https://kotlinlang.org logo
Join Slack
Powered by
Hi, 1. when does a Job get completed? once all th...
# coroutines
v
Hi, 1. when does a Job get completed? once all the `.launch`ed job children get completed? 2. can a job indefinitely block a parent context if it is still .isActive`
Untitled
e
When you call 
job.compelete()
(or 
job.cancel()
) and all its launched children complete, too.
v
my bad, slack posted this before I was able to post the other questions; 2. can and will a job block the parent from completion if it is still 
.isActive
3. I am trying to implement a simple worker pool, post all the `.launch`es complete, the job is still 
.isActive
, but (see 4) 4. the job is blocking the parent only if the job is spawned with 
parent=coroutineContext
rather than 
parent=Dispatchers.Default
5. I am not able to access the 
.complete()
function on the job interface
e
complete
is available on 
CompletableJob
that the 
Job(...)
constructor returns.
It is convenient and encouraged to wrap your 
Job
into a 
CoroutineScope
(use 
CoroutineScope(...)
constructor). This way you can do 
myScope.launch { ... }
which is more idiomatic.
v
ah I figured why I am not able to access `.complete`; I’m stuck on 
kotlinx-coroutines-core:1.1.1
e
Update to 
1.3.0
, then. Do you have any roadblocks?
v
versions are locked by a single parent across the org ¯\_(ツ)_/¯
e
The workaround is to use 
CompletableDeferred<Unit>()
as your job.
v
ok, so I changed the logic a bit; since I realized I do not need a specific (context + Job) 1. switched to 
: CoroutineScope by CoroutineScope(context)
constructor 2. stored the `.launch`’ed workers in a 
val workers: List<Job>
3. on my 
.shutdown()
I do a 
workers.joinAll()
which waits for completion and 4. 
.shutdownNow
does 
workers.map { it.cancelAndJoin() }
is there anything that does not seem very idiomatic for the same?
Untitled
maybe I do not need to wait for a 
.cancelAndJoin()
and just firing 
.cancel
would suffice for 
shutdownNow
e
2-4 are not needed. You can simply call 
cancel
on the scope/job (it cancels all the children) and do 
job.children.forEach { it.await() }
. No need to keep track of your children manually.
v
but since I am passing 
coroutineContext
in the constructor, will that try to cancel children of that context as well?
e
v
ah, so it’s a new job for the given CoroutineScope that gets independently cancelled?
e
There's no new job if you already had a job in the context when you called 
CoroutineScope(context)
Jobs form a hierarchy. Cancel parent and all children get cancelled automatically. You don't have to manually keep track of them. It won't harm, but there's simply no need to write all that code
v
I am a bit confused; There is 
class Foo(context) : CoroutineScope by CoroutineScope(context)
will have 
.[Job]
as whatever is the 
coroutineContext
passed to the ctor’s 
.[Job]
is calling 
Foo().cancel
not going to cancel the other running Job’s in parent?
e
It will cancel everything in parent. If you want a specific job to encompass coroutines created inside 
Foo
, write:
Copy code
class Foo(context) : CoroutineScope by CoroutineScope(context + Job(context[Job]))
v
yes! I have ended up with the following:
Copy code
suspend fun shutdown() {
        tasks.close()
        coroutineContext[Job]?.children?.forEach { it.join() }
        coroutineContext.cancel()
    }

    fun shutdownNow() {
        tasks.close()
        coroutineContext.cancel()
    }
and this cancellation does not mess with my parent spawn’ed jobs
cool, thanks for the inputs and patience 🙂
7 Views
Open in Slack
PreviousNext
Powered by