EOFError: marshal data too short -- causes?

[Glenn Linderman]

On 12/30/2015 12:43 PM, Glenn Linderman wrote:
> On 12/29/2015 1:00 PM, Terry Reedy wrote:
>> I updated to 2.7.11, 3.4.4, and 3.5.1 a couple of weeks ago, so the 
>> timestamps are all fresh.  So I don't know what happened with 3.4.3 
>> timestamps from last April and whether Windows itself touches the 
>> files.  I just tried importing a few and Python did not. 
>
> I'm a Windows user, too, generally, but the web host runs Linux.
>
> I suppose, since the install does the compileall, that I could set all 
> the __pycache__ files to read-only, even for "owner".  Like you said, 
> those files _can't_ be updated without Admin/root permission when it 
> is a root install... so there would be no need, once compileall has 
> been done, for the files to be updated until patches would be 
> applied.  This isn't a root install, though, but a "user" install.
>
> Level1 support at the web host claims they never touch user files 
> unless the user calls and asks them to help with something that 
> requires it. And maybe Level1 support religiously follows that policy, 
> but other files have changed, so that policy doesn't appear to be 
> universally applied for all personnel there... so the answer isn't 
> really responsive to the question, but the tech I talked to was as 
> much a parrot as a tech...
>
> Glenn

So this morning the problem happens again.  48 files or directories 
modified at 5:47am, while I'm sound asleep so the web site is down for 
over 3 hours until I wake up and notice (both because my bootup process 
always checks, and because I had several emails about it).

2 of the files had the suspicious 4096 EOF, and deleting the first 
caused it to be rebuilt and the second to be noticed, and deleting the 
second caused it to be rebuilt and cured the  site.

But all the touched files are .pyc files (and the directories 
__pycache__ directories).  None of the source files were modified. So 
why would any .pyc files ever be updated if the source files are not?  
Are there _any_  Python-specific reasons?

My only speculation is a problem accessing the .pyc file on first 
attempt, which would be a file system problem, not a Python problem?

Are there other speculative reasons?

And then why would a short file be built?  Conflict with multiple 
processes trying to rebuild it at the same time?  Or another file system 
problem?  Or???

This-could-be-annoying-if-it-keeps-happening-ly yours,
Glenn