1 Embedded Solaris
This chapter describes the OS specific parts of OTP which relate to Solaris.
1.1 Memory Usage
Solaris takes about 17 Mbyte of RAM on a system with 64 Mbyte of total RAM. This leaves about 47 Mbyte for the applications. If the system utilizes swapping, these figures cannot be improved because unnecessary daemon processes are swapped out. However, if swapping is disabled, or if the swap space is of limited resource in the system, it becomes necessary to kill off unnecessary daemon processes.
1.2 Disk Space Usage
The disk space required by Solaris can be minimized by using the Core User support installation. It requires about 80 Mbyte of disk space. This installs only the minimum software required to boot and run Solaris. The disk space can be further reduced by deleting unnecessary individual files. However, unless disk space is a critical resource the effort required and the risks involved may not be justified.
1.3 Installation
This section is about installing an embedded system. The following topics are considered,
-
Creation of user and installation directory,
-
Installation of embedded system,
-
Configuration for automatic start at reboot,
-
Making a hardware watchdog available,
-
Changing permission for reboot,
-
Patches,
-
Configuration of the OS_Mon application.
Several of the procedures described below require expert knowledge of the Solaris 2 operating system. For most of them super user privilege is needed.
Creation of User and Installation Directory
It is recommended that the Embedded Environment is run by an ordinary user, i.e. a user who does not have super user privileges.
Throughout this section we assume that the user name is otpuser, and that the home directory of that user is,
/export/home/otpuser
Furthermore, we assume that in the home directory of otpuser, there is a directory named otp, the full path of which is,
/export/home/otpuser/otp
This directory is the installation directory of the Embedded Environment.
Installation of an Embedded System
The procedure for installation of an embedded system does not differ from that of an ordinary system (see the Installation Guide), except for the following:
-
the (compressed) tape archive file should be extracted in the installation directory as defined above, and,
-
there is no need to link the start script to a standard directory like /usr/local/bin.
Configuration for Automatic Start at Boot
A true embedded system has to start when the system boots. This section accounts for the necessary configurations needed to achieve that.
The embedded system and all the applications will start automatically if the script file shown below is added to the /etc/rc3.d directory. The file must be owned and readable by root, and its name cannot be arbitrarily assigned. The following name is recommended,
S75otp.system
For further details on initialization (and termination) scripts, and naming thereof, see the Solaris documentation.
#!/bin/sh # # File name: S75otp.system # Purpose: Automatically starts Erlang and applications when the # system starts # Author: janne@erlang.ericsson.se # Resides in: /etc/rc3.d # if [ ! -d /usr/bin ] then # /usr not mounted exit fi killproc() { # kill the named process(es) pid=`/usr/bin/ps -e | /usr/bin/grep -w $1 | /usr/bin/sed -e 's/^ *//' -e 's/ .*//'` [ "$pid" != "" ] && kill $pid } # Start/stop processes required for Erlang case "$1" in 'start') # Start the Erlang emulator # su - otpuser -c "/export/home/otpuser/otp/bin/start" & ;; 'stop') killproc beam ;; *) echo "Usage: $0 { start | stop }" ;; esac
The file /export/home/otpuser/otp/bin/start referred to in the above script, is precisely the script start described in the section Starting Erlang below. The script variable OTP_ROOT in that start script corresponds to the example path
/export/home/otpuser/otp
used in this section. The start script should be edited accordingly.
Use of the killproc procedure in the above script could be combined with a call to erl_call, e.g.
$SOME_PATH/erl_call -n Node init stop
In order to take Erlang down gracefully see the erl_call(1) reference manual page for further details on the use of erl_call. That however requires that Erlang runs as a distributed node which is not always the case.
The killproc procedure should not be removed: the purpose is here to move from run level 3 (multi-user mode with networking resources) to run level 2 (multi-user mode without such resources), in which Erlang should not run.
Hardware Watchdog
For Solaris running on VME boards from Force Computers, there is a possibility to activate the onboard hardware watchdog, provided a VME bus driver is added to the operating system (see also Installation Problems below).
See also the heart(3) reference manual page in Kernel.
Changing Permissions for Reboot
If the HEART_COMMAND environment variable is to be set in the start script in the section, Starting Erlang, and if the value shall be set to the path of the Solaris reboot command, i.e.
HEART_COMMAND=/usr/sbin/reboot
the ownership and file permissions for /usr/sbin/reboot must be changed as follows,
chown 0 /usr/sbin/reboot chmod 4755 /usr/sbin/reboot
See also the heart(3) reference manual page in Kernel.
The TERM Environment Variable
When the Erlang runtime system is automatically started from the S75otp.system script the TERM environment variable has to be set. The following is a minimal setting,
TERM=sun
which should be added to the start script described in the section.
Patches
For proper functioning of flushing file system data to disk on Solaris 2.5.1, the version specific patch with number 103640-02 must be added to the operating system. There may be other patches needed, see the release README file <ERL_INSTALL_DIR>/README.
Installation of Module os_sup in Application OS_Mon
The following four installation procedures require super user privilege.
-
Make a copy the Solaris standard configuration file for syslogd.
-
Make a copy the Solaris standard configuration file for syslogd. This file is usually named syslog.conf and found in the /etc directory.
-
The file name of the copy must be syslog.conf.ORIG but the directory location is optional. Usually it is /etc.
A simple way to do this is to issue the command
cp /etc/syslog.conf /etc/syslog.conf.ORIG
-
-
Make an Erlang specific configuration file for syslogd.
-
Make an edited copy of the back-up copy previously made.
-
The file name must be syslog.conf.OTP and the path must be the same as the back-up copy.
-
The format of the configuration file is found in the man page for syslog.conf(5), by issuing the command man syslog.conf.
-
Usually a line is added which should state:
-
which types of information that will be supervised by Erlang,
-
the name of the file (actually a named pipe) that should receive the information.
-
-
If e.g. only information originating from the unix-kernel should be supervised, the line should begin with kern.LEVEL (for the possible values of LEVEL see syslog.conf(5)).
-
After at least one tab-character, the line added should contain the full name of the named pipe where syslogd writes its information. The path must be the same as for the syslog.conf.ORIG and syslog.conf.OTP files. The file name must be syslog.otp.
-
If the directory for the syslog.conf.ORIG and syslog.conf.OTP files is /etc the line in syslog.conf.OTP will look like:
kern.LEVEL /etc/syslog.otp
-
-
Check the file privileges of the configuration files.
-
The configuration files should have rw-r--r-- file privileges and be owned by root.
-
A simple way to do this is to issue the commands
chmod 644 /etc/syslog.conf chmod 644 /etc/syslog.conf.ORIG chmod 644 /etc/syslog.conf.OTP
-
Note: If the syslog.conf.ORIG and syslog.conf.OTP files are not in the /etc directory, the file path in the second and third command must be modified.
-
-
Modify file privileges and ownership of the mod_syslog utility.
-
The file privileges and ownership of the mod_syslog utility must be modified.
-
The full name of the binary executable file is derived from the position of the os_mon application if the file system by adding /priv/bin/mod_syslog. The generic full name of the binary executable file is thus
<OTP_ROOT>/lib/os_mon-<REV>/priv/bin/mod_syslog
Example: If the path to the otp-root is /usr/otp, thus the path to the os_mon application is /usr/otp/lib/os_mon-1.0 (assuming revision 1.0) and the full name of the binary executable file is /usr/otp/lib/os_mon-1.0/priv/bin/mod_syslog.
-
The binary executable file must be owned by root, have rwsr-xr-x file privileges, in particular the setuid bit of user must be set.
-
A simple way to do this is to issue the commands
cd <OTP_ROOT>/lib/os_mon-<REV>/priv/bin/mod_syslog chmod 4755 mod_syslog chown root mod_syslog
-
The following procedure does not require root privilege.
-
Ensure that the configuration parameters for the os_sup module in the os_mon application are correct.
-
Browse the application configuration file (do not edit it). The full name of the application configuration file is derived from the position of the OS_Mon application if the file system by adding /ebin/os_mon.app.
The generic full name of the file is thus
<OTP_ROOT>/lib/os_mon-<REV>/ebin/os_mon.app.
Example: If the path to the otp-root is /usr/otp, thus the path to the os_mon application is /usr/otp/lib/os_mon-1.0 (assuming revision 1.0) and the full name of the binary executable file is /usr/otp/lib/os_mon-1.0/ebin/os_mon.app.
-
Ensure that the following configuration parameters are bound to the correct values.
Parameter | Function | Standard value |
start_os_sup | Specifies if os_sup will be started or not. | truefor the first instance on the hardware; falsefor the other instances. |
os_sup_own | The directory for (1)the back-up copy, (2) the Erlang specific configuration file for syslogd. | "/etc" |
os_sup_syslogconf | The full name for the Solaris standard configuration file for syslogd | "/etc/syslog.conf" |
error_tag | The tag for the messages that are sent to the error logger in the Erlang runtime system. | std_error |
If the values listed in the os_mon.app do not suit your needs, you should not edit that file. Instead you should override values in a system configuration file, the full pathname of which is given on the command line to erl.
Example: The following is an example of the contents of an application configuration file.
[{os_mon, [{start_os_sup, true}, {os_sup_own, "/etc"}, {os_sup_syslogconf, "/etc/syslog.conf"}, {os_sup_errortag, std_error}]}].
See also the os_mon(3), application(3) and erl(1) reference manual pages.
Installation Problems
The hardware watchdog timer which is controlled by the heart port program requires the FORCEvme package, which contains the VME bus driver, to be installed. This driver, however, may clash with the Sun mcp driver and cause the system to completely refuse to boot. To cure this problem, the following lines should be added to /etc/system:
- exclude: drv/mcp
- exclude: drv/mcpzsa
- exclude: drv/mcpp
It is recommended that these lines be added to avoid the clash described, which may make it completely impossible to boot the system.
1.4 Starting Erlang
This section describes how an embedded system is started. There are four programs involved, and they all normally reside in the directory <ERL_INSTALL_DIR>/bin. The only exception is the program start, which may be located anywhere, and also is the only program that must be modified by the user.
In an embedded system there usually is no interactive shell. However, it is possible for an operator to attach to the Erlang system by giving the command to_erl. He is then connected to the Erlang shell, and may give ordinary Erlang commands. All interaction with the system through this shell is logged in a special directory.
Basically, the procedure is as follows. The program start is called when the machine is started. It calls run_erl, which sets things up so the operator can attach to the system. It calls start_erl which calls the correct version of erlexec (which is located in <ERL_INSTALL_DIR>/erts-EVsn/bin) with the correct boot and config files.
1.5 Programs
start
This program is called when the machine is started. It may be modified or re-written to suit a special system. By default, it must be called start and reside in <ERL_INSTALL_DIR>/bin. Another start program can be used, by using the configuration parameter start_prg in the application sasl.
The start program must call run_erl as shown below. It must also take an optional parameter which defaults to <ERL_INSTALL_DIR>/releases/start_erl.data.
This program should set static parameters and environment variables such as -sname Name and HEART_COMMAND to reboot the machine.
The <RELDIR> directory is where new release packets are installed, and where the release handler keeps information about releases. See release_handler(3) in the application sasl for further information.
The following script illustrates the default behaviour of the program.
#!/bin/sh # Usage: start [DataFile] # ROOTDIR=/usr/local/otp if [ -z "$RELDIR" ] then RELDIR=$ROOTDIR/releases fi START_ERL_DATA=${1:-$RELDIR/start_erl.data} $ROOTDIR/bin/run_erl /tmp/ $ROOTDIR/log "exec $ROOTDIR/bin/start_erl \\ $ROOTDIR $RELDIR $START_ERL_DATA" > /dev/null 2>&1 &
The following script illustrates a modification where the node is given the name cp1, and the environment variables HEART_COMMAND and TERM have been added to the above script.
#!/bin/sh # Usage: start [DataFile] # HEART_COMMAND=/usr/sbin/reboot TERM=sun export HEART_COMMAND TERM ROOTDIR=/usr/local/otp if [ -z "$RELDIR" ] then RELDIR=$ROOTDIR/releases fi START_ERL_DATA=${1:-$RELDIR/start_erl.data} $ROOTDIR/bin/run_erl /tmp/ $ROOTDIR/log "exec $ROOTDIR/bin/start_erl \\ $ROOTDIR $RELDIR $START_ERL_DATA -heart -sname cp1" > /dev/null 2>&1 &
If a diskless and/or read-only client node is about to start the start_erl.data file is located in the client directory at the master node. Thus, the START_ERL_DATA line should look like:
CLIENTDIR=$ROOTDIR/clients/clientname START_ERL_DATA=${1:-$CLIENTDIR/bin/start_erl.data}
run_erl
This program is used to start the emulator, but you will not be connected to the shell. to_erl is used to connect to the Erlang shell.
Usage: run_erl pipe_dir/ log_dir "exec command [parameters ...]"
Where pipe_dir/ should be /tmp/ (to_erl uses this name by default) and log_dir is where the log files are written. command [parameters] is executed, and everything written to stdin and stdout is logged in the log_dir.
In the log_dir, log files are written. Each logfile has a name of the form: erlang.log.N where N is a generation number, ranging from 1 to 5. Each logfile holds up to 100kB text. As time goes by the following logfiles will be found in the logfile directory
erlang.log.1 erlang.log.1, erlang.log.2 erlang.log.1, erlang.log.2, erlang.log.3 erlang.log.1, erlang.log.2, erlang.log.3, erlang.log.4 erlang.log.2, erlang.log.3, erlang.log.4, erlang.log.5 erlang.log.3, erlang.log.4, erlang.log.5, erlang.log.1 ...
with the most recent logfile being the right most in each row of the above list. That is, the most recent file is the one with the highest number, or if there are already four files, the one before the skip.
When a logfile is opened (for appending or created) a time stamp is written to the file. If nothing has been written to the log files for 15 minutes, a record is inserted that says that we're still alive.
to_erl
This program is used to attach to a running Erlang runtime system, started with run_erl.
Usage: to_erl [pipe_name | pipe_dir]
Where pipe_name defaults to /tmp/erlang.pipe.N.
To disconnect from the shell without exiting the Erlang system, type Ctrl-D.
start_erl
This program starts the Erlang emulator with parameters -boot and -config set. It reads data about where these files are located from a file called start_erl.data which is located in the <RELDIR>. Each new release introduces a new data file. This file is automatically generated by the release handler in Erlang.
The following script illustrates the behaviour of the program.
#!/bin/sh # # This program is called by run_erl. It starts # the Erlang emulator and sets -boot and -config parameters. # It should only be used at an embedded target system. # # Usage: start_erl RootDir RelDir DataFile [ErlFlags ...] # ROOTDIR=$1 shift RELDIR=$1 shift DataFile=$1 shift ERTS_VSN=`awk '{print $1}' $DataFile` VSN=`awk '{print $2}' $DataFile` BINDIR=$ROOTDIR/erts-$ERTS_VSN/bin EMU=beam PROGNAME=`echo $0 | sed 's/.*\\///'` export EMU export ROOTDIR export BINDIR export PROGNAME export RELDIR exec $BINDIR/erlexec -boot $RELDIR/$VSN/start -config $RELDIR/$VSN/sys $*
If a diskless and/or read-only client node with the sasl configuration parameter static_emulator set to true is about to start the -boot and -config flags must be changed. As such a client cannot read a new start_erl.data file (the file is not possible to change dynamically) the boot and config files are always fetched from the same place (but with new contents if a new release has been installed). The release_handler copies this files to the bin directory in the client directory at the master nodes whenever a new release is made permanent.
Assuming the same CLIENTDIR as above the last line should look like:
exec $BINDIR/erlexec -boot $CLIENTDIR/bin/start \\ -config $CLIENTDIR/bin/sys $*