[Debian-science-sagemath] 84 failed tests

[Ximin Luo]

Ximin Luo:
> Jeroen Demeyer:
>> On 2016-10-06 21:12, Ximin Luo wrote:
>>> Unfortunately the docbuild (even with _wip_docbuild2) still runs out of memory for me.
>>
>> Are you running the docbuild with many processes? On my laptop with 6GB of RAM, I can use at most 3 processes for docbuilding (in vanilla Sage) before it starts swapping. So I guess with 8GB, you should use at most 4 processes.
>>
>> What does "top" say? How much memory does a single Python process use when docbuilding?
>>
> 
> Oh, thanks for that reminder. Actually I disabled parallel sphinx building temporarily [1], because it was making things hard to debug. But it looks like I only disabled it for the reference-manual building and not for the AllBuilder; I will try that next.
> 
> Once everything builds without parallel-build [..]

I got this to work too, by disabling the parallel docbuild completely. I've pushed this to our git now, as the master branch.

My build uses up about 5GB for a single thread.

For quite a while, it does indeed use 2GB:

  PID USER      PR  NI    VIRT    RES    SHR S  %CPU %MEM     TIME+ COMMAND
23336 infinit+  20   0 5903056 1.949g 114796 R 100.0 25.3  10:20.83 python

but then after "Merging environment/index files", it keeps growing:

  PID USER      PR  NI    VIRT    RES    SHR S  %CPU %MEM     TIME+ COMMAND
23336 infinit+  20   0 8886072 4.678g  71240 R 100.3 60.8  29:59.80 python

> [..] I'll see if I can do the tweak that others mentioned, of using fewer processes for the docbuild.
> 
> On a side note, I'm pretty sure that this:
> 
>             except Exception:
>                 pool.terminate()
>                 if ABORT_ON_ERROR:
>                     raise
> 
> in both places that it occurs in docbuild/__init__.py, does not work as intended. And then this messes up subsequent builds, and I get "not found" errors. This is partly why I (tried to) disabled parallel docbuilding in the first place. But I'll try to explore this in more depth and come back with some more details.
> 
> X
> 
> [1] https://anonscm.debian.org/cgit/debian-science/packages/sagemath.git/tree/debian/patches/temp-disable-parallel-sphinx.patch
> 

I'll pause this line of investigation for now, the rest of the tests are a higher short-term priority.

X

-- 
GPG: ed25519/56034877E1F87C35
GPG: rsa4096/1318EFAC5FBBDBCE
https://github.com/infinity0/pubkeys.git