emqx4.3/bin/emqx

#!/bin/bash
# -*- tab-width:4;indent-tabs-mode:nil -*-
# ex: ts=4 sw=4 et

set -e

ROOT_DIR="$(cd "$(dirname "$(readlink "$0" || echo "$0")")"/..; pwd -P)"
# shellcheck disable=SC1090
. "$ROOT_DIR"/releases/emqx_vars

RUNNER_SCRIPT="$RUNNER_BIN_DIR/$REL_NAME"
CODE_LOADING_MODE="${CODE_LOADING_MODE:-embedded}"
REL_DIR="$RUNNER_ROOT_DIR/releases/$REL_VSN"

WHOAMI=$(whoami)

# Make sure log directory exists
mkdir -p "$RUNNER_LOG_DIR"

# Make sure data directory exists
mkdir -p "$RUNNER_DATA_DIR"

# cuttlefish try to read environment variables starting with "EMQX_", if not specified
export CUTTLEFISH_ENV_OVERRIDE_PREFIX="${CUTTLEFISH_ENV_OVERRIDE_PREFIX:-EMQX_}"

relx_usage() {
    command="$1"

    case "$command" in
        unpack)
            echo "Usage: $REL_NAME unpack [VERSION]"
            echo "Unpacks a release package VERSION, it assumes that this"
            echo "release package tarball has already been deployed at one"
            echo "of the following locations:"
            echo "      releases/<relname>-<version>.tar.gz"
            echo "      releases/<relname>-<version>.zip"
            ;;
        install)
            echo "Usage: $REL_NAME install [VERSION]"
            echo "Installs a release package VERSION, it assumes that this"
            echo "release package tarball has already been deployed at one"
            echo "of the following locations:"
            echo "      releases/<relname>-<version>.tar.gz"
            echo "      releases/<relname>-<version>.zip"
            echo ""
            echo "     --no-permanent   Install release package VERSION but"
            echo "                      don't make it permanent"
            ;;
        uninstall)
            echo "Usage: $REL_NAME uninstall [VERSION]"
            echo "Uninstalls a release VERSION, it will only accept"
            echo "versions that are not currently in use"
            ;;
        upgrade)
            echo "Usage: $REL_NAME upgrade [VERSION]"
            echo "Upgrades the currently running release to VERSION, it assumes"
            echo "that a release package tarball has already been deployed at one"
            echo "of the following locations:"
            echo "      releases/<relname>-<version>.tar.gz"
            echo "      releases/<relname>-<version>.zip"
            echo ""
            echo "     --no-permanent   Install release package VERSION but"
            echo "                      don't make it permanent"
            ;;
        downgrade)
            echo "Usage: $REL_NAME downgrade [VERSION]"
            echo "Downgrades the currently running release to VERSION, it assumes"
            echo "that a release package tarball has already been deployed at one"
            echo "of the following locations:"
            echo "      releases/<relname>-<version>.tar.gz"
            echo "      releases/<relname>-<version>.zip"
            echo ""
            echo "     --no-permanent   Install release package VERSION but"
            echo "                      don't make it permanent"
            ;;
        *)
            echo "Usage: $REL_NAME {start|start_boot <file>|ertspath|foreground|stop|restart|reboot|pid|ping|console|console_clean|console_boot <file>|attach|remote_console|upgrade|downgrade|install|uninstall|versions|escript|rpc|rpcterms|eval|root_dir}"
            ;;
    esac
}

# Simple way to check the correct user and fail early
check_user() {
    # Validate that the user running the script is the owner of the
    # RUN_DIR.
    if [ "$RUNNER_USER" ] && [ "x$WHOAMI" != "x$RUNNER_USER" ]; then
        if [ "x$WHOAMI" != "xroot" ]; then
            echo "You need to be root or use sudo to run this command"
            exit 1
        fi
        CMD="\"$RUNNER_SCRIPT\" "
        for ARG in "$@"; do
            CMD="${CMD} \"$ARG\""
        done
        # This will drop priviledges into the runner user
        # It exec's in a new shell and the current shell will exit
        exec su - "$RUNNER_USER" -c "$CMD"
    fi
}


# Make sure the user running this script is the owner and/or su to that user
check_user "$@"
ES=$?
if [ "$ES" -ne 0 ]; then
    exit $ES
fi

if [ -z "$WITH_EPMD" ]; then
    EPMD_ARG="-start_epmd false -epmd_module ekka_epmd -proto_dist ekka"
else
    EPMD_ARG="-start_epmd true $PROTO_DIST_ARG"
fi

# Warn the user if ulimit -n is less than 1024
ULIMIT_F=$(ulimit -n)
if [ "$ULIMIT_F" -lt 1024 ]; then
    echo "!!!!"
    echo "!!!! WARNING: ulimit -n is ${ULIMIT_F}; 1024 is the recommended minimum."
    echo "!!!!"
fi

# Echo to stderr on errors
echoerr() { echo "$@" 1>&2; }

# By default, use cuttlefish to generate app.config and vm.args
CUTTLEFISH="${USE_CUTTLEFISH:-yes}"

SED_REPLACE="sed -i "
case $(sed --help 2>&1) in
    *GNU*) SED_REPLACE="sed -i ";;
    *BusyBox*) SED_REPLACE="sed -i ";;
    *) SED_REPLACE="sed -i '' ";;
esac

# Get node pid
relx_get_pid() {
    if output="$(relx_nodetool rpcterms os getpid)"
    then
        # shellcheck disable=SC2001 # Escaped quote taken as closing quote in editor
        echo "$output" | sed -e 's/"//g'
        return 0
    else
        echo "$output"
        return 1
    fi
}

relx_get_nodename() {
    id="longname$(relx_gen_id)-${NAME}"
    "$BINDIR/erl" -boot "$REL_DIR/start_clean" -eval '[Host] = tl(string:tokens(atom_to_list(node()),"@")), io:format("~s~n", [Host]), halt()' -noshell "${NAME_TYPE}" "$id"
}

# Connect to a remote node
relx_rem_sh() {
    # Generate a unique id used to allow multiple remsh to the same node
    # transparently
    id="remsh$(relx_gen_id)-${NAME}"
    # Get the node's ticktime so that we use the same thing.
    TICKTIME="$(relx_nodetool rpcterms net_kernel get_net_ticktime)"

    # shellcheck disable=SC2086 # $EPMD_ARG is supposed to be split by whitespace
    # Setup remote shell command to control node
    exec "$BINDIR/erl" "$NAME_TYPE" "$id" -remsh "$NAME" -boot "$REL_DIR/start_clean" \
         -boot_var ERTS_LIB_DIR "$ERTS_LIB_DIR" \
         -setcookie "$COOKIE" -hidden -kernel net_ticktime "$TICKTIME" $EPMD_ARG
}

# Generate a random id
relx_gen_id() {
    od -t x -N 4 /dev/urandom | head -n1 | awk '{print $2}'
}

# Control a node
relx_nodetool() {
    command="$1"; shift

    ERL_FLAGS="$ERL_FLAGS $EPMD_ARG" \
    "$ERTS_DIR/bin/escript" "$ROOTDIR/bin/nodetool" "$NAME_TYPE" "$NAME" \
                                -setcookie "$COOKIE" "$command" "$@"
}

# Run an escript in the node's environment
relx_escript() {
    shift; scriptpath="$1"; shift
    export RUNNER_ROOT_DIR

    "$ERTS_DIR/bin/escript" "$ROOTDIR/$scriptpath" "$@"
}

# Output a start command for the last argument of run_erl
relx_start_command() {
    printf "exec \"%s\" \"%s\"" "$RUNNER_SCRIPT" \
           "$START_OPTION"
}

# Function to generate app.config and vm.args
generate_config() {
    ## Delete the *.siz files first or it cann't start after
    ## changing the config 'log.rotation.size'
    rm -rf "${RUNNER_LOG_DIR}"/*.siz

    if [ "$CUTTLEFISH" != "yes" ]; then
        # Note: we have added a parameter '-vm_args' to this. It
        # appears redundant but it is not! the erlang vm allows us to
        # access all arguments to the erl command EXCEPT '-args_file',
        # so in order to get access to this file location from within
        # the vm, we need to pass it in twice.
        CONFIG_ARGS=" -config $RUNNER_ETC_DIR/app.config -args_file $RUNNER_ETC_DIR/vm.args -vm_args $RUNNER_ETC_DIR/vm.args "
    else
        CONFIG_ARGS=$("$ERTS_PATH"/escript "$RUNNER_ROOT_DIR"/bin/cuttlefish -i "$REL_DIR"/emqx.schema -c "$RUNNER_ETC_DIR"/emqx.conf -d "$RUNNER_DATA_DIR"/configs generate)

        ## Merge cuttlefish generated *.args into the vm.args
        CUTTLE_GEN_ARG_FILE=$(echo "$CONFIG_ARGS" | sed -n 's/^.*\(vm_args[[:space:]]\)//p' | awk '{print $1}')
        TMP_ARG_FILE="$RUNNER_DATA_DIR/configs/vm.args.tmp"
        cp "$RUNNER_ETC_DIR/vm.args" "$TMP_ARG_FILE"
        echo "" >> "$TMP_ARG_FILE"
        sed '/^#/d' "$CUTTLE_GEN_ARG_FILE" | sed '/^$/d' | while IFS='' read -r ARG_LINE || [ -n "$ARG_LINE" ]; do
            ARG_KEY=$(echo "$ARG_LINE" | awk '{$NF="";print}')
            ARG_VALUE=$(echo "$ARG_LINE" | awk '{print $NF}')
            TMP_ARG_VALUE=$(grep "^$ARG_KEY" "$TMP_ARG_FILE" | awk '{print $NF}')
            if [ "$ARG_VALUE" != "$TMP_ARG_VALUE" ] ; then
                if [ -n "$TMP_ARG_VALUE" ]; then
                    sh -c "$SED_REPLACE 's/^$ARG_KEY.*$/$ARG_LINE/' $TMP_ARG_FILE"
                else
                    echo "$ARG_LINE" >> "$TMP_ARG_FILE"
                fi
            fi
        done
        mv -f "$TMP_ARG_FILE" "$CUTTLE_GEN_ARG_FILE"
    fi

    # shellcheck disable=SC2086
    if ! relx_nodetool chkconfig $CONFIG_ARGS; then
        echoerr "Error reading $CONFIG_ARGS"
        exit 1
    fi
}

# Call bootstrapd for daemon commands like start/stop/console
bootstrapd() {
    if [ -e "$RUNNER_DATA_DIR/.erlang.cookie" ]; then
        chown "$RUNNER_USER" "$RUNNER_DATA_DIR"/.erlang.cookie
    fi
}

# Use $CWD/etc/sys.config if exists
if [ -z "$RELX_CONFIG_PATH" ]; then
    if [ -f "$RUNNER_ETC_DIR/sys.config" ]; then
        RELX_CONFIG_PATH="-config $RUNNER_ETC_DIR/sys.config"
    else
        RELX_CONFIG_PATH=""
    fi
fi

if [ -z "$NAME_ARG" ]; then
    NODENAME="${EMQX_NODE_NAME:-}"
    # check if there is a node running, inspect its name
    # shellcheck disable=SC2009 # pgrep does not support Extended Regular Expressions
    [ -z "$NODENAME" ] && NODENAME=$(ps -ef | grep -E '\-progname\s.*emqx\s' | grep -o -E '\-name (\S*)' | awk '{print $2}')
    [ -z "$NODENAME" ] && NODENAME=$(grep -E '^[ \t]*node.name[ \t]*=[ \t]*' "$RUNNER_ETC_DIR/emqx.conf" 2> /dev/null | tail -1 | cut -d = -f 2-)
    if [ -z "$NODENAME" ]; then
        echoerr "vm.args needs to have a -name parameter."
        echoerr "  -sname is not supported."
        echoerr "perhaps you do not have read permissions on $RUNNER_ETC_DIR/emqx.conf"
        exit 1
    else
        NAME_ARG="-name ${NODENAME# *}"
    fi
fi

# Extract the name type and name from the NAME_ARG for REMSH
NAME_TYPE="$(echo "$NAME_ARG" | awk '{print $1}')"
NAME="$(echo "$NAME_ARG" | awk '{print $2}')"
NODENAME="$(echo "$NAME" | awk -F'@' '{print $1}')"
export ESCRIPT_NAME="$NODENAME"

PIPE_DIR="${PIPE_DIR:-/$RUNNER_DATA_DIR/${WHOAMI}_erl_pipes/$NAME/}"

# Extract the target cookie
if [ -z "$COOKIE_ARG" ]; then
    COOKIE="${EMQX_NODE_COOKIE:-}"
    # check if there is a node running, steal its cookie
    # shellcheck disable=SC2009 # pgrep does not support Extended Regular Expressions
    [ -z "$COOKIE" ] && COOKIE=$(ps -ef | grep -E '\-progname\s.*emqx\s' | grep -o -E '\-setcookie (\S*)' | awk '{print $2}')
    [ -z "$COOKIE" ] && COOKIE=$(grep -E '^[ \t]*node.cookie[ \t]*=[ \t]*' "$RUNNER_ETC_DIR/emqx.conf" 2> /dev/null | tail -1 | cut -d = -f 2-)
    if [ -z "$COOKIE" ]; then
        echoerr "vm.args needs to have a -setcookie parameter."
        echoerr "please check $RUNNER_ETC_DIR/emqx.conf"
        exit 1
    else
        COOKIE_ARG="-setcookie $COOKIE"
    fi
fi

# Extract cookie name from COOKIE_ARG
COOKIE="$(echo "$COOKIE_ARG" | awk '{print $2}')"

# Support for IPv6 Dist. See: https://github.com/emqtt/emqttd/issues/1460
PROTO_DIST=$(grep -E '^[ \t]*cluster.proto_dist[ \t]*=[ \t]*' "$RUNNER_ETC_DIR/emqx.conf" 2> /dev/null | tail -1 | cut -d = -f 2-)
if [ -z "$PROTO_DIST" ]; then
    PROTO_DIST_ARG=""
else
    PROTO_DIST_ARG="-proto_dist $PROTO_DIST"
fi

export ROOTDIR="$RUNNER_ROOT_DIR"
export ERTS_DIR="$ROOTDIR/erts-$ERTS_VSN"
export BINDIR="$ERTS_DIR/bin"
export EMU="beam"
export PROGNAME="erl"
export LD_LIBRARY_PATH="$ERTS_DIR/lib:$LD_LIBRARY_PATH"
ERTS_LIB_DIR="$ERTS_DIR/../lib"
MNESIA_DATA_DIR="$RUNNER_DATA_DIR/mnesia/$NAME"

cd "$ROOTDIR"

# User can specify an sname without @hostname
# This will fail when creating remote shell
# So here we check for @ and add @hostname if missing
case $NAME in
    *@*)
        # Nothing to do
        ;;
    *)
        NAME=$NAME@$(relx_get_nodename)
        ;;
esac

# Check the first argument for instructions
case "$1" in
    start|start_boot)
        # Make sure a node IS not running
        if relx_nodetool "ping" >/dev/null 2>&1; then
            echo "Node is already running!"
            exit 1
        fi
        # Bootstrap daemon command (check perms & drop to $RUNNER_USER)
        bootstrapd

        # Save this for later.
        CMD=$1
        case "$1" in
            start)
                shift
                START_OPTION="console"
                HEART_OPTION="start"
                ;;
            start_boot)
                shift
                START_OPTION="console_boot"
                HEART_OPTION="start_boot"
                ;;
        esac
        RUN_PARAM="$*"

        # Set arguments for the heart command
        set -- "$RUNNER_SCRIPT" "$HEART_OPTION"
        [ "$RUN_PARAM" ] && set -- "$@" "$RUN_PARAM"

        # Export the HEART_COMMAND
        HEART_COMMAND="$RUNNER_SCRIPT $CMD"
        export HEART_COMMAND

        ## See: http://erlang.org/doc/man/run_erl.html
        # Export the RUN_ERL_LOG_GENERATIONS
        export RUN_ERL_LOG_GENERATIONS=${RUN_ERL_LOG_GENERATIONS:-"5"}

        # Export the RUN_ERL_LOG_MAXSIZE
        export RUN_ERL_LOG_MAXSIZE=${RUN_ERL_LOG_MAXSIZE:-"10485760"}

        mkdir -p "$PIPE_DIR"

        "$BINDIR/run_erl" -daemon "$PIPE_DIR" "$RUNNER_LOG_DIR" \
                          "$(relx_start_command)"

        WAIT_TIME=${WAIT_FOR_ERLANG:-15}
        while [ "$WAIT_TIME" -gt 0 ]; do
            if ! relx_nodetool "ping" >/dev/null 2>&1; then
                WAIT_TIME=$((WAIT_TIME - 1))
                sleep 1
                continue
            fi
            sleep 1
            if relx_nodetool "ping" >/dev/null 2>&1; then
                echo "$EMQX_DISCR $REL_VSN is started successfully!"
                exit 0
            fi
        done && echo "$EMQX_DISCR $REL_VSN failed to start within ${WAIT_FOR_ERLANG:-15} seconds,"
        echo "see the output of '$0 console' for more information."
        echo "If you want to wait longer, set the environment variable"
        echo "WAIT_FOR_ERLANG to the number of seconds to wait."
        exit 1
        ;;

    stop)
        # Wait for the node to completely stop...
        PID="$(relx_get_pid)"
        if ! relx_nodetool "stop"; then
            exit 1
        fi
        while kill -s 0 "$PID" 2>/dev/null; do
            sleep 1
        done
        ;;

    restart|reboot)
        echo "$EMQX_DISCR $REL_VSN is stopped: $("$RUNNER_BIN_DIR"/emqx stop)"
        "$RUNNER_BIN_DIR"/emqx start
        ;;

    pid)
        ## Get the VM's pid
        if ! relx_get_pid; then
            exit 1
        fi
        ;;

    ping)
        ## See if the VM is alive
        if ! relx_nodetool "ping"; then
            exit 1
        fi
        ;;

    escript)
        ## Run an escript under the node's environment
        if ! relx_escript "$@"; then
            exit 1
        fi
        ;;

    attach)
        # Make sure a node IS running
        if ! relx_nodetool "ping" > /dev/null; then
            echo "Node is not running!"
            exit 1
        fi

        # Bootstrap daemon command (check perms & drop to $RUNNER_USER)
        bootstrapd

        shift
        exec "$BINDIR/to_erl" "$PIPE_DIR"
        ;;

    remote_console)
        # Make sure a node IS running
        if ! relx_nodetool "ping" > /dev/null; then
            echo "Node is not running!"
            exit 1
        fi

        # Bootstrap daemon command (check perms & drop to $RUNNER_USER)
        bootstrapd

        shift
        relx_rem_sh
        ;;

    upgrade|downgrade|install|unpack|uninstall)
        if [ -z "$2" ]; then
            echo "Missing version argument"
            echo "Usage: $REL_NAME $1 {version}"
            exit 1
        fi

        COMMAND="$1"; shift

        # Make sure a node IS running
        if ! relx_nodetool "ping" > /dev/null; then
            echo "Node is not running!"
            exit 1
        fi

        ERL_FLAGS="$ERL_FLAGS $EPMD_ARG" \
        exec "$BINDIR/escript" "$ROOTDIR/bin/install_upgrade.escript" \
             "$COMMAND" "{'$REL_NAME', \"$NAME_TYPE\", '$NAME', '$COOKIE'}" "$@"
        ;;

    versions)
        # Make sure a node IS running
        if ! relx_nodetool "ping" > /dev/null; then
            echo "Node is not running!"
            exit 1
        fi

        COMMAND="$1"; shift

        ERL_FLAGS="$ERL_FLAGS $EPMD_ARG" \
        exec "$BINDIR/escript" "$ROOTDIR/bin/install_upgrade.escript" \
             "versions" "{'$REL_NAME', \"$NAME_TYPE\", '$NAME', '$COOKIE'}" "$@"
        ;;

    console|console_clean|console_boot)
        # Bootstrap daemon command (check perms & drop to $RUNNER_USER)
        bootstrapd

        # .boot file typically just $REL_NAME (ie, the app name)
        # however, for debugging, sometimes start_clean.boot is useful.
        # For e.g. 'setup', one may even want to name another boot script.
        case "$1" in
            console)
                if [ -f "$REL_DIR/$REL_NAME.boot" ]; then
                  BOOTFILE="$REL_DIR/$REL_NAME"
                else
                  BOOTFILE="$REL_DIR/start"
                fi
                ;;
            console_clean)
                BOOTFILE="$REL_DIR/start_clean"
                ;;
            console_boot)
                shift
                BOOTFILE="$1"
                shift
                ;;
        esac

        #generate app.config and vm.args
        generate_config

        # Setup beam-required vars
        EMU="beam"
        PROGNAME="${0#*/}"

        export EMU
        export PROGNAME

        # Store passed arguments since they will be erased by `set`
        ARGS="$*"

        # shellcheck disable=SC2086 # $RELX_CONFIG_PATH $CONFIG_ARGS $EPMD_ARG are supposed to be split by whitespace
        # Build an array of arguments to pass to exec later on
        # Build it here because this command will be used for logging.
        set -- "$BINDIR/erlexec" -boot "$BOOTFILE" -mode "$CODE_LOADING_MODE" \
            -boot_var ERTS_LIB_DIR "$ERTS_LIB_DIR" \
            -mnesia dir "\"${MNESIA_DATA_DIR}\"" \
            $RELX_CONFIG_PATH $CONFIG_ARGS $EPMD_ARG

        # Dump environment info for logging purposes
        echo "Exec: $*" -- ${1+$ARGS}
        echo "Root: $ROOTDIR"

        # Log the startup
        echo "$RUNNER_ROOT_DIR"
        logger -t "${REL_NAME[$$]}" "Starting up"

        # Start the VM
        exec "$@" -- ${1+$ARGS}
        ;;

    foreground)
        # Bootstrap daemon command (check perms & drop to $RUNNER_USER)
        bootstrapd
        # start up the release in the foreground for use by runit
        # or other supervision services

        #generate app.config and vm.args
        generate_config

        [ -f "$REL_DIR/$REL_NAME.boot" ] && BOOTFILE="$REL_NAME" || BOOTFILE=start
        FOREGROUNDOPTIONS="-noshell -noinput +Bd"

        # Setup beam-required vars
        EMU=beam
        PROGNAME="${0#*/}"

        export EMU
        export PROGNAME

        # Store passed arguments since they will be erased by `set`
        ARGS="$*"

        # shellcheck disable=SC2086 # $RELX_CONFIG_PATH $CONFIG_ARGS $EPMD_ARG are supposed to be split by whitespace
        # Build an array of arguments to pass to exec later on
        # Build it here because this command will be used for logging.
        set -- "$BINDIR/erlexec" $FOREGROUNDOPTIONS \
            -boot "$REL_DIR/$BOOTFILE" -mode "$CODE_LOADING_MODE" \
            -boot_var ERTS_LIB_DIR "$ERTS_LIB_DIR" \
            -mnesia dir "\"${MNESIA_DATA_DIR}\"" \
            $RELX_CONFIG_PATH $CONFIG_ARGS $EPMD_ARG

        # Dump environment info for logging purposes
        echo "Exec: $*" -- ${1+$ARGS}
        echo "Root: $ROOTDIR"

        # Start the VM
        exec "$@" -- ${1+$ARGS}
        ;;
    ertspath)
        echo "$ERTS_PATH"
        ;;
    rpc)
        # Make sure a node IS running
        if ! relx_nodetool "ping" > /dev/null; then
            echo "Node is not running!"
            exit 1
        fi

        shift

        relx_nodetool rpc "$@"
        ;;
    rpcterms)
        # Make sure a node IS running
        if ! relx_nodetool "ping" > /dev/null; then
            echo "Node is not running!"
            exit 1
        fi

        shift

        relx_nodetool rpcterms "$@"
        ;;
    root_dir)
        # Make sure a node IS running
        if ! relx_nodetool "ping" > /dev/null; then
            echo "Node is not running!"
            exit 1
        fi

        shift
        relx_nodetool "eval" 'code:root_dir()'
        ;;
    eval)
        # Make sure a node IS running
        if ! relx_nodetool "ping" > /dev/null; then
            echo "Node is not running!"
            exit 1
        fi

        shift
        relx_nodetool "eval" "$@"
        ;;
    *)
        relx_usage "$1"
        exit 1
        ;;
esac

exit 0