Can read() be non-blocking?

[Thomas Christensen]

This issue has been raised a couple of times I am sure. But I have yet
to find a satisfying answer.

I am reading from a subprocess and this subprocess sometimes hang, in
which case a call to read() call will block indefinite, keeping me from
killing it.

The folloing sample code illustrates the problem:

proc = subprocess.Popen(['/usr/bin/foo', '/path/to/some/file'],
stdout=subprocess.PIPE)
output = StringIO.StringIO()
while True:
r = select.select([proc.stdout.fileno()], [], [], 5)[0]
if r:
# NOTE: This will block since it reads until EOF
data = proc.stdout.read()
if not data:
break # EOF from process has been reached
else:
output.write(data)
else:
os.kill(proc.pid, signal.SIGKILL)
proc.wait()

<Process the output...>

As the NOTE: comment above suggests the call to read() will block here.

I see two solutions:

1. Read one byte at a time, meaning call read(1).
2. Read non-blocking.

I think reading one byte at a time is a waste of CPU, but I cannot find
a way to read non-blocking.

Is there a way to read non-blocking? Or maybe event a better way in
generel to handle this situation?

Thanks

Thomas

 

[Lawrence D'Oliveiro]

r = select.select([proc.stdout.fileno()], [], [], 5)[0]
if r:
# NOTE: This will block since it reads until EOF
data = proc.stdout.read()

No, it will read what data is available.

 

[Lawrence D'Oliveiro]

r = select.select([proc.stdout.fileno()], [], [], 5)[0]
if r:
# NOTE: This will block since it reads until EOF
data = proc.stdout.read()

No, it will read what data is available.

Sorry, maybe not. But you can set O_NOBLOCK on the fd.

 

[Ò»Ê×Ê«]

This issue has been raised a couple of times I am sure.  But I have yet
to find a satisfying answer.

I am reading from a subprocess and this subprocess sometimes hang, in
which case a call to read() call will block indefinite, keeping me from
killing it.

The folloing sample code illustrates the problem:

  proc = subprocess.Popen(['/usr/bin/foo', '/path/to/some/file'],
                          stdout=subprocess.PIPE)
  output = StringIO.StringIO()
  while True:
      r = select.select([proc.stdout.fileno()], [], [], 5)[0]
      if r:
          # NOTE: This will block since it reads until EOF
          data = proc.stdout.read()
          if not data:
              break  # EOF from process has been reached
          else:
              output.write(data)
      else:
          os.kill(proc.pid, signal.SIGKILL)
  proc.wait()

  <Process the output...>

As the NOTE: comment above suggests the call to read() will block here.

I see two solutions:

1. Read one byte at a time, meaning call read(1).
2. Read non-blocking.

I think reading one byte at a time is a waste of CPU, but I cannot find
a way to read non-blocking.

Is there a way to read non-blocking?  Or maybe event a better way in
generel to handle this situation?

Thanks

                Thomas

As far as I know, you can use '''fctnl''' to make a file handle non-
blocking.

But :

1. I don't know if it works on standard io
2. If it works in python

 

[Karthik Gurusamy]

This issue has been raised a couple of times I am sure.  But I have yet
to find a satisfying answer.

I am reading from a subprocess and this subprocess sometimes hang, in
which case a call to read() call will block indefinite, keeping me from
killing it.

The folloing sample code illustrates the problem:

  proc = subprocess.Popen(['/usr/bin/foo', '/path/to/some/file'],
                          stdout=subprocess.PIPE)
  output = StringIO.StringIO()
  while True:
      r = select.select([proc.stdout.fileno()], [], [], 5)[0]
      if r:
          # NOTE: This will block since it reads until EOF
          data = proc.stdout.read()
          if not data:
              break  # EOF from process has been reached
          else:
              output.write(data)
      else:
          os.kill(proc.pid, signal.SIGKILL)
  proc.wait()

  <Process the output...>

As the NOTE: comment above suggests the call to read() will block here.

I see two solutions:

1. Read one byte at a time, meaning call read(1).
2. Read non-blocking.

I think reading one byte at a time is a waste of CPU, but I cannot find
a way to read non-blocking.

Is there a way to read non-blocking?  Or maybe event a better way in
generel to handle this situation?

From what I understand, you want a way to abort waiting on a blocking
read if the process is hung.
There are some challenges about how you decide if the process is hung
or just busy doing work without generating output for a while (or may
be the system is busy and the process didn't get enough CPU due to
other CPU hungry processes).
Assuming you have a valid way to figure this out, one option is to
have a timeout on the read.
If the timeout exceeds, you abort the read call. No, the read doesn't
provide a timeout, you can build one using alarm.

def alarm_handler(*args):
""" This signal stuff may not work in non unix env """
raise Exception("timeout")

signal.signal(signal.SIGALRM, alarm_handler)

try:
signal.alarm(timeout) # say timeout=60 for a max wait of 1
minute
data = proc.stdout.read()
except Exception, e:
if not str(e) == 'timeout': # something else went wrong ..
raise
# got the timeout exception from alarm .. proc is hung; kill it

Karthik

 

[saju.pillai]

In message <[email protected]>, Thomas
Christensen wrote:

      r = select.select([proc.stdout.fileno()], [], [], 5)[0]
      if r:
          # NOTE: This will block since it reads until EOF
          data = proc.stdout.read()

No, it will read what data is available.

Sorry, maybe not. But you can set O_NOBLOCK on the fd.

Set O_NONBLOCK on proc.fileno() and try using os.read() on that
descriptor.

-srp