Skip to content

cli: make ^C print a JS stack trace #24937

Description

@bnoordhuis

It would be exceedingly nice for debugging non-responsive scripts if terminating the process with ^C / SIGINT prints a JS stack trace leading up to the currently executing code.

The use case is scripts stuck in (for example) infinite loops. Right now you don't really get an indication when that's happening except CPU usage in top.

It might be hard to pull off because signal handlers but maybe someone has a clever idea. My attempt at being clever:

  1. Route signals to a watchdog thread instead of the main thread
  2. Call isolate->RequestInterrupt() when a signal is received.
  3. Collect the stack trace in the interrupt handler.
  4. Either re-raise the signal on the main thread to terminate as usual, or raise a termination exception to kill the script.

Activity

  1. joyeecheung commented on Dec 10, 2018

    @joyeecheung
    Member

    Would it make more sense if we only do this when inspector is enabled? Or instead of SIGINT, use another signal.

  2. targos commented on Dec 10, 2018

    @targos
    Member

    If inspector is enabled, we don't need this. You can use the inspector API to pause the script and get the stack trace.

  3. added
    feature requestIssues requesting new Node.js features.
    cliIssues and PRs related to the Node.js command-line interface.
    on Dec 10, 2018
  4. joyeecheung commented on Dec 10, 2018

    @joyeecheung
    Member

    If inspector is enabled, we don't need this. You can use the inspector API to pause the script and get the stack trace.

    Makes sense.

    1. Route signals to a watchdog thread instead of the main thread
    2. Call isolate->RequestInterrupt() when a signal is received.
    3. Collect the stack trace in the interrupt handler.
    4. Either re-raise the signal on the main thread to terminate as usual, or raise a termination exception to kill the script.

    I remember there was a similar discussion when we were talking about implementing a special option for util.inspect to break when it's taking too long? The opinions were we should enable that as a global option instead of making it specific to util.inspect. Would a new CLI option (instead of --inspect) makes more sense for this?
    Also either way we should already be able to do the watchdog refactoring even if we do not implement 3?

  5. bnoordhuis commented on Dec 10, 2018

    @bnoordhuis
    MemberAuthor

    If inspector is enabled, we don't need this. You can use the inspector API to pause the script and get the stack trace.

    I'm aware. This is specifically for the case where the inspector is not enabled.

    I got the idea while working on some python code that occasionally got stuck. Tapping ^C and getting a useful stack trace is just amazingly useful.

    Also either way we should already be able to do the watchdog refactoring even if we do not implement 3?

    Yes, there are more reasons why a watchdog would be useful.

  6. targos commented on Dec 10, 2018

    @targos
    Member

    I'm aware. This is specifically for the case where the inspector is not enabled.

    I was replying to Joyee. I would really like to see a stack trace with ^C.

  7. bnoordhuis commented on Jan 29, 2020

    @bnoordhuis
    MemberAuthor

    Fixed now that #29207 has (finally!) landed. Congrats, @legendecas!

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    cliIssues and PRs related to the Node.js command-line interface.feature requestIssues requesting new Node.js features.

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions