Often hangs when called from gnome-software

[iainlane]

[ Sorry, I don't have a backtrace to hand ]

gnome-software recently introduced support for cancelling long-running plugin operations. That is, when it calls a vfunc on a plugin, if that takes more than x seconds, the request is cancelled. This was added mainly because the snap plugin was behaving badly and hanging, meaning that entire operations were not returning.

Now that's fixed (and this work is soon to be uploaded to Artful), I thought I'd file this bug to track investigations as to /why/ the operations are hanging in the first place.

[iainlane]

I think this article about using GMainContext from libraries is relevant. I've implemented some of the recommendations contained in it, and got back to a state where the testsuite is passing. Haven't yet tried it in GNOME Software, but will do tomorrow and then I'll let this issue know what my findings are.

[iainlane]

I had this on my desktop!

Thread 7 (Thread 0x7fffcd84e700 (LWP 8978)):
#0  syscall () at ../sysdeps/unix/sysv/linux/x86_64/syscall.S:38
#1  0x00007ffff748b75f in g_cond_wait (cond=cond@entry=0x55555587e018, mutex=mutex@entry=0x55555587e010) at ../../../../glib/gthread-posix.c:1395
#2  0x00007ffff7444a30 in g_main_context_wait (context=0x55555587e010, cond=0x55555587e018, mutex=0x55555587e010) at ../../../../glib/gmain.c:3331
#3  0x00007ffff7445ab3 in g_main_loop_run (loop=0x7fffc40525e0) at ../../../../glib/gmain.c:4054
#4  0x00007fffd79d858c in snapd_client_list_one_sync (client=client@entry=0x7fffd80105c0, name=name@entry=0x7fffc405abe0 "anbox-installer", cancellable=cancellable@entry=0x7fffd800c870, error=error@entry=0x0)
    at snapd-client.c:2809
#5  0x00007fffd7bf66c8 in gs_plugin_refine_app (plugin=0x555555a7be80, app=0x7fffc405c000, flags=2097184, cancellable=0x7fffd800c870, error=0x7fffcd84d938) at ../plugins/snap/gs-plugin-snap.c:428
#6  0x00005555555c36f9 in gs_plugin_loader_call_vfunc (helper=helper@entry=0x7fffb807ae00, plugin=plugin@entry=0x555555a7be80, app=app@entry=0x7fffc405c000, list=0x5555586a1740, list@entry=0x0, 
    cancellable=cancellable@entry=0x7fffd800c870, error=error@entry=0x7fffcd84dae8) at ../lib/gs-plugin-loader.c:577
#7  0x00005555555c40e9 in gs_plugin_loader_run_refine_internal (helper=helper@entry=0x7fffb807ae00, list=list@entry=0x5555586a1740, cancellable=cancellable@entry=0x7fffd800c870, 
    error=error@entry=0x7fffcd84dae8) at ../lib/gs-plugin-loader.c:764
#8  0x00005555555c46c0 in gs_plugin_loader_run_refine (helper=helper@entry=0x5555586a1780, list=list@entry=0x5555586a1740, cancellable=cancellable@entry=0x7fffd800c870, error=error@entry=0x7fffcd84dae8)
    at ../lib/gs-plugin-loader.c:898
#9  0x00005555555c59a5 in gs_plugin_loader_process_thread_cb (task=0x7fffc0005ae0, object=<optimised out>, task_data=0x5555586a1780, cancellable=0x7fffd800c870) at ../lib/gs-plugin-loader.c:3105
#10 0x00007ffff70eab2d in g_task_thread_pool_thread (thread_data=0x7fffc0005ae0, pool_data=<optimised out>) at ../../../../gio/gtask.c:1328
#11 0x00007ffff746e07e in g_thread_pool_thread_proxy (data=<optimised out>) at ../../../../glib/gthreadpool.c:307
#12 0x00007ffff746d685 in g_thread_proxy (data=0x55555852c2d0) at ../../../../glib/gthread.c:784
#13 0x00007ffff499b6da in start_thread (arg=0x7fffcd84e700) at pthread_create.c:456
#14 0x00007ffff46d517f in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:105

[robert-ancell]

Is that backtrace from after you patched it or before?

Yes, the GMainContext stuff is a lot harder than I initially thought. I was discussing it with Allison but that only got half way...

[iainlane]

Before - that's how it hangs. I'll try the patch in the morning.

…

On 24 July 2017 22:16:29 BST, Robert Ancell ***@***.***> wrote: Is that backtrace from after you patched it or before? Yes, the GMainContext stuff is a lot harder than I initially thought. I was discussing it with Allison but that only got half way...

[robert-ancell]

I'm going to close this assumed fixed with the recent threading changes. Please re-open if not...