914120 - [Chrome Workers] Chrome workers shouldn't be killed abruptly during shutdown

Status: RESOLVED WONTFIX

(Core :: DOM: Workers, defect)

Description

[David Teller [:Yoric] - still alive but not very active]

During shutdown, we kill Workers, regardless of what they may be doing. For DOM Workers, this is a desired behavior. For Chrome Workers, however, not so much, as Chrome Workers may very well be in the process of doing something important, such as a system call.

At the moment, to save our workers from being killed, we have to spin the event loop from JavaScript during shutdown until we are satisfied that they are done with any important work. That's both ugly and error-prone.

I believe that we should change the behavior to ensure that shutdown does not kill a Chrome Worker while it is handling a message. Rather, shutdown should dispatch a call |shutdown()| on the worker and the worker runtime should wait until that call has been properly handled before proceeding with xpcom thread shutdown.

Comment 1

[David Teller [:Yoric] - still alive but not very active]

Note: Current behavior of Chrome Workers could cause data loss with Session Restore. Our fault for misunderstanding the semantics of workers, but this is an example of unintended consequences of this design choice in Chrome Workers.

Comment 2

[Ben Turner (not reading bugmail, use the needinfo flag!)]

Just keep in mind that all I/O needs to be done at 'profile-before-change' (before workers shut down).

Comment 3

[David Teller [:Yoric] - still alive but not very active]

Actually, it doesn't seem to be so clear - we have profile-before-change and profile-before-change2.

Regardless, are you suggesting that we spin the event loop at profile-before-change (or profile-before-change2) to ensure that off main thread I/O always finishes before we kill chrome workers?

Comment 4

[Ben Turner (not reading bugmail, use the needinfo flag!)]

All I/O must complete during 'profile-before-change', spinning the event loop if needed. That is a totally separate problem from whether or not ChromeWorkers get canceled at shutdown.

Comment 5

[David Teller [:Yoric] - still alive but not very active]

Good point. I believe that this bug is still valid, but 1/ with a lower priority than I thought; 2/ not for the I/O reasons I mentioned initially.

Comment 6

[David Teller [:Yoric] - still alive but not very active]

Let's forget about that bug. We'll reopen it if we realize that it is actually useful.