Call stopGracefully() when batch ingestion is killed explicitly

[ankit0811]

Fixes #7962.

Fixed by calling task.stopGracefully() from ForkingTaskRunner when hadoop ingestion task is killed, as the control goes to the ForkingTaskRunner::shutdown
and not to SingleTaskBackgroundRunner::stop

---

This PR has:

• been self-reviewed.

• been tested in a test Druid cluster.

---

[jihoonson]

Hmm, @ankit0811 do you know why stopGracefully didn't get called by ForkingTaskRunner? It's supposed to be called when the hadoop index task is killed. And how did you kill the hadoop index task?

[ankit0811]

The ingestion task was issued a kill command by clicking the kill link on the UI
This did not trigger calling the stopGracefully method

[jihoonson]

The ingestion task was issued a kill command by clicking the kill link on the UI

Do you mean the Druid task issued a kill command to kill Hadoop job? or you issued a kill command to kill the Druid task? If it's the later case, then it sounds like something went wrong with the kill command issued to the Druid task. How did you know stopGracefully wasn't called? I guess Tried killing job: [%s], status: [%s] wasn't printed?

[ankit0811]

Yes, it is the later case. And yes the log line is not printed

[clintropolis]

I looked into this a bit over the weekend, and have what I think is an alternative fix to this (that also fixes a logging issue). It seemed easier to just open a separate PR to capture all of the logs and details from the change I made there than to hijack this PR.

#8085 fixes a logging issue so we can be sure that lifecycle stop is happening, but I don't believe it fixes the issue with the Hadoop task not getting killed, see here for more details.