Task visualiser: time values of D3 graph edges don't make sense?

[MichaelGrupp]

When using the D3 graph visualization, the time values at the edges don't seem to correspond to the actual duration of that task. The code indicates that it's a duration time.

Here's an example script. It runs a TestTask which wraps 4 DummyTasks that each create a temporary file and sleep for 1 second.

So, running the whole pipeline should take a bit more than 4 seconds in total.

import time

import luigi


class DummyTask(luigi.Task):
    tmp_path = luigi.Parameter()

    def output(self):
        return luigi.LocalTarget(path=self.tmp_path, is_tmp=True)

    def run(self):
        open(self.tmp_path, 'w').write("")
        time.sleep(1)


class TestTask(luigi.WrapperTask):
    def requires(self):
        for path in ["/tmp/a", "/tmp/b", "/tmp/c", "/tmp/d"]:
            yield DummyTask(tmp_path=path)


if __name__ == "__main__":
    luigi.build([TestTask()])

When I run a luigid daemon and then start the script:

time python test.py

time reports a bit more than 4 seconds as expected.

However, the D3 graph in the web viewer shows this:

To add to the confusion, not even the luigid daemon was running as long as any of the shown times in this example.

Could this be a time conversion bug or am I understanding something wrong?

[MichaelGrupp]

I'm using luigi 2.8.2 from PYPI on Ubuntu 16.04

[dlstadther]

@MichaelGrupp Could you @-blame the author of that section of code? I'm not familiar with the visualization integrations.

[MichaelGrupp]

@hadesbox can you help here?

The oldest occurence of this duration I could track down across file changes is here: 755064b#diff-c32982ed29d088f880fba9a28a99c1a7R294

[MichaelGrupp]

I think I fixed it. Will open a PR.

[Tarrasch]

Would be nice if the strings eventually get human formatted. But it's a good start to not have incorrect data. :)