Supervisor: A System for Allowing the Control of Process State on UNIX
7/3/2006: updated for version 2.0
8/30/2006: updated for version 2.1
3/31/2007: updated for version 2.2
The supervisor is a client/server system that allows its users to control a number of processes on UNIX-like operating systems. It was inspired by the following:
- It is often inconvenient to need to write "rc.d" scripts for every single process instance. rc.d scripts are a great lowest-common-denominator form of process initialization/autostart/management, but they can be painful to write and maintain. Additionally, rc.d scripts cannot automatically restart a crashed process and many programs do not restart themselves properly on a crash. Supervisord starts processes as its subprocesses, and can be configured to automatically restart them on a crash. It can also automatically be configured to start processes on its own invocation.
- It's often difficult to get accurate up/down status on processes on UNIX. Pidfiles often lie. Supervisord starts processes as subprocesses, so it always knows the true up/down status of its children and can be queried conveniently for this data.
- Users who need to control process state often need only to do that. They don't want or need full-blown shell access to the machine on which the processes are running. Supervisorctl allows a very limited form of access to the machine, essentially allowing users to see process status and control supervisord-controlled subprocesses by emitting "stop", "start", and "restart" commands from a simple shell or web UI.
- Users often need to control processes on many machines. Supervisor provides a simple, secure, and uniform mechanism for interactively and automatically controlling processes on groups of machines.
- Processes which listen on "low" TCP ports often need to be started and restarted as the root user (a UNIX misfeature). It's usually the case that it's perfectly fine to allow "normal" people to stop or restart such a process, but providing them with shell access is often impractical, and providing them with root access or sudo access is often impossible. It's also (rightly) difficult to explain to them why this problem exists. If supervisord is started as root, it is possible to allow "normal" users to control such processes without needing to explain the intricacies of the problem to them.
- Processes often need to be started and stopped in groups, sometimes even in a "priority order". It's often difficult to explain to people how to do this. Supervisor allows you to assign priorities to processes, and allows user to emit commands via the supervisorctl client like "start all", and "restart all", which starts them in the preassigned priority order.
Supervisor has been tested and is known to run on Linux (Fedora Core 5, Ubuntu 6), Mac OS X (10.4), and Solaris (10 for Intel) and FreeBSD 6.1. It will likely work fine on most UNIX systems.
Supervisor will not run at all under any version of Windows.
Supervisor requires Python 2.3 or better.
Run "python setup.py install", then copy the "sample.conf" file to /etc/supervisord.conf and modify to your liking. If you'd rather not put the supervisord.conf file in /etc, you can place it anywhere and start supervisord and point it at the configuration file via the -c flag, e.g. "python supervisord.py -c /path/to/sample/conf" or, if you use the shell script named "supervisord", "supervisord -c /path/to/sample.conf".
I make reference below to a "$BINDIR" when explaining how to run supervisord and supervisorctl. This is the "bindir" directory that your Python installation has been configured with. For example, for an installation of Python installed via "./configure --prefix=/usr/local/python; make; make install", $BINDIR would be "/usr/local/python/bin". Python interpreters on different platforms use different $BINDIRs. Look at the output of "setup.py install" if you can't figure out where yours is.
To start supervisord, run $BINDIR/supervisord. The resulting process will daemonize itself and detach from the terminal. It keeps an operations log at "/tmp/supervisor.log" by default.
You can start supervisord in the foreground by passing the "-n" flag on its command line. This is useful to debug startup problems.
To change the set of programs controlled by supervisord, edit the supervisord.conf file and kill -HUP or otherwise restart the supervisord process. This file has several example program definitions.
Supervisord accepts a number of command-line overrides. Type 'supervisord -h' for an overview.
To start supervisorctl, run $BINDIR/supervisorctl. A shell will be presented that will allow you to control the processes that are currently managed by supervisord. Type "help" at the prompt to get information about the supported commands.
supervisorctl may be invoked with "one time" commands when invoked with arguments from a command line. An example: "supervisorctl stop all". If arguments are present on the supervisorctl command-line, it will prevent the interactive shell from being invoked. Instead, the command will be executed and supervisorctl will exit.
If supervisorctl is invoked in interactive mode against a supervisord that requires authentication, you will be asked for authentication credentials.
The server piece of the supervisor is named "supervisord". It is responsible for responding to commands from the client process as well as restarting crashed processes. It is meant to be run as the root user in most production setups. NOTE: see "Security Notes" at the end of this document for caveats!
The server process uses a configuration file. This is typically located in "/etc/supervisord.conf". This configuration file is an "Windows-INI" style config file. It is important to keep this file secure via proper filesystem permissions because it may contain unencrypted usernames and passwords.
The command-line client piece of the supervisor is named "supervisorctl". It provides a shell-like interface to the features provided by supervisord. From supervisorctl, a user can connect to different supervisord processes, get status on the subprocesses controlled by a supervisord, stop and start subprocesses of a supervisord, and get lists of running processes of a supervisord.
The command-line client talks to the server across a UNIX domain socket or an Internet socket. The server can assert that the user of a client should present authentication credentials before it allows him to perform commands. The client process may use the same configuration file as the server; any configuration file with a [supervisorctl] section in it will work.
Web ServerA (sparse) web user interface with functionality comparable to supervisorctl may be accessed via a browser if you start supervisord against an internet socket. Visit the server URL (e.g. http://localhost:9001/) to view and control process status through the web interface after changing the configuration file's 'http_port' parameter appropriately.
The same HTTP server which serves the web UI serves up an XML-RPC interface that can be used to interrogate and control supervisor and the programs it runs. To use the XML-RPC interface, connect to supervisor's http port with any XML-RPC client library and run commands against it. An example of doing this using Python's xmlrpclib client library:import xmlrpclib server = xmlrpclib.Server('http://localhost:9001')
Call methods against the supervisor and its subprocesses by using the 'supervisor' namespace:server.supervisor.getState()
You can get a list of methods supported by supervisor's XML-RPC interface by using the XML-RPC 'system.listMethods' API:server.system.listMethods()
You can see help on a method by using the 'system.methodHelp' API against the method:print server.system.methodHelp('supervisor.shutdown')
Supervisor's XML-RPC interface also supports the nascent XML-RPC multicall API described at http://www.xmlrpc.com/discuss/msgReader$1208.
Configuration File '[supervisord]' Section Settings
The supervisord.conf log file contains a section named '[supervisord]' in which global settings for the supervisord process should be inserted. These are:
'http_port' -- Either a TCP host:port value or (e.g. 127.0.0.1:9001) or a path to a UNIX domain socket (e.g. /tmp/supervisord.sock) on which supervisor will listen for HTTP/XML-RPC requests. Supervisorctl itself uses XML-RPC to communicate with supervisord over this port.
'sockchmod' -- Change the UNIX permission mode bits of the http_port UNIX domain socket to this value (ignored if using a TCP socket). Default: 0700.
'sockchown' -- Change the user and group of the socket file to this value. May be a username (e.g. chrism) or a username and group separated by a dot (e.g. chrism.wheel) Default: do not change.
'umask' -- The umask of the supervisord process. Default: 022.
'logfile' -- The path to the activity log of the supervisord process.
'logfile_maxbytes' -- The maximum number of bytes that may be consumed by the activity log file before it is rotated (suffix multipliers like "KB", "MB", and "GB" can be used in the value). Set this value to 0 to indicate an unlimited log size. Default: 50MB.
'logfile_backups' -- The number of backups to keep around resulting from activity log file rotation. Set this to 0 to indicate an unlimited number of backups. Default: 10.
'loglevel' -- The logging level, dictating what is written to the activity log. One of 'critical', 'error', 'warn', 'info', 'debug' or 'trace'. At log level 'trace', the supervisord log file will record the stderr/stdout output of its child processes, which is useful for debugging. Default: info.
'pidfile' -- The location in which supervisord keeps its pid file.
'nodaemon' -- If true, supervisord will start in the foreground instead of daemonizing. Default: false.
'minfds' -- The minimum number of file descriptors that must be available before supervisord will start successfully. Default: 1024.
'minprocs' -- The minimum nymber of process descriptors that must be available before supervisord will start successfully. Default: 200.
'nocleanup' -- prevent supervisord from clearing old "AUTO" log files at startup time. Default: false.
'http_username' -- the username required for authentication to our HTTP server. Default: none.
'http_password' -- the password required for authentication to our HTTP server. Default: none.
'childlogdir' -- the directory used for AUTO log files. Default: value of Python's tempfile.get_tempdir().
'user' -- if supervisord is run as root, switch users to this UNIX user account before doing any meaningful processing. This value has no effect if supervisord is not run as root. Default: do not switch users.
'directory' -- When supervisord daemonizes, switch to this directory. Default: do not cd.
- 'environment' -- A list of key/value pairs in the form "KEY=val,KEY2=val2"
- that will be placed in the supervisord process' environment (and as a result in all of its child process' environments). Default: none.
Configuration File '[supervisorctl]' Section Settings
The configuration file may contain settings for the supervisorctl interactive shell program. These options are listed below.
'serverurl' -- The URL that should be used to access the supervisord server, e.g. "http://localhost:9001". For UNIX domain sockets, use "unix:///absolute/path/to/file.sock".
'username' -- The username to pass to the supervisord server for use in authentication (should be same as 'http_username' in supervisord config). Optional.
'password' -- The password to pass to the supervisord server for use in authentication (should be the same as 'http_password' in supervisord config). Optional.
'prompt' -- String used as supervisorctl prompt. Default: supervisor.
Configuration File '[program:x]' Section Settings
The .INI file must contain one or more 'program' sections in order for supervisord to know which programs it should start and control. A sample program section has the following structure, the options of which are described below it:[program:programname] command=/path/to/programname priority=1 autostart=true autorestart=true startsecs=1 startretries=3 exitcodes=0,2 stopsignal=TERM stopwaitsecs=10 user=nobody log_stdout=true log_stderr=false logfile=/tmp/programname.log logfile_maxbytes=10MB logfile_backups=2 environment=A=1,B=2
'[program:programname]' -- the section header, required for each program. 'programname' is a descriptive name (arbitrary) used to describe the program being run.
'command' -- the command that will be run when this program is started. The command can be either absolute, e.g. ('/path/to/programname') or relative ('programname'). If it is relative, the PATH will be searched for the executable. Programs can accept arguments, e.g. ('/path/to/program foo bar'). The command line can used double quotes to group arguments with spaces in them to pass to the program, e.g. ('/path/to/program/name -p "foo bar"'). Controlled programs should themselves not be daemons, as supervisord assumes it is responsible for daemonizing its subprocesses.
'priority' -- the relative priority of the program in the start and shutdown ordering. Lower priorities indicate programs that start first and shut down last at startup and when aggregate commands are used in various clients (e.g. "start all"/"stop all"). Higher priorities indicate programs that start last and shut down first. Default: 999.
'autostart' -- If true, this program will start automatically when supervisord is started. Default: true.
'autorestart' -- If true, when the program exits "unexpectedly", supervisor will restart it automatically. "unexpected" exits are those which happen when the program exits with an "unexpected" exit code (see 'exitcodes'). Default: true.
'startsecs' -- The total number of seconds which the program needs to stay running after a startup to consider the start successful. If the program does not stay up for this many seconds after it is started, even if it exits with an "expected" exit code, the startup will be considered a failure. Set to 0 to indicate that the program needn't stay running for any particular amount of time. Default: 1
'startretries' -- The number of serial failure attempts that supervisord will allow when attempting to start the program before giving up and puting the process into an ERROR state. Default: 3.
'exitcodes' -- The list of 'expected' exit codes for this program. A program is considered 'failed' (and will be restarted, if autorestart is set true) if it exits with an exit code which is not in this list and a stop of the program has not been explicitly requested. Default: 0,2.
'stopsignal' -- The signal used to kill the program when a stop is requested. This can be any of TERM, HUP, INT, QUIT, KILL, USR1, or USR2. Default: TERM.
'stopwaitsecs' -- The number of seconds to wait for the program to return a SIGCHILD to supervisord after the program has been sent a stopsignal. If this number of seconds elapses before supervisord receives a SIGCHILD from the process, supervisord will attempt to kill it with a final SIGKILL. Default: 10.
'user' -- If supervisord is running as root, this UNIX user account will be used as the account which runs the program. If supervisord is not running as root, this option has no effect. Defaut: do not switch users.
'log_stdout' -- Send process stdout output to the process logfile. Default: true.
'log_stderr' -- Send process stderr output to the process logfile. Default: false.
'logfile' -- Keep process output as determined by log_stdout and log_stderr in this file. NOTE: if both log_stderr and log_stdout are true, chunks of output from the process' stderr and stdout will be intermingled more or less randomly in the log. If 'logfile' is unset or set to 'AUTO', supervisor will automatically choose a file location. If this is set to 'NONE', supervisord will create no log file. AUTO log files and their backups will be deleted when supervisord restarts. Default: AUTO.
'logfile_maxbytes' -- The maximum number of bytes that may be consumed by the process log file before it is rotated (suffix multipliers like "KB", "MB", and "GB" can be used in the value). Set this value to 0 to indicate an unlimited log size. Default: 50MB.
'logfile_backups' -- The number of backups to keep around resulting from process log file rotation. Set this to 0 to indicate an unlimited number of backups. Default: 10.
'environment' -- A list of key/value pairs in the form "KEY=val,KEY2=val2" that will be placed in the child process' environment. Default: none.
Nondaemonizing of Subprocesses
Programs run under supervisor should not daemonize themselves. Instead, they should run in the foreground and not detach from the "terminal" that starts them. The easiest way to tell if a command will run in the foreground is to run the command from a shell prompt. If it gives you control of the terminal back, it's daemonizing itself and that will be the wrong way to run it under supervisor. You want to run a command that essentially requires you to press Ctrl-C to get control of the terminal back. If it gives you a shell prompt back after running it without needing to press Ctrl-C, it's not useful under supervisor. All programs have options to be run in the foreground but there's no standard way to do it; you'll need to read the documentation for each program you want to do this with.
Examples of Program Configurations
Apache 2.0.54:[program:apache] command=/usr/sbin/httpd -DNO_DETACH
Postgres 8.14:[program:postgres] command=/path/to/postmaster ; we use the "fast" shutdown signal SIGINT stopsignal=INT log_stderr=true
Zope 2.8 instances and ZEO:[program:zeo] command=/path/to/runzeo priority=1 [program:zope1] command=/path/to/instance/home/bin/runzope priority=2 log_stderr=true [program:zope2] command=/path/to/another/instance/home/bin/runzope priority=2 log_stderr=true
OpenLDAP slapd:[program:slapd] command=/path/to/slapd -f /path/to/slapd.conf -h ldap://0.0.0.0:8888
A process controlled by supervisord will be in one of the below states at any given time. You may see these state names in various user interface elements.
- STOPPED (0) -- The process has been stopped due to a stop request or
- has never been started.
STARTING (10) -- The process is starting due to a start request.
RUNNING (20) -- The process is running.
BACKOFF (30) -- The process is waiting to restart after a nonfatal error.
STOPPING (40) -- The process is stopping due to a stop request.
EXITED (100) -- The process exited with an expected exit code.
FATAL (200) -- The process could not be started successfully.
UNKNOWN (1000) -- The process is in an unknown state (programming error).
Process progress through these states as per the following directed graph:--> STOPPED / | | | | | STOPPING | ^ V | STARTING <-----> BACKOFF | / ^ | | V | | \-- RUNNING / \ | | / \ V V / \ ----- FATAL EXITED
A process is in the STOPPED state if it has been stopped adminstratively or if it has never been started.
When an autorestarting process is in the BACKOFF state, it will be automatically restarted by supervisord. It will switch between STARTING and BACKOFF states until it becomes evident that it cannot be started because the number of startretries has exceeded the maximum, at which point it will transition to the FATAL state. Each start retry will take progressively more time.
An autorestarted process will never be automtatically restarted if it ends up in the FATAL state (it must be manually restarted from this state).
A process transitions into the STOPPING state via an administrative stop request, and will then end up in the STOPPED state.
A process that cannot be stopped successfully will stay in the STOPPING state forever. This situation should never be reached during normal operations as it implies that the process did not respond to a final SIGKILL, which is "impossible" under UNIX.
State transitions which always require user action to invoke are these:
FATAL -> STARTING
RUNNING -> STOPPING
State transitions which typically, but not always, require user action to invoke are these, with exceptions noted:
- STOPPED -> STARTING (except at supervisord startup if process is
- configured to autostart)
EXITED -> STARTING (except if process is configured to autorestart)
All other state transitions are managed by supervisord automatically.
Killing supervisord with SIGHUP will stop all processes, reload the configuration from the config file, and restart all processes.
Killing supervisord with SIGUSR2 will close and reopen the supervisord activity log and child log files.
The UNIX permissions on the socket effectively control who may send commands to the server. HTTP basic authentication provides access control for internet and UNIX domain sockets as necessary.
I have done my best to assure that use of a supervisord process running as root cannot lead to unintended privilege escalation, but caveat emptor. Particularly, it is not as paranoid as something like DJ Bernstein's "daemontools", inasmuch as "supervisord" allows for arbitrary path specifications in its configuration file to which data may be written. Allowing arbitrary path selections can create vulnerabilities from symlink attacks. Be careful when specifying paths in your configuration. Ensure that supervisord's configuration file cannot be read from or written to by unprivileged users and that all files installed by the supervisor package have "sane" file permission protection settings. Additionally, ensure that your PYTHONPATH is sane and that all Python standard library files have adequate file permission protections. Then, pray to the deity of your choice.
Some examples of shell scripts to start services under supervisor can be found "here":http://www.thedjbway.org/services.html. These examples are actually for daemontools but the premise is the same for supervisor. Another collection of recipes for starting various programs in the foreground is "here":http://smarden.org/runit/runscripts.html .
Some processes (like mysqld) ignore signals sent to the actual process/thread which is created by supervisord. Instead, a "special" thread/process is created by these kinds of programs which is responsible for handling signals. This is problematic, because supervisord can only kill a pid which it creates itself, not any child thread or process of the program it creates. Fortunately, these programs typically write a pidfile which is meant to be read in order to kill the process. As a workaround for this case, a special "pidproxy" program can handle startup of these kinds of processes. The pidproxy program is a small shim that starts a process, and upon the receipt of a signal, sends the signal to the pid provided in a pidfile. A sample supervisord configuration program entry for a pidproxy-enabled program is provided here:[program:mysql] command=/path/to/pidproxy /path/to/pidfile /path/to/mysqld_safe
The pidproxy program is named 'pidproxy.py' and is in the distribution.
My program never starts and supervisor doesn't indicate any error: Make sure the "x" bit is set on the executable file you're using in the command= line.
How can I tell if my program is running under supervisor? Supervisor and its subprocesses share an environment variable "SUPERVISOR_ENABLED". When a process is run under supervisor, your program can check for the presence of this variable to determine whether it is running under supervisor (new in 2.0).
Please report bugs at http://www.plope.com/software/collector .
Chris McDonough (email@example.com) http://www.plope.com