ctrueden
commented
1 year ago A minimal example to reproduce:
public static void main(String[] args) throws Exception {
Context context = new Context();
// context.close();
System.out.println("Done");
}and it hangs, because the ExecutorService (created by Executors.newCachedThreadPool) spins out non-daemon threads by default, which sit around waiting for new jobs, blocking JVM shutdown, until the ExecutorService is shut down.
Unfortunately, just starting up the Context seems to result in calls to threadService.run, I believe via eventService.publishLater, which is why the simple example above hangs.
Changing to daemon threads is not sufficient though, because we do not want to allow JVM shutdown while these threads are still actively doing work. The Guava library handles this pitfall by adding a shutdown hook that waits for the daemon threads to finish pending work. I will try to implement that solution here in SciJava Common.
Daniel-Alievsky
I already created an issue in SCIFIO project: https://github.com/scifio/scifio/issues/486 But probably it should be here.
I'm using SCIFIO for reading and parsing large TIFF files with help of your TiffParser class.
I've notices that almost any usage of io.scif SCIFIO (current version 0.44) creates a background non-daemon thread, named like SciJava-6b53e23f-Thread-0@1742
This is a problem: this thread prevents an application to be normally closed.
This thread appears while any attempt to create SCIFIO context, both by
new Context(false)orSCIFIO scifio = new SCIFIO(); Context context = scifio.getContext();operators. The only variant when there is no problem is creating empty context bynew Context(true), but such a context is useless: it leads to NullPointerException when passed to TiffParser constructor.I've found a possible workaround: I can manually close the created context by
context.close()call, and the extra thread will be closed. However, it is not a good solution due to at least 2 reasons. 1) Contexts are created in invisible manner by a lot of SCIFIO methods, like IO.opener(), IO.save(some_image) etc. If at least one of such contexts will not be closed manually, parasite thread remains. 2) I'm using SCIFIO as a maven library A, which is used inside another library B, which is used inside another library C, etc. My end user does not know anything about the fact that a low-level library A uses SCIFIO, when the processed file is TIFF - it is an implementation detail. So, at the high level there is no suitable point where we could close SCIFIO contexts.Is it possible to resolve this problem? In particular, is it possible to make the service thread "SciJava-..." a daemon thread, not preventing normal usage of this library? Or, maybe, you could add an ability to create working context without creating any threads?
Note that this problem was not so critical in old SCIFIO version 0.11.0. In that version, the call
SCIFIO scifio = new SCIFIO()did not create an additional thread, and we could freely use scifio.getContext() But calls likeIO.openImgs(...)were problematic also in that version.