SonarScanner into multiple jobs declaration in CI

Suggest new features
, ,
1

Hi,

I would like to better understand under the hood of the dotnet sonarscanner tool

I know there are many remarks about running the build and tests tasks inside the same job on a CI and that’s the way it works.

For what I understand, the sonarscanner begin and end will get all the report output and send it to the server for analysis.

So I was asking myself, would it possible to send part of data per job in a CI ?

What I am saying is, for a gitlab CI example, we could have something like :

init:
  stage: .pre
  script:
    - dotnet sonarscanner begin /step:true /id:$CI_COMMIT_SHORT_SHA

build:
  stage: build
  script:
    - dotnet sonarscanner beginstep /id:$CI_COMMIT_SHORT_SHA
    - dotnet build
    - dotnet sonarscanner endstep /id:$CI_COMMIT_SHORT_SHA

test:
  stage: test
  script:
    - dotnet sonarscanner beginstep /id:$CI_COMMIT_SHORT_SHA
    - dotnet test
    - dotnet sonarscanner endstep /id:$CI_COMMIT_SHORT_SHA

cleanup:
  stage: .post
  script:
    - dotnet sonarscanner end /id:$CI_COMMIT_SHORT_SHA

In this case, we could keep separated job for each task and avoid an unique sonar task that do everything (and take time)

I am not saying this is a solution, and surely there are a lot of scenario to resolve and it would be a long-term implementation.

I just want to know if hypothetically this could be possible

Thanks.

PS: I was sure this feature was already asked, but didn’t find it

2

Hi,

I’m not sure I understand what you’re asking. I guess you want to isolate build from test, each with its own begin and end? If that’s what you’re after, I wouldn’t recommend it. begin essentially sets up a listening context, and causes the rules to be run as part of / during the build. The end step wraps up analysis and submits the analysis report to the server.

Run two different begin / end combinations on the same project and you’ll get two different half-analyses that continually overwrite each other because analysis is not additive. It all has to be part of the same analysis.

 
HTH,
Ann

3

Hi,

Thanks for your answer.

It was not thinking about pure isolation, sorry for my english. For me, the scanner begin and the scanner end actions are at the pipeline level

A block like beginstep and endstep are temporary listening/rules/analysis/… running during a job.

When all the jobs are done, the server would be able to “concat” the different reports and do a final one

For my better personal understanding, how does the scanner listen those 2 tasks below ?

test:
  stage: test
  script:
    - dotnet sonarscanner begin
    - dotnet build
    - dotnet test
    - dotnet sonarscanner end

Does it get all the output in one step ? The scanner does not see any difference between one line, two or three lines ?

Or does it parse the first command output dotnet build, then do internal things, and then parse the dotnet test output, finalize, and send all to the server ?

4

Hi,

Again, the concatenation happens during the scanner run, not server-side. If you send multiple reports to the server, it won’t concatenate them; it will replace them.

You’re asking for the low-level, technical details? There’s a hint the the docs I linked. Did you check them?

To be honest, I’m not sure what changes are made to the test step. I suspect it’s just an order-of-operations requirement that puts the test step where it is in the sequence.

 
HTH,
Ann

5

If you send multiple reports to the server, it won’t concatenate them; it will replace them.

That was the initial question. In theory could it be possible to concatenate instead of replace ?

That was all the idea behind the concept of beginstep and endstep. In theory, set a series of functions server-side that add data and not replace them. Then tell the server “ok no more data”, then the server digest everything and “voilà”! (again, just in theory, I can imagine the amount of work, of questions, problems, …)

Through the link you provided I read about the end command:

It […], collects the analysis data generated by the build, the test results, and the code coverage, and then uploads everything to SonarQube Server.

So I assume the scanner does separate task for each dotnetcommand ?

Sorry, this is going off-topic of a simple “suggest feature” :sweat_smile:

6

Hi,

No.

Okay, in theory if we totally re-engineered everything about how analysis works, then sure. :sweat_smile:

Again, all the concatenation is done CI-side.

 
Ann

7

Ha, I thought parts were more isolate

But that was really instructive.

Thank you :+1: