Apply proper "scientific method" with written down hypotheses, experiments and conclusions in tickets, follow https://progress.opensuse.org/projects/openqav3/wiki#Further-decision-steps-working-on-test-issues
  • Keep salt states to describe what should not be there
  • Try out older btrfs snapshots in systems for crosschecking and boot with disabled salt. In the kernel cmdline append systemd.mask=salt-minion.service
  • Team should conduct a work backlog check on a daily base, e.g. look for urgent tickets related to infrastructure problems
  • For hardware replacement order replacement with the help of line managers, let the components be delivered to the according place, e.g. SUSE Nuremberg datacenter and create SUSE IT EngInfra ticket to have them conduct the physical component replacement
  • Test reboot stability of machines with commands like in https://progress.opensuse.org/issues/78010#note-31 e.g.
for run in {01..30}; do for host in $host; do echo -n "run: $run, $host: ping .. " && timeout -k 5 600 sh -c "until ping -c30 $host >/dev/null; do :; done" && echo -n "ok, ssh .. " && timeout -k 5 600 sh -c "until nc -z -w 1 $host 22; do :; done" && echo -n "ok, uptime/reboot: " && ssh $host "uptime && sudo reboot" && sleep 120 || break; done || break; done

Updated by okurz almost 3 years ago · 137 revisions