I noticed that often MM WUs get stuck in my systems. It looks like the wrapper loses sight of the worker application, which seems to finish normally, and then fails to report progress and status to the BOINC client, thus remains taking a slot until way past its due time (e.g., this WU).

I suspect that it happens when a MM WU is suspended, albeit kept in memory. It seems that the worker application keeps on running and the wrapper, suspended, misses the completion signal from it.

Killing the wrapper solves things, but it still happens with about 10% of the WUs.

Please, advise.
____________

Here's an update on something that I just observed: 2 MM WUs, one running and the other suspended in memory.

Observing the system using BOINCView and Process Explorer, suddenly, the other suspended WU vanished from Process Explorer (all, the wrapper, the worker and the watchdog), yet both BOINCView and the official BOINC Manager report that the other is still active.

I wonder if the watchdog killed the other WU even though it was suspended...

Here's the other's error output:

Now that the one WU finished, the other WU was restarted, as can be seen by its error output:

After restarting, the other WU finished successfully.

I'll continue keeping an eye out for hanged wrappers.
____________

Here's the error output of a stuck process:

It looks like the zombie processes are eventually purged.

____________

It looks like the zombie processes are eventually purged.

• boinc_lockfile
  
• stderr.txt
  

The latter is updated every few seconds with a new line: