Option to hide `stdout` for non-task
gaow opened this issue · comments
I seem to recall we have talked about this before -- for non-tasks, is there a way to hide, or better, redirect stdout / stderr elsewhere just like for task? The output of some programs are so overwhelming that it drastically slows down the entire process because of huge amount of text printed to the terminal ...
-v1 would suppress output of script execution...
-v1 would suppress output of script execution...
Indeed, but not stderr unless I use -v0. Also because there are many concurrent processes going on, ideally if we can configure stderr and stdout to write to files it would be easier to check them later.
The line you pointed to is for task. this line is for regular execution, which says,
stderr=None if env.verbosity > 0 else subprocess.DEVNULL,
stdout=None if env.verbosity > 1 else subprocess.DEVNULL
So v1 shows stderr but not stdout. This makes sense because stderr is meant for errors and should have higher priority.
I agree. But unfortunately some programs prints debug information or warning messages to stderr which can be a bit overwhelming. If there is really an error then the step should fail and throw an exception anyways. For R and python packages it is possible to handle warning messages in the code, but not for bash executables. For such cases I was hoping that SoS can redirect the stderr output to a file called {_output}.log via some options, eg:
R: stdout = f'{_output}.out', stderr = f'{_output}.err'
...
Not sure if it is a good idea, but it looks useful to me. After all the only reason to use non-task mode is performance of job submission; that often means there involve lots of jobs, and such messages will easily get overwhelming.
Sounds like a good idea ...
but wondering what should happen for
R: stdout = f'{_output}.log', stderr = f'{_output}.log'
...
And for
input: for_each ...
R: stdout='samefile.log'
...
Should we let users take the risk?
Should we let users take the risk?
I guess it is fair enough that users take the risk. And we make it w mode not a for append? I am thinking of scenarios when there are more than 2 actions in the same step like #880; but maybe users should just name them differently ...
I am not sure about the w and a part because stdout and stderr are streams and are supposed to be in a mode.
I see. I guess my question is whether or not to remove those files, if exist, before running the code that will write to them, thus preventing repeated runs adding too much to it. I assumed this is natural to do. But the corner case would be :
R: stdout = f'{_output}.log', stderr = f'{_output}.log'
...
python: stdout = f'{_output}.log', stderr = f'{_output}.log'
...
So removing existing f'{_output}.log' should only happen at the first action, not the 2nd one. Then after that everything should append.
I used ab to open file without any error checking. Please test.
These options right now only accept a file, but users might want to use stdout=subprocess.DEVNULL.
Great! It works perfectly.
These options right now only accept a file, but users might want to use stdout=subprocess.DEVNULL.
Right, I think it would be useful to have. Then it'd be stdout = None?
The problem is that stdout=None is the subprocess.Popen default for standard output.
Right. But from my reading of the code we are parsing the options anyways. So can we change the behavior such that we keep the default behavior if stdout/stderr is not seen in kwargs, and when seen None will get translated to DEVNULL? Or other keywords such as /dev/null will do, too?
We can certainly implement\ the behavior (missing as standard, None as DEVNULL) but this is not very pythonic so I am trying to see if there can be a better even more general method (e.g. can handle non-path object like DEVNULL and allow users to pass arbitrary object with write attribute).
But I agree that stdout=None is the most obvious method for users.
We actually have another choice, stdout=False, which seems to translate better to no standard output.
Okey, I changed it from None to False, and added a line in the error message prompt to point to the error file location. Should I push the patch below and document? I do not think it impacts unit test as far as I can tell.
diff --git a/src/sos/actions.py b/src/sos/actions.py
index 4b28e5a6..38d99d51 100644
--- a/src/sos/actions.py
+++ b/src/sos/actions.py
@@ -323,14 +323,14 @@ class SoS_ExecuteScript:
stderr=subprocess.PIPE, bufsize=0)
out, err = child.communicate()
if 'stdout' in kwargs:
- if kwargs['stdout'] is not None:
+ if kwargs['stdout']:
with open(kwargs['stdout'], 'ab') as so:
so.write(out)
else:
sys.stdout.write(out.decode())
if 'stderr' in kwargs:
- if kwargs['stderr'] is not None:
+ if kwargs['stderr']:
with open(kwargs['stderr'], 'ab') as se:
se.write(err)
else:
@@ -343,7 +343,7 @@ class SoS_ExecuteScript:
elif '__std_out__' in env.sos_dict and '__std_err__' in env.sos_dict:
if 'stdout' in kwargs or 'stderr' in kwargs:
if 'stdout' in kwargs:
- if kwargs['stdout'] is None:
+ if kwargs['stdout'] is False:
so = subprocess.DEVNULL
else:
so = open(kwargs['stdout'], 'ab')
@@ -353,7 +353,7 @@ class SoS_ExecuteScript:
so = subprocess.DEVNULL
if 'stderr' in kwargs:
- if kwargs['stderr'] is None:
+ if kwargs['stderr'] is False:
se = subprocess.DEVNULL
else:
se = open(kwargs['stderr'], 'ab')
@@ -379,7 +379,7 @@ class SoS_ExecuteScript:
ret = p.wait()
else:
if 'stdout' in kwargs:
- if kwargs['stdout'] is None:
+ if kwargs['stdout'] is False:
so = subprocess.DEVNULL
else:
so = open(kwargs['stdout'], 'ab')
@@ -389,7 +389,7 @@ class SoS_ExecuteScript:
so = subprocess.DEVNULL
if 'stderr' in kwargs:
- if kwargs['stderr'] is None:
+ if kwargs['stderr'] is False:
se = subprocess.DEVNULL
else:
se = open(kwargs['stderr'], 'ab')
@@ -412,11 +412,12 @@ class SoS_ExecuteScript:
debug_args = '{filename:q}'
else:
debug_args = self.args
- cmd = interpolate(f'{self.interpreter} {debug_args}',
+ cmd = interpolate(f'{self.interpreter.strip()} ``{debug_args}``',
{'filename': sos_targets(debug_script_file), 'script': self.script})
- raise RuntimeError('Failed to execute commmand ``{}`` (ret={}, workdir={}{})'.format(
+ raise RuntimeError('Failed to execute commmand "{}" (ret={}, workdir={}{}{})'.format(
cmd, ret, os.getcwd(),
- f', task={os.path.basename(env.sos_dict["__std_err__"]).split(".")[0]}' if '__std_out__' in env.sos_dict else ''))
+ f', task={os.path.basename(env.sos_dict["__std_err__"]).split(".")[0]}' if '__std_out__' in env.sos_dict else '',
+ f', err=``{kwargs["stderr"]}``' if 'stderr' in kwargs and os.path.isfile(kwargs['stderr']) else ''))
except RuntimeError:
raise
except Exception as e:Ok with me.