Fix crash on continue in missing package dialog #5937
Conversation
Due to the way how exceptions propagate from the Anaconda installation tasks we ended
up with a non-critical error exception interrupting the iteration of an important task
iteration loop installing the payload.
Due to not being handled "deep enough" but instead "bubbling up" to a top level
error handler the loop apparently gets interrupted & remaining tasks
skipped.
This resulted in an unrelated crash ("kernel version list no available")
due to all the important installation tasks being skipped, packages not
being installed and installation related data not being populated.
The end result was that a non-critical error (such as a missing package)
would trigger a dialog asking the user to quit or continue - but clicking
"continue" would result in a weird crash.
So move the error handler check closer to the task execution, to
prevent the loop from being interrupted. That way the loop will resume
its iteration if "continue" is clicked in the UI.
Also convert the non-critical error if it gets raised (user deciding
*not* to continue after the non-critical error) to a fatal one. This is
necessary, as otherwise the top level error handler would get triggered,
asking the user again to quit or continue.
NOTE: Longer term we really should clean this up & have all installation
tasks gathered, ordered and executed from a single place. Then all the
error handling could be in a single place, making things much simpler.
Resolves: INSTALLER-4045
Related: rhbz#2238045
Related: RHEL-57699
M4rtinK
temporarily deployed
to
gh-cockpituous
GitHub Actions
Inactive
|
This took quite a bit of time to track down what the heck is actually happening, why & what to do about it - ideally without having to change some of the tricky Anaconda innards :P Thankfully I seems to have arrive on a fairly minimal solution that seems to fix the issue as far without visibly breaking something else, as far as I can tell. This is how it looks like - first the user will see a continue yes/no dialog - same as before, nothing new here: Clicking "Yes" should no longer crash. Clicking "No" will now show this: While this is a side effect of making sure the user is not asked again to continue after pressing "No" (as the top level error handler is still active) it actually IMHO looks a lot better than the previous messy exception dialog that was shown after pressing the "No" button: Feedback welcome - if you have an idea to do this better or managed to hit some edge case this has introduced, let me know! :) |
|
NOTE, to easily reproduce the original issue, just create a kickstart like this: And run the installation with it (by for example injecting it to an updated boot.iso via the local development scripts). |
|
I'll open a RHEL 10 version of this PR once we are sure this is what we want to do to fix this issue (which is IMHO quite likely :) ). |
M4rtinK
changed the title
|
/kickstart-test --skip-testtypes whatever |
|
/build-image --boot.iso |
|
Images built based on commit 73b412f:
Download the images from the bottom of the job status page. |
The failed tests seem all like known flakes. When I run scripts/classify-failures (its latest version from rhinstaller/kickstart-tests#1315) on a folder containing only the failures I am getting: ie known issues, I guess I think we need to support the |
|
/kickstart-test --testtype smoke |
|
Thank you very much for fixing this. |
Due to the way how exceptions propagate from the Anaconda installation tasks we ended up with a non-critical error exception interrupting the iteration of an important task iteration loop installing the payload.
Due to not being handled "deep enough" but instead "bubbling up" to a top level error handler the loop apparently gets interrupted & remaining tasks skipped.
This resulted in an unrelated crash ("kernel version list no available") due to all the important installation tasks being skipped, packages not being installed and installation related data not being populated.
The end result was that a non-critical error (such as a missing package) would trigger a dialog asking the user to quit or continue - but clicking "continue" would result in a weird crash.
So move the error handler check closer to the task execution, to prevent the loop from being interrupted. That way the loop will resume its iteration if "continue" is clicked in the UI.
Also convert the non-critical error if it gets raised (user deciding not to continue after the non-critical error) to a fatal one. This is necessary, as otherwise the top level error handler would get triggered, asking the user again to quit or continue.
NOTE: Longer term we really should clean this up & have all installation tasks gathered, ordered and executed from a single place. Then all the error handling could be in a single place, making things much simpler.
Resolves: INSTALLER-4045
Related: rhbz#2238045
Related: RHEL-57699