On page 145 of Operating System Concepts (Avi Silberschatz, 9th edition), it states that
Note that ordinary pipes require a parent–child relationship between the communicating processes on both UNIX and Windows systems. This
means that these pipes can be used only for communication between
processes on the same machine.
However, when I execute the command ps -el | grep 0 on a linux machine, both processes ps and grep have bash as their parent process. Other resources confirms that the shell pipe is an ordinary pipe that is unidirectional. If a parent-child relationship is required, how come the ps and grep process are siblings under bash? Just to confirm my understanding about pipes, is the statement in the textbook simply wrong or have I misunderstood something here?
The statement is true, but must be interpreted mixing with descriptor inheritance. What it means is that to setup a piping you need a parent-child relationship, not that communication with pipe is only possible between parent and children. So in your case, bash creates the pipe and transmit it to both children ps and grep (one writing in it while the other reads). Thus, both children communicates with the a pipe created by their parent.
I'm using Zabbix 3.2. I want to calculate the traffic statistics on network interface based on the program name.
Like for getting total incoming traffic, we use net.if.in[if,] , by same way is it possible to retreive traffic utilized by each running process like in Nethogs. If so, pls share the Item key. Or, if there is any sh script to do the same.
Thanks in advance.
You haven't specified the operating system, but the question is tagged 'unix' and you mention nethogs and shell scripts - I'll assume Linux.
It might be a bit too much to monitor traffic for all of the processes - there could be hundreds of them, and even though many would not use the network, on a server system many would.
It is also important how you want to structure the data. For example, do you want to split it up per process name, or per individual process? Or maybe even process name and its parameters - in case of running several Java JVMs on the same box. You would have to decide on all this, as it will affect the data collection.
As sending data to Zabbix, the simplest way on the Zabbix side would be monitoring by process name only, and creating items in advance, if you know all the process names you will be interested in. If you do not know them, you will have to use Zabbix low level discovery to automatically create items as new processes appear.
And we finally get to the data collection part. Here, it indeed might be the easiest to use nethogs (keeping in mind that UDP is not supported). You can run nethogs in "trace" mode, which is pretty much the same as the "batch" mode for top. In this mode, output is simply printed to stdout.
nethogs -c 1 -d 60 -t
Here, the parameters mean:
-c - how many times to print output
-d - for how long to sleep between iterations, including the time before the first output
-t - tracing or batch mode
Nethogs also supports setting traffic output type with the -v flag. You'd have to decide how you want to visualise this:
0 - KB/s
1 - total KB
2 - total B
3 - total MB
With Zabbix, you probably will not want to use modes 1 or 3 - it is better to store data in bytes and allow Zabbix to add the multiplier as needed. In case of the KB/s mode (0), it is probably worth adding an item multiplier of 1024 to store data in bytes and again benefiting from the automatic unit application at Zabbix. Note that in any case you will want to run nethog instances back-to-back, to avoid windows where you are not collecting data. One way to minimise possibility of any windows would be running nethogs constantly (without supplying -c option) and redirecting output to a file. A script would then parse the file and send the data to Zabbix with zabbix_sender.
You wouldn't run this as a normal Zabbix user parameter, neither as an active, nor passive check - it would block for too long. Consider using atd (see this howto) or nohup to launch a script that sends data to Zabbix with zabbix_sender instead.
Note that you must run nethogs as root - use sudo for that.
I'm not aware of any existing scripts for this, but the following might get you started:
nethogs -c 1 -d 1 -t | awk 'BEGIN {FS="[[:space:]/]+"}; /Refreshing/,0 \
{if ($1 != "Refreshing:" && $1 != "unknown") {print $(NF-4), $(NF-1), $NF}}'
Here, awk grabs only program lines and prints out program name and sent/received traffic.
Stdin and stdout are single files that are shared by multiple processes to take in input from the users. So how does the OS make sure that only the input give to a particular program is visible in the stdin for than program?
Your assumption that stdin/stdout (while having the same logical name) are shared among all processes is wrong at best.
stdin/stdout are logical names for open files that are forwarded (or initialized) by the process that has started a given process. Actually, with the standard fork-and-exec pattern the setup of those may occur already in the new process (after fork) before exec is being called.
stdin/stdout usually are just inherited from parent. So, yes there exist groups of processes that share stdin and/or stdout for a given filenode.
Also, as a filedescriptor may be a side of a pipe, you need not have file from a filesystem (or a device node) linked to any of the well known standard channels (you also should include stderr into your considerations).
The normal way of setup is:
the parent (e.g. your shell) is calling fork
the forked process (child) is setting up environment, standard I/O channels and anything else.
the child then executes exec to overlay the process with the target image to be executed.
When setting up: it either will keep the existing channels or replace them with new ones e.g. creating a pipe and linking the endpoints appropriately. (To be honest, creating the pipe need to happen before the fork in that simplified description)
This way, most of the process have their own I/O channels.
Nevertheless, multiple processes may write into a channel they are connected to (have a valid filedescriptor to). When reading each junk of data (usually lines with terminals or blocks with files) is being read by a single reader only. So if you have several (running) processes reading from a terminal as stdin, only one will read your typing, while the other(s) will not see this typing at all.
my program has two processes.one process writes to FIFO(named pipe) and must wait until another process read from FIFO and then it waits for that process to return the result(writing to FIFO to be read by this process).
my question is that how to know the data is read form FIFO by another process and then call read() system call for result?
Most UNIXen have bidirectional pipes (man pipe)
Linux, IIRC hasn't got those, so you need to use socket_pair, which can conveniently use UNIX domain sockets giving roughly the same functionality.
In my experience porting code often required little else beyond replacing the call to pipe by a call to socket_pair
I created a program that iterates over a bunch of files and invokes for some of them:
scp <file> user#host:<remotefile>
However, in my case, there may be thousands of small files that need to transferred, and scp is opening a new ssh connection for each of them, which has quite some overhead.
I was wondering if there is no solution where I keep one process running that maintains the connection and I can send it "requests" to copy over single files.
Ideally, I'm looking for a combination of some sender and receiver program, such that I can start a single process (1) at the beginning:
ssh user#host receiverprogram
And for each file, I invoke a command (2):
senderprogram <file> <remotefile>
and pipe the output of (2) to the input of (1), and this would cause the file to be transferred. In the end, I can just send process (1) some signal to terminate.
Preferably the sender and receiver programs are open source C programs for Unix. They may communicate using a socket instead of a pipe, or any other creative solution.
However, it is an important constraint that each file gets transferred at the moment I iterate over it: it is not acceptable to collect a list of files and then invoke one instance of scp to transfer all the files at once at the end. Also, I have only simple shell access to the receiving host.
Update: I found a solution for the problem of the connection overhead using the multiplexing features of ssh, see my own answer below. Yet, I'm starting a bounty because I'm curious to find if there exists a sender/receiver program as I describe here. It seems there should exist something that can be used, e.g. xmodem/ymodem/zmodem?
I found a solution from another angle. Since version 3.9, OpenSSH supports session multiplexing: a single connection can carry multiple login or file transfer sessions. This avoids the set-up cost per connection.
For the case of the question, I can first open a connection with sets up a control master (-M) with a socket (-S) in a specific location. I don't need a session (-N).
ssh user#host -M -S /tmp/%r#%h:%p -N
Next, I can invoke scp for each file and instruct it to use the same socket:
scp -o 'ControlPath /tmp/%r#%h:%p' <file> user#host:<remotefile>
This command starts copying almost instantaneously!
You can also use the control socket for normal ssh connections, which will then open immediately:
ssh user#host -S /tmp/%r#%h:%p
If the control socket is no longer available (e.g. because you killed the master), this falls back to a normal connection. More information is available in this article.
This way would work, and for other things, this general approach is more or less right.
(
iterate over file list
for each matching file
echo filename
) | cpio -H newc -o | ssh remotehost cd location \&\& | cpio -H newc -imud
It might work to use sftp instead of scp, and to place it into batch mode. Make the batch command file a pipe or UNIX domain socket and feed commands to it as you want them executed.
Security on this might be a little tricky at the client end.
Have you tried sshfs?
You could:
sshfs remote_user#remote_host:/remote_dir /mnt/local_dir
Where
/remote_dir was the directory you want to send files to on the system you are sshing into
/mnt/local_dir was the local mount location
With this setup you can just cp a file into the local_dir and it would be sent over sftp to remote_host in its remote_dir
Note that there is a single connection, so there is little in the way of overhead
You may need to use the flag -o ServerAliveInterval=15 to maintain an indefinite connection
You will need to have fuse installed locally and an SSH server supporting (and configured for) sftp
May be you are looking for this:
ZSSH
zssh (Zmodem SSH) is a program for interactively transferring files to a remote machine while using the secure shell (ssh). It is intended to be a convenient alternative to scp , allowing to transfer files without having to open another session and re-authenticate oneself.
Use rsync over ssh if you can collect all the files to send in a single directory (or hierarchy of directories).
If you don't have all the files in a single place, please give some more informations as to what you want to achieve and why you can't pack all the files into an archive and send that over. Why is it so vital that each file is sent immediately? Would it be OK if the file was sent with a short delay (like when 4K worth of data has accumulated)?
It's a nice little problem. I'm not aware of a prepackaged solution, but you could do a lot with simple shell scripts. I'd try this at the receiver:
#!/bin/ksh
# this is receiverprogram
while true
do
typeset -i length
read filename # read filename sent by sender below
read size # read size of file sent
read -N $size contents # read all the bytes of the file
print -n "$contents" > "$filename"
done
At the sender side I would create a named pipe and read from the pipe, e.g.,
mkfifo $HOME/my-connection
ssh remotehost receiver-script < $HOME/my-connection
Then to send a file I'd try this script
#!/bin/ksh
# this is senderprogram
FIFO=$HOME/my-connection
localname="$1"
remotename="$2"
print "$remotename" > $FIFO
size=$(stat -c %s "$localname")
print "$size" > $FIFO
cat "$localname" > $FIFO
If the file size is large you probably don't want to read it at one go, so something on the order of
BUFSIZ=8192
rm -f "$filename"
while ((size >= BUFSIZ)); do
read -N $BUFSIZE buffer
print -n "$buffer" >> "$filename"
size=$((size - BUFSIZ))
done
read -N $size buffer
print -n "$contents" >> "$filename"
Eventually you'll want to extend the script so you can pass through chmod and chgrp commands. Since you trust the sending code, it's probably easiest to structure the thing so that the receiver simply calls shell eval on each line, then send stuff like
print filename='"'"$remotename"'"' > $FIFO
print "read_and_copy_bytes " '$filename' "$size" > $FIFO
and then define a local function read_and_copy_bytes. Getting the quoting right is a bear, but otherwise it should be straightforward.
Of course, none of this has been tested! But I hope it gives you some useful ideas.
Seems like a job for tar? Pipe its output to ssh, and on the other side pipe the ssh output back to tar.
I think that the GNOME desktop uses a single SSH connection when accessing a share through SFTP (SSH). I'm guessing that this is what's happening because I see a single SSH process when I access a remote share this way. So if this is true you should be able to use the same program for this purpose.
The new version of GNOME used GVFS through GIO in order to perform all kind of I/O through different backends. The Ubuntu package gvfs-bin provides various command line utilities that let you manipulate the backends from the command line.
First you will need to mount your SSH folder:
gvfs-mount sftp://user#host/
And then you can use the gvfs-copy to copy your files. I think that all file transfers will be performed through a single SSH process. You can even use ps to see which process is being used.
If you feel more adventurous you can even write your own program in C or in some other high level language that provides an API to GIO.
One option is Conch is a SSH client and server implementation written in Python using the Twsited framework. You could use it to write a tool which accepts requests via some other protocol (HTTP or Unix domain sockets, FTP, SSH or whatever) and triggers file transfers over a long running SSH connection. In fact, I have several programs in production which use this technique to avoid multiple SSH connection setups.
There was a very similar question here a couple of weeks ago. The accepted answer proposed to open a tunnel when ssh'ing to the remote machine and to use that tunnel for scp transfers.
Perhapse CurlFTPFS might be a valid solution for you.
It looks like it just mounts an external computer's folder to your computer via SFTP. Once that's done, you should be able to use your regular cp commands and everything will be done securely.
Unfortunately I was not able to test it out myself, but let me know if it works for ya!
Edit 1: I have been able to download and test it. As I feared it does require that the client have a FTP server. However, I have found another program which does has exactly the same concept as what you are looking for. sshfs allows you to connect to your client computer without needing any special server. Once you have mounted one of their folders, you can use your normal cp commands to move whatever files you need to more. Once you are done, it should then be a smile matter of umount /path/to/mounted/folder. Let me know how this works out!
rsync -avlzp user#remotemachine:/path/to/files /path/to/this/folder
This will use SSH to transfer files, in a non-slow way
Keep it simple, write a little wrapper script that does something like this.
tar the files
send the tar-file
untar on the other side
Something like this:
tar -cvzf test.tgz files ....
scp test.tgz user#other.site.com:.
ssh user#other.site.com tar -xzvf test.tgz
/Johan