Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Unhandled rejection overtakes synchronous exception through done() #471

Closed
rogierschouten opened this issue Jan 28, 2015 · 15 comments
Closed
Labels

Comments

@rogierschouten
Copy link

Seen with Bluebird 2.9.3.

The behavior below caused a programming error to become hidden, making debugging difficult. The done() method will terminate the application on a thrown error within a promise chain, but it might wait to do so for quite some time, even processing more promises. In the meantime, the application continues to run, generating all kinds of unexpected side effects - after all, we're already in an error state.

I'm aware that the whole thing is a-synchronous and that you cannot terminate the application immediately on an exception. However, it seems that arbitrarily long promise chains are still executed before the done() - they seem to overtake each other. I would at least expect some execution order to be preserved.

What happened in reality (below is a smaller example) is that an assertion error occurred within a promise chain that ended in .done(). After that, the application was in a state where it behaved wrongly and eventually caused another un-handled rejection. The un-handled rejection overtook the assertion error and only the rejection was shown, obscuring the real cause. Note that the rejection was quite far removed in terms of nodejs cycles.

The example below shows the problem

var Promise = require("bluebird");

// kill the process on unhandled rejections
Promise.onPossiblyUnhandledRejection(function(reason) {
        // do some custom logging
    console.log("possibly unhandled rejected Promise", reason);
    process.exit(1);
});

Promise.resolve()
    .then(function() {
        // longer chain expected to give error long after the process is dead
                // not because .done() but because of possibly unhandled rejection handler.
        Promise.resolve()
            .then(function() {              
                Promise.resolve()
                    .then(function() {              
                        Promise.reject(new Error("unhandled rejection"));
                    });
            });

        // exception expected to terminate the process
        throw new Error("unhandled thrown error which should end the process because of done()");
    })
    .done();

Console output:

possibly unhandled rejected Promise [Error: unhandled rejection]

Is this expected behavior?

@spion
Copy link
Collaborator

spion commented Jan 28, 2015

The problem is in async.js using setTimeout to throw the error, if available. By that time the possibly unhandled rejection has already happened and the process will exit. Maybe it can be fixed by a patch that rethrows both synchronously and asynchronously.

@rogierschouten
Copy link
Author

We also have this:

Promise.setScheduler(function(callback) {
    setImmediate.apply(null, [callback]);
});

I would at least have expected that to be used also for .done(). But I agree that throwing synchronously would be even better.

EDIT: note that reproducing the error was done without this scheduler so in that sense it does not matter.

@petkaantonov
Copy link
Owner

I want to throw synchronously but if the application recovers from the exception (e.g. always in browser and sometimes in node if there is domain or uncaughtException handler that doesn't exit process), it will leave bluebird in inconsistent state

@rogierschouten
Copy link
Author

Would that matter once you've made a programmer error?

@petkaantonov
Copy link
Owner

In practice, yes.. just open the console of any website, it's full of uncaught errors and the applications appear to be keep working just fine.

@rogierschouten
Copy link
Author

Could we make it an option, then? So we can have a sync exception server-side, or while not in production?

@petkaantonov
Copy link
Owner

What if in node it did the same as event emitters... write stack trace to stderr and exit immediately? shouldn't that be effectively the same as throwing synchronously but without the risk that the application recovers?

@rogierschouten
Copy link
Author

Yes that would be nice. But in that case I would expect a handler similar to onPossiblyUnhandledRejection() so that I can flush the logs.

@petkaantonov
Copy link
Owner

You mean process.on("exit"?

@rogierschouten
Copy link
Author

Good point

@rogierschouten
Copy link
Author

No wait, that means I cannot log the exception to my own logging system anymore and also I cannot flush the logs because process.on("exit" only allows sync actions.

@petkaantonov
Copy link
Owner

So if you need to do async stuff before the exception can be thrown, it will not be any different from the status quo because while you are waiting for the logs to flush, the unhandled rejection handler will be called and call process.exit

@petkaantonov
Copy link
Owner

@rogierschouten 3.0.0 will be soon released 2-4 weeks, do you have any input? I think you can synchronously flush logs with writeSync using stdout file descriptor.

Otherwise this will be left as (as it doesn't change anything if you need to flush logs asynchronously) and the next time it can be changed is 4.0.

@rogierschouten
Copy link
Author

Hi Petka,

I agree with you, you can throw synchronously.Thank you.

Rogier

fearenales added a commit to saitodisse/azk that referenced this issue May 31, 2015
fearenales added a commit to saitodisse/azk that referenced this issue Jun 1, 2015
fearenales added a commit to saitodisse/azk that referenced this issue Jun 7, 2015
@petkaantonov
Copy link
Owner

Fixed in 3.0 release

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
Projects
None yet
Development

No branches or pull requests

3 participants