Submitter guide

How to submit build and test reports with KCIDB

Here’s what you need to do to submit your reports.

  1. Get submission credentials and parameters.
  2. Install KCIDB.
  3. Generate some report data.
  4. Submit report data.
  5. Go to 3, if you have more.

You don’t need to run a daemon, just execute command-line tools (or use the Python 3 library, if you’re feeling fancy).

1. Get submission credentials and parameters

Write to, introduce yourself, and explain what you want to submit (better, show preliminary report data). Once your request is approved, you will get a JSON credentials file, which you can use to authenticate yourself with KCIDB tools/library.

Export the file location (here ~/.kcidb-credentials.json) into environment like this:

export GOOGLE_APPLICATION_CREDENTIALS=~/.kcidb-credentials.json

We will also need to agree on the “origin” string identifying your system among other submitters. We’ll use submitter in examples below.

Finally you will need to specify some or all of the following to the tools/library:

  • Google Cloud project: kernelci-production
  • Submission queue topic: playground_kcidb_new

The above refers to the special “playground” setup we have, where you can freely experiment with your submissions, without worrying about any negative effects on the system or other submitters. This setup has a separate database, which can be selected on the dashboard. We’ll use playground parameters in the examples below.

Once you feel comfortable and ready, we’ll add extra permissions for your account, and you can start using the production parameters:

  • Google Cloud project: kernelci-production
  • Submission queue topic: kcidb_new

The submitted data will appear in our production database, visible on the dashboard by default.

2. Install KCIDB

KCIDB employs continuous integration and delivery, and aims to keep the code working at all times.

Please install the latest version from GitHub:

pip3 install --user 'git+'

Then make sure your PATH includes the ~/.local/bin directory, e.g. with:

export PATH="$PATH":~/.local/bin

See Installation for alternatives, and if you know your Python, feel free to do it your way!

To test your installation, authentication, and the parameters you received, submit an empty report:

$ echo '{"version":{"major":4,"minor":3}}' |
        kcidb-submit -p kernelci-production -t playground_kcidb_new

The command should execute without errors, produce the submitted message ID on output, and finish with zero exit status.

3. Generate some report data

kcidb-schema tool will output the current schema version.

However, all tools will accept data complying with older schema versions. Pipe your data into kcidb-validate tool to check if it will be accepted.

Here’s a minimal report, containing no data:

    "version": {
        "major": 4,
        "minor": 3

You can submit such a report, it will be accepted, but will have no effect on the database or notifications.


The schema describes five types of objects which can be submitted independently or in any combination:

  • “checkout” - a checkout of the code being built and tested
  • “build” - a build of a specific checkout
  • “test” - an execution of a test on a specific build in specific environment
  • “issue” - an issue found either in the kernel code, a test, or a CI system
  • “incident” - a record of an issue appearing in a build or a test run

Each of these object types refers to on or two of the others IDs. The only required fields for each object are their own IDs, IDs of the parent objects (except for checkouts and issues), and the origin. Objects of each type are stored in their own top-level array named respectively (in plural).

Here’s an example of a report, containing only the required fields for a checkout with one build and one test, as well as one issue and one incident:

    "checkouts": [
            "id": "submitter:32254",
            "origin": "submitter"
    "builds": [
            "id": "submitter:32254",
            "checkout_id": "submitter:c9c9735c46f589b9877b7fc00c89ef1b61a31e18",
            "origin": "submitter"
    "tests": [
            "id": "submitter:114353810",
            "build_id": "submitter:956769",
            "origin": "submitter"
    "issue": [
            "id": "submitter:124853810",
            "version": 1,
            "origin": "submitter"
    "incident": [
            "id": "submitter:1084645810",
            "issue_id": "submitter:956769",
            "origin": "submitter",
            "issue_version": 0
    "version": {
        "major": 4,
        "minor": 3

Object IDs

All object IDs have to start with your “origin” string, followed by the colon : character, followed by your origin-local ID. The origin-local ID can be any string, but must identify the object uniquely among all objects of the same type you submit. E.g.:



Once you get the required properties (IDs and origins) generated, and have your objects accepted by KCIDB, you can start adding the optional fields. Some good starting candidates are described below. See the schema for more.



True if the checkout is valid, i.e. if the source code was successfully checked out. False if not, e.g. if its patches failed to apply. Set to True if you successfully checked out a git commit.

Example: true


The URL of the Git repository which contains the checked out base code. The shortest possible https:// URL, or, if that’s not available, the shortest possible git:// URL.

Example: ""


The full commit hash of the checked out base code. Note that until a checkout has the git_commit_hash property it may not appear in reports or on the dashboard.

Example: "db14560cba31b9fdf8454d097e5cb9e488c621fd"


The full hash of the patches applied on top of the commit, or an empty string, if there were no patches. Note that until a checkout has the patchset_hash property it may not appear in reports or on the dashboard.

The hash is a sha256 hash over newline-terminated sha256 hashes of each patch, in order of application. If your patch file alphabetic order matches the application order (which is true for patches generated with git format-patch or git send-email), and you only have the patchset you’re hashing in the current directory, you can generate the hash with this command:

sha256sum *.patch | cut -c-64 | sha256sum | cut -c-64

Example: "a86ef57bf15cd35ba4da4e719e0874c8dd9432bb05d9fb5e45b716d43561d2b8"


The time the checkout was started by the CI system. As described by RFC3339, 5.6 Internet Date/Time Format.

Example: "2020-08-14T23:08:06.967000+00:00"



True if the build is valid, i.e. if it succeeded. False if not.

Example: true


Target architecture of the build. Not standardized yet.

Example: "x86_64"


Name and version of the compiler used to make the build.

Example: "gcc (GCC) 10.1.1 20200507 (Red Hat 10.1.1-1)"


The time the build was started, according to RFC3339, 5.6 Internet Date/Time Format.

Example: "2020-08-14T23:08:10.008000+00:00"



The test status string, one of the following:

  • “FAIL” - the test completed and reported the tested code as faulty.
  • “ERROR” - the test didn’t complete due to a failure in its code, and the status of the tested code is unknown.
  • “MISS” - the test didn’t run due to a failure in the test harness, and the status of both the test and the tested code is unknown.
  • “PASS” - the test completed and reported the tested code as correct.
  • “DONE” - the test completed and had not reported the status of the tested code, but, for example, produced a performance measurement result.
  • “SKIP” - the test did not run or complete, because it was not applicable, and the status of both the test and the tested code is unknown.

The status names above are listed in priority order (highest to lowest), which can be used to produce a summary status for a collection of test runs.

For example, the summary status for all testing done on a build would be the highest-priority status across all its tests.

You can break down the testing stack into three layers: the tested code, the test, and the harness running the test (and everything above it). With that in mind, you can then express each status as one of three outcomes for each layer: failure, success, or no data. Like this:


FAIL        ❌   ✅   ✅                 ❌ - failure
ERROR       ➖   ❌   ✅                 ✅ - success
MISS        ➖   ➖   ❌                 ➖ - no data
PASS        ✅   ✅   ✅
DONE        ➖   ✅   ✅
SKIP        ➖   ➖   ✅
            ➖   ➖   ➖

E.g. an ERROR status would mean that the harness succeeded in running the test, but the test code itself failed, and the test didn’t complete, so we cannot make any conclusions about the tested code.

OTOH, the similar DONE status would mean that both the harness and the test worked alright, but the test simply didn’t produce sufficient data to make a conclusion about the code. E.g. it made a performance measurement that needs analyzing over a span of revisions (to determine if it’s dropping or rising), and is meaningless alone.

Going one layer up, the MISS status means that the harness has failed to even execute the test, so it didn’t run, and we don’t have any data on its outcome, and obviously no data on the tested code either. And the similar SKIP status means that the harness worked alright, and either it, or the test itself, decided not to run this specific test, because it was inapplicable.

Finally, the last line in the table above (without a status) corresponds to the absent “status” field, and means that the test is either only scheduled, or is still executing, and so we have no status data yet.

Another way to visualize the status values and the three layered testing stack, is to put them into an execution outcome chart:

    columns 5

    B1("SKIP\ntest not applicable")
    C1("DONE\nno test result required")
    D1("PASS\ncode fulfilled requirements")


    A["schedule test"]
    B["run test"]
    C["evaluate result"]


    B2("MISS\ntest could not be run\n") space
    C2("ERROR\ntest could not be completed") space
    D2("FAIL\ncode failed requirements")

    B3["responsibility: CI maintainer"]
    C3["responsibility: test maintainer"]
    D3["responsibility: kernel developer"]

    A -.-> B
    A --> B1
    A --> B2

    B -.-> C
    B --> C1
    B --> C2

    C --> D1
    C --> D2

    style A  fill:#F5F5F5,stroke:#666666,stroke-width:1px,color:#000
    style B  fill:#F5F5F5,stroke:#666666,stroke-width:1px,color:#000
    style C  fill:#F5F5F5,stroke:#666666,stroke-width:1px,color:#000
    style B1 fill:#D9EAD3,stroke:#82B366,stroke-width:1px,color:#000
    style C1 fill:#D9EAD3,stroke:#82B366,stroke-width:1px,color:#000
    style D1 fill:#D9EAD3,stroke:#82B366,stroke-width:1px,color:#000
    style B2 fill:#FFF2CC,stroke:#D6B656,stroke-width:1px,color:#000
    style C2 fill:#FFE6CC,stroke:#D79B00,stroke-width:1px,color:#000
    style D2 fill:#f4cccc,stroke:#B85450,stroke-width:1px,color:#000

    style B3 fill:transparent,stroke:transparent,color:#6C8EBF
    style C3 fill:transparent,stroke:transparent,color:#6C8EBF
    style D3 fill:transparent,stroke:transparent,color:#6C8EBF

Example: "FAIL"


True if the test status should be ignored.

Could be used for reporting test results without affecting the overall test status and alerting the subscribers. For example, for collecting test reliability statistics when the test is first introduced, or is being fixed.

Example: false


Dot-separated path to the node in the test classification tree the executed test belongs to. The empty string signifies the root of the tree, i.e. all tests for the build, executed by the origin CI system.

Please consult the catalog of recognized tests for picking the top-level name for your test, and submit a PR adding it if you don’t find it.

Example: "ltp.sem01"


The time the test run was started, according to RFC3339, 5.6 Internet Date/Time Format.

Example: "2020-08-14T23:41:54+00:00"


“Issues” describe an issue with either the kernel code being tested, the test, or anything running the test, such as test harness, framework, or just the CI system as a whole.


The version number of the issue (required). The system always uses the issue with the largest version number, so if you want to change your issue, submit a new one with the same ID and larger version.

Example: 20240502101105


The URL pointing to the issue report: e.g. an issue in a bug tracker, a thread on a mailing list, and so on. Anything helping describe and identify the issue to humans.



The subject, or title of the issue report, helping identify the issue in reports or dashboards, without following the report URL.

Example: C-media USB audio device stops working from 5.2.0-rc3 onwards


An object with boolean attributes pointing out the origin, or the “culprit” of the issue: code - if the bug is in the kernel itself, tool - if the bug is in the test, or e.g. the build toolchain, and harness - if the bug is in the test framework, or the CI system in general. These fields help the system decide who to notify when the issue is discovered somewhere.

A missing attribute would indicate the unknown status (not “false”), so please include each attribute with a value, when you know it.

Example: `{“code”: true, “tool”: false, “harness”: false}'


“Incidents” record issue occurrences in builds and tests. They always refer to a particular version of the issue. This allows the system to use results of triaging the previous version of the issue, while triaging of the new one is ongoing.


The version of the issue this incident refers to (required, in addition to issue_id).

Example: 20240502101105


ID of the build the issue was found in, for issues found during a build.

Example: submitter:32254


ID of the test run the issue was found in, for issues found during a test.

Example: submitter:114353810


True if the issue did occur in the linked build or test. False if it did not. An absent attribute means that the occurence status is unknown, and can signify ongoing triaging.

Example: true

Extra data

If you have some data you’d like to provide developers with, but the schema doesn’t accommodate it, put it as arbitrary JSON under the misc field, which is supported for every object. Then contact KCIDB developers with your requirements.

For example, here’s a misc field from KernelCI-native builds:

    "build_platform" : [
        "#102~16.04.1-Ubuntu SMP Tue Jul 14 20:28:23 UTC 2020",
    "kernel_image_size" : 9042816,
    "vmlinux_bss_size" : 1019904,
    "vmlinux_data_size" : 1595008

The kernel developers would already be able to see it in the dashboard, and the KCIDB developers would have samples of your data which would help them support it in the schema.

You can also put any debugging information you need into the misc fields. E.g. you can add IDs of your internal objects corresponding to reported ones, so you can track where they came from, like CKI does:

    "beaker_recipe_id": 8687594,
    "beaker_task_id": 114353810,
    "job_id": 956777,
    "pipeline_id": 612127

While the misc field is primarily aimed at computers, the comment field for every object can contain free-form text helping people understand the data.

4. Submit report data

As soon as you have your report data pass validation (e.g. with the kcidb-validate tool), you should be able to submit it to the database.

If you’re using shell, and e.g. have your data in file report.json, pipe it to the kcidb-submit tool like this:

kcidb-submit -p kernelci-production \
             -t playground_kcidb_new < report.json

If you’re using Python 3, and e.g. have variable report holding standard JSON representation of your report, you can submit it like this:

import kcidb

client = kcidb.Client(project_id="kernelci-production",

Your data could take up to a few minutes to reach the database, but after that you should be able to find it in our dashboard.

Submitting directly

If for any reason you cannot use the command-line tools, and you don’t use Python 3 (e.g. you are using another language in a “serverless” environment), you can interface with KCIDB submission system directly.

NOTE: this interface is less stable than the command-line, and the library interfaces, and is more likely to change in the future.

You will have to use one of the Google Cloud Pub/Sub client libraries or service APIs to publish your reports to the Pub/Sub topic specified above, using the provided credentials. Please make sure to validate each report against the schema output by kcidb-schema before publishing it.

Submitting objects multiple times

If you submit an object with the same ID more than once, then the database will still consider them as one object, but will pick the value for each of its properties randomly, from across all submitted objects, wherever present.

This can be used to submit object properties gradually. E.g. you can send a test object without the duration and status properties when you start the test. Then, when it finishes, you can send a report with a test containing the same id, and only the duration and status properties, to mark its completion.

5. Go to 3, if you have more

Do not hesitate to start submitting your reports as soon as you can. This will let you, kernel developers, and KCIDB developers see how it works, what changes/additions might be needed to both your data and KCIDB, and improve our reporting faster!

Last modified November 18, 2021