Re: [PATCH] process: Flush server stdout/stderr before reporting failure

[Jan Kiszka <jan.kiszka@siemens.com>]

On 2018-09-04 14:34, Richard Purdie wrote:
> On Tue, 2018-09-04 at 13:15 +0100, Richard Purdie wrote:
>> On Sun, 2018-09-02 at 10:43 +0200, Jan Kiszka wrote:
>>> We need to print exceptions and flush output channels before
>>> telling
>>> the
>>> client that the server failed. Otherwise the daemon log file may
>>> still
>>> be empty when the client opens it for printing.
>>>
>>> This fixes error reporting when the output is redirected to a pipe,
>>> e.g.
>>> "bitbake target | cat" with a syntax error in a recipes or config
>>> file.
>>> That's specifically annoying when bitbake is under the control of a
>>> frontend process such as kas.
>>
>> Hi,
>>
>> Could you check if this was already fixed by:
>>
>> http://git.yoctoproject.org/cgit.cgi/poky/commit/?id=cbe2d3cb0af6c7a5
>> cd368e7dc489960a13648bd0

Looks similar on first glance, but it only partly helps with exceptions. 
We get from

$ bitbake | cat
ERROR: Unable to start bitbake server

to

$ bitbake | cat
ERROR: Unable to start bitbake server
ERROR: Server log for this session (/work/build/bitbake-cookerdaemon.log):
--- Starting bitbake server pid 149 at 2018-09-04 13:02:15.144583 ---

while it should have

$ bitbake
ERROR: Unable to start bitbake server
ERROR: Last 10 lines of server log for this session 
(/work/build/bitbake-cookerdaemon.log):
Traceback (most recent call last):
   File "/work/isar/bitbake/lib/bb/daemonize.py", line 77, in createDaemon
     function()
   File "/work/isar/bitbake/lib/bb/server/process.py", line 433, in 
_startServer
     self.cooker = bb.cooker.BBCooker(self.configuration, self.featureset)
   File "/work/isar/bitbake/lib/bb/cooker.py", line 178, in __init__
     self.configwatcher = pyinotify.WatchManager()
   File "/work/isar/bitbake/lib/pyinotify.py", line 1764, in __init__
     raise OSError(err % self._inotify_wrapper.str_errno())
OSError: Cannot initialize new instance of inotify, Errno=Too many open 
files (EMFILE)

>>
>> ?
>>
>> I really want to get these flushes as close to the source of the
>> problems as we can.
> 
> Looking at the code further, I suspect we should simply be doing:
> 
>          writer = ConnectionWriter(self.readypipein)
>          self.cooker = bb.cooker.BBCooker(self.configuration, self.featureset)
>          writer.send("ready")
>          os.close(self.readypipein)
>          server.cooker = self.cooker
> 
> with no exception handling. The ready pipe would then get closed by the
> os._exit() in daemonize which is after the flush. It simplifies the
> code as an added bonus...

Tried that, but it does not work that way, it just hangs.

As I wrote, the test case is trivial: just direct the output of bitbake 
into a pipe (bitbake | cat).

Jan