|
|
@@ -4,11 +4,26 @@
|
|
|
|
|
|
set -euo pipefail
|
|
|
|
|
|
-BASE="$(cd "$(dirname "$(readlink "$0" || echo "$0")")"/..; pwd -P)"
|
|
|
-# shellcheck disable=SC1090,SC1091
|
|
|
-source "$BASE/bin/common_defs.sh"
|
|
|
-# shellcheck disable=SC1090,SC1091
|
|
|
-source "$BASE/bin/common_functions.sh"
|
|
|
+DEBUG="${DEBUG:-0}"
|
|
|
+if [ "$DEBUG" -eq 1 ]; then
|
|
|
+ set -x
|
|
|
+fi
|
|
|
+
|
|
|
+ROOT_DIR="$(cd "$(dirname "$(readlink "$0" || echo "$0")")"/..; pwd -P)"
|
|
|
+# shellcheck disable=SC1090
|
|
|
+. "$ROOT_DIR"/releases/emqx_vars
|
|
|
+
|
|
|
+# defined in emqx_vars
|
|
|
+export RUNNER_ROOT_DIR
|
|
|
+export RUNNER_ETC_DIR
|
|
|
+export REL_VSN
|
|
|
+
|
|
|
+RUNNER_SCRIPT="$RUNNER_BIN_DIR/$REL_NAME"
|
|
|
+CODE_LOADING_MODE="${CODE_LOADING_MODE:-embedded}"
|
|
|
+REL_DIR="$RUNNER_ROOT_DIR/releases/$REL_VSN"
|
|
|
+SCHEMA_MOD=emqx_conf_schema
|
|
|
+
|
|
|
+WHOAMI=$(whoami)
|
|
|
|
|
|
# Make sure log directory exists
|
|
|
mkdir -p "$RUNNER_LOG_DIR"
|
|
|
@@ -17,8 +32,49 @@ mkdir -p "$RUNNER_LOG_DIR"
|
|
|
mkdir -p "$RUNNER_DATA_DIR"
|
|
|
|
|
|
# Make sure data/configs exists
|
|
|
+CONFIGS_DIR="$RUNNER_DATA_DIR/configs"
|
|
|
mkdir -p "$CONFIGS_DIR"
|
|
|
|
|
|
+# hocon try to read environment variables starting with "EMQX_"
|
|
|
+export HOCON_ENV_OVERRIDE_PREFIX='EMQX_'
|
|
|
+
|
|
|
+export ROOTDIR="$RUNNER_ROOT_DIR"
|
|
|
+export ERTS_DIR="$ROOTDIR/erts-$ERTS_VSN"
|
|
|
+export BINDIR="$ERTS_DIR/bin"
|
|
|
+export EMU="beam"
|
|
|
+export PROGNAME="erl"
|
|
|
+export ERTS_LIB_DIR="$ERTS_DIR/../lib"
|
|
|
+DYNLIBS_DIR="$RUNNER_ROOT_DIR/dynlibs"
|
|
|
+
|
|
|
+# Echo to stderr on errors
|
|
|
+echoerr() { echo "ERROR: $*" 1>&2; }
|
|
|
+
|
|
|
+die() {
|
|
|
+ echoerr "ERROR: $1"
|
|
|
+ errno=${2:-1}
|
|
|
+ exit "$errno"
|
|
|
+}
|
|
|
+
|
|
|
+assert_node_alive() {
|
|
|
+ if ! relx_nodetool "ping" > /dev/null; then
|
|
|
+ die "node_is_not_running!" 1
|
|
|
+ fi
|
|
|
+}
|
|
|
+
|
|
|
+
|
|
|
+# Echo to stderr on errors
|
|
|
+echoerr() { echo "$*" 1>&2; }
|
|
|
+
|
|
|
+check_erlang_start() {
|
|
|
+ # RELEASE_LIB is used by Elixir
|
|
|
+ "$BINDIR/$PROGNAME" \
|
|
|
+ -noshell \
|
|
|
+ -boot_var RELEASE_LIB "$ERTS_LIB_DIR/lib" \
|
|
|
+ -boot "$REL_DIR/start_clean" \
|
|
|
+ -s crypto start \
|
|
|
+ -s erlang halt
|
|
|
+}
|
|
|
+
|
|
|
usage() {
|
|
|
local command="$1"
|
|
|
|
|
|
@@ -69,14 +125,14 @@ usage() {
|
|
|
echo "Print path to Erlang runtime dir"
|
|
|
;;
|
|
|
rpc)
|
|
|
- echo "Usage $REL_NAME rpc MODULE FUNCTION [ARGS, ...]"
|
|
|
+ echo "Usge $REL_NAME rpc MODULE FUNCTION [ARGS, ...]"
|
|
|
echo "Connect to the $EMQX_DESCRIPTION node and make an Erlang RPC"
|
|
|
echo "The result of the RPC call must be 'ok'"
|
|
|
echo "This command blocks for at most 60 seconds in case the node"
|
|
|
echo "does not reply the call in time"
|
|
|
;;
|
|
|
rpcterms)
|
|
|
- echo "Usage $REL_NAME rpcterms MODULE FUNCTION [ARGS, ...]"
|
|
|
+ echo "Usge $REL_NAME rpcterms MODULE FUNCTION [ARGS, ...]"
|
|
|
echo "Connect to the $EMQX_DESCRIPTION node and make an Erlang RPC"
|
|
|
echo "The result of the RPC call is pretty-printed as an Erlang term"
|
|
|
;;
|
|
|
@@ -166,6 +222,30 @@ if ! check_erlang_start >/dev/null 2>&1; then
|
|
|
echoerr "WARNING: There seem to be missing dynamic libs from the OS. Using libs from ${DYNLIBS_DIR}"
|
|
|
fi
|
|
|
|
|
|
+## backward compatible
|
|
|
+if [ -d "$ERTS_DIR/lib" ]; then
|
|
|
+ export LD_LIBRARY_PATH="$ERTS_DIR/lib:$LD_LIBRARY_PATH"
|
|
|
+fi
|
|
|
+
|
|
|
+# 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="DEBUG=$DEBUG \"$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=$?
|
|
|
@@ -173,9 +253,224 @@ if [ "$ES" -ne 0 ]; then
|
|
|
exit $ES
|
|
|
fi
|
|
|
|
|
|
-## IS_BOOT_COMMAND is set to be later used by `common_defs2.sh` to
|
|
|
-## inspect node name and cookie from hocon config (or env variable),
|
|
|
-## which also resides in `common_defs2.sh`.
|
|
|
+# EPMD_ARG="-start_epmd true $PROTO_DIST_ARG"
|
|
|
+NO_EPMD="-start_epmd false -epmd_module ekka_epmd -proto_dist ekka"
|
|
|
+EPMD_ARG="${EPMD_ARG:-${NO_EPMD}}"
|
|
|
+
|
|
|
+# Warn the user if ulimit -n is less than 1024
|
|
|
+ULIMIT_F=$(ulimit -n)
|
|
|
+if [ "$ULIMIT_F" -lt 1024 ]; then
|
|
|
+ echo "!!!!"
|
|
|
+ echo "!!!! WARNING: ulimit -n is ${ULIMIT_F}; 1024 is the recommended minimum."
|
|
|
+ echo "!!!!"
|
|
|
+fi
|
|
|
+
|
|
|
+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
|
|
|
+}
|
|
|
+
|
|
|
+# 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 # $EPMD_ARG is supposed to be split by whitespace
|
|
|
+ # Setup remote shell command to control node
|
|
|
+ if [ "$IS_ELIXIR" = "yes" ]
|
|
|
+ then
|
|
|
+ exec "$REL_DIR/iex" \
|
|
|
+ --remsh "$NAME" \
|
|
|
+ --boot-var RELEASE_LIB "$ERTS_LIB_DIR" \
|
|
|
+ --cookie "$COOKIE" \
|
|
|
+ --hidden \
|
|
|
+ --erl "-kernel net_ticktime $TICKTIME" \
|
|
|
+ --erl "$EPMD_ARG" \
|
|
|
+ --erl "$NAME_TYPE $id" \
|
|
|
+ --boot "$REL_DIR/start_clean"
|
|
|
+ else
|
|
|
+ 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
|
|
|
+ fi
|
|
|
+}
|
|
|
+
|
|
|
+# 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" "$command" "$@"
|
|
|
+}
|
|
|
+
|
|
|
+call_hocon() {
|
|
|
+ "$ERTS_DIR/bin/escript" "$ROOTDIR/bin/nodetool" hocon "$@" \
|
|
|
+ || die "call_hocon_failed: $*" $?
|
|
|
+}
|
|
|
+
|
|
|
+# Run an escript in the node's environment
|
|
|
+relx_escript() {
|
|
|
+ shift; scriptpath="$1"; shift
|
|
|
+ "$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() {
|
|
|
+ local name_type="$1"
|
|
|
+ local node_name="$2"
|
|
|
+ ## Delete the *.siz files first or it cann't start after
|
|
|
+ ## changing the config 'log.rotation.size'
|
|
|
+ rm -rf "${RUNNER_LOG_DIR}"/*.siz
|
|
|
+
|
|
|
+ EMQX_LICENSE_CONF_OPTION=""
|
|
|
+ if [ "${EMQX_LICENSE_CONF:-}" != "" ]; then
|
|
|
+ EMQX_LICENSE_CONF_OPTION="-c ${EMQX_LICENSE_CONF}"
|
|
|
+ fi
|
|
|
+
|
|
|
+ ## timestamp for each generation
|
|
|
+ local NOW_TIME
|
|
|
+ NOW_TIME="$(call_hocon now_time)"
|
|
|
+
|
|
|
+ ## ths command populates two files: app.<time>.config and vm.<time>.args
|
|
|
+ ## NOTE: the generate command merges environment variables to the base config (emqx.conf),
|
|
|
+ ## but does not include the cluster-override.conf and local-override.conf
|
|
|
+ ## meaning, certain overrides will not be mapped to app.<time>.config file
|
|
|
+ ## disable SC2086 to allow EMQX_LICENSE_CONF_OPTION to split
|
|
|
+ # shellcheck disable=SC2086
|
|
|
+ call_hocon -v -t "$NOW_TIME" -I "$CONFIGS_DIR/" -s $SCHEMA_MOD -c "$RUNNER_ETC_DIR"/emqx.conf $EMQX_LICENSE_CONF_OPTION -d "$RUNNER_DATA_DIR"/configs generate
|
|
|
+
|
|
|
+ ## filenames are per-hocon convention
|
|
|
+ local CONF_FILE="$CONFIGS_DIR/app.$NOW_TIME.config"
|
|
|
+ local HOCON_GEN_ARG_FILE="$CONFIGS_DIR/vm.$NOW_TIME.args"
|
|
|
+
|
|
|
+ # This is needed by the Elixir scripts.
|
|
|
+ # Do NOT append `.config`.
|
|
|
+ RELEASE_SYS_CONFIG="$CONFIGS_DIR/app.$NOW_TIME"
|
|
|
+ export RELEASE_SYS_CONFIG
|
|
|
+
|
|
|
+ CONFIG_ARGS="-config $CONF_FILE -args_file $HOCON_GEN_ARG_FILE"
|
|
|
+
|
|
|
+ ## Merge hocon generated *.args into the vm.args
|
|
|
+ TMP_ARG_FILE="$CONFIGS_DIR/vm.args.tmp"
|
|
|
+ cp "$RUNNER_ETC_DIR/vm.args" "$TMP_ARG_FILE"
|
|
|
+ echo "" >> "$TMP_ARG_FILE"
|
|
|
+ echo "-pa ${REL_DIR}/consolidated" >> "$TMP_ARG_FILE"
|
|
|
+ ## read lines from generated vm.<time>.args file
|
|
|
+ ## drop comment lines, and empty lines using sed
|
|
|
+ ## pipe the lines to a while loop
|
|
|
+ sed '/^#/d' "$HOCON_GEN_ARG_FILE" | sed '/^$/d' | while IFS='' read -r ARG_LINE || [ -n "$ARG_LINE" ]; do
|
|
|
+ ## in the loop, split the 'key[:space:]value' pair
|
|
|
+ ARG_KEY=$(echo "$ARG_LINE" | awk '{$NF="";print}')
|
|
|
+ ARG_VALUE=$(echo "$ARG_LINE" | awk '{print $NF}')
|
|
|
+ ## use the key to look up in vm.args file for the value
|
|
|
+ TMP_ARG_VALUE=$(grep "^$ARG_KEY" "$TMP_ARG_FILE" || true | awk '{print $NF}')
|
|
|
+ ## compare generated (to override) value to original (to be overriden) value
|
|
|
+ if [ "$ARG_VALUE" != "$TMP_ARG_VALUE" ] ; then
|
|
|
+ ## if they are different
|
|
|
+ if [ -n "$TMP_ARG_VALUE" ]; then
|
|
|
+ ## if the old value is present, replace it with generated value
|
|
|
+ sh -c "$SED_REPLACE 's|^$ARG_KEY.*$|$ARG_LINE|' $TMP_ARG_FILE"
|
|
|
+ else
|
|
|
+ ## otherwise append generated value to the end
|
|
|
+ echo "$ARG_LINE" >> "$TMP_ARG_FILE"
|
|
|
+ fi
|
|
|
+ fi
|
|
|
+ done
|
|
|
+ echo "$name_type $node_name" >> "$TMP_ARG_FILE"
|
|
|
+ ## rename the generated vm.<time>.args file
|
|
|
+ mv -f "$TMP_ARG_FILE" "$HOCON_GEN_ARG_FILE"
|
|
|
+
|
|
|
+ # shellcheck disable=SC2086
|
|
|
+ if ! relx_nodetool chkconfig $CONFIG_ARGS; then
|
|
|
+ die "failed_to_check_config $CONFIG_ARGS"
|
|
|
+ fi
|
|
|
+}
|
|
|
+
|
|
|
+# check if a PID is down
|
|
|
+is_down() {
|
|
|
+ PID="$1"
|
|
|
+ if ps -p "$PID" >/dev/null; then
|
|
|
+ # still around
|
|
|
+ # shellcheck disable=SC2009 # this grep pattern is not a part of the progra names
|
|
|
+ if ps -p "$PID" | grep -q 'defunct'; then
|
|
|
+ # zombie state, print parent pid
|
|
|
+ parent="$(ps -o ppid= -p "$PID" | tr -d ' ')"
|
|
|
+ echo "WARN: $PID is marked <defunct>, parent:"
|
|
|
+ ps -p "$parent"
|
|
|
+ return 0
|
|
|
+ fi
|
|
|
+ return 1
|
|
|
+ fi
|
|
|
+ # it's gone
|
|
|
+ return 0
|
|
|
+}
|
|
|
+
|
|
|
+wait_for() {
|
|
|
+ local WAIT_TIME
|
|
|
+ local CMD
|
|
|
+ WAIT_TIME="$1"
|
|
|
+ shift
|
|
|
+ CMD="$*"
|
|
|
+ while true; do
|
|
|
+ if $CMD >/dev/null 2>&1; then
|
|
|
+ return 0
|
|
|
+ fi
|
|
|
+ if [ "$WAIT_TIME" -le 0 ]; then
|
|
|
+ return 1
|
|
|
+ fi
|
|
|
+ WAIT_TIME=$((WAIT_TIME - 1))
|
|
|
+ sleep 1
|
|
|
+ done
|
|
|
+}
|
|
|
+
|
|
|
+latest_vm_args() {
|
|
|
+ local hint_var_name="$1"
|
|
|
+ local vm_args_file
|
|
|
+ vm_args_file="$(find "$CONFIGS_DIR" -type f -name "vm.*.args" | sort | tail -1)"
|
|
|
+ if [ -f "$vm_args_file" ]; then
|
|
|
+ echo "$vm_args_file"
|
|
|
+ else
|
|
|
+ echoerr "ERRRO: node not initialized?"
|
|
|
+ echoerr "Generated config file vm.*.args is not found for command '$COMMAND'"
|
|
|
+ echoerr "in config dir: $CONFIGS_DIR"
|
|
|
+ echoerr "In case the file has been deleted while the node is running,"
|
|
|
+ echoerr "set environment variable '$hint_var_name' to continue"
|
|
|
+ exit 1
|
|
|
+ fi
|
|
|
+}
|
|
|
+
|
|
|
+## IS_BOOT_COMMAND is set for later to inspect node name and cookie from hocon config (or env variable)
|
|
|
case "${COMMAND}" in
|
|
|
start|console|console_clean|foreground)
|
|
|
IS_BOOT_COMMAND='yes'
|
|
|
@@ -184,10 +479,62 @@ case "${COMMAND}" in
|
|
|
IS_BOOT_COMMAND='no'
|
|
|
;;
|
|
|
esac
|
|
|
-export IS_BOOT_COMMAND
|
|
|
|
|
|
-# shellcheck disable=SC1090,SC1091
|
|
|
-source "$BASE/bin/common_defs2.sh"
|
|
|
+## make EMQX_NODE_COOKIE right
|
|
|
+if [ -n "${EMQX_NODE_NAME:-}" ]; then
|
|
|
+ export EMQX_NODE__NAME="${EMQX_NODE_NAME}"
|
|
|
+ unset EMQX_NODE_NAME
|
|
|
+fi
|
|
|
+## Possible ways to configure emqx node name:
|
|
|
+## 1. configure node.name in emqx.conf
|
|
|
+## 2. override with environment variable EMQX_NODE__NAME
|
|
|
+## Node name is either short-name (without '@'), e.g. 'emqx'
|
|
|
+## or long name (with '@') e.g. 'emqx@example.net' or 'emqx@127.0.0.1'
|
|
|
+NAME="${EMQX_NODE__NAME:-}"
|
|
|
+if [ -z "$NAME" ]; then
|
|
|
+ if [ "$IS_BOOT_COMMAND" = 'yes' ]; then
|
|
|
+ # for boot commands, inspect emqx.conf for node name
|
|
|
+ NAME="$(call_hocon -s $SCHEMA_MOD -I "$CONFIGS_DIR/" -c "$RUNNER_ETC_DIR"/emqx.conf get node.name | tr -d \")"
|
|
|
+ else
|
|
|
+ vm_args_file="$(latest_vm_args 'EMQX_NODE__NAME')"
|
|
|
+ NAME="$(grep -E '^-s?name' "${vm_args_file}" | awk '{print $2}')"
|
|
|
+ fi
|
|
|
+fi
|
|
|
+
|
|
|
+# force to use 'emqx' short name
|
|
|
+[ -z "$NAME" ] && NAME='emqx'
|
|
|
+MNESIA_DATA_DIR="$RUNNER_DATA_DIR/mnesia/$NAME"
|
|
|
+
|
|
|
+case "$NAME" in
|
|
|
+ *@*)
|
|
|
+ NAME_TYPE='-name'
|
|
|
+ ;;
|
|
|
+ *)
|
|
|
+ NAME_TYPE='-sname'
|
|
|
+esac
|
|
|
+SHORT_NAME="$(echo "$NAME" | awk -F'@' '{print $1}')"
|
|
|
+export ESCRIPT_NAME="$SHORT_NAME"
|
|
|
+
|
|
|
+PIPE_DIR="${PIPE_DIR:-/$RUNNER_DATA_DIR/${WHOAMI}_erl_pipes/$NAME/}"
|
|
|
+
|
|
|
+## make EMQX_NODE_COOKIE right
|
|
|
+if [ -n "${EMQX_NODE_COOKIE:-}" ]; then
|
|
|
+ export EMQX_NODE__COOKIE="${EMQX_NODE_COOKIE}"
|
|
|
+ unset EMQX_NODE_COOKIE
|
|
|
+fi
|
|
|
+COOKIE="${EMQX_NODE__COOKIE:-}"
|
|
|
+if [ -z "$COOKIE" ]; then
|
|
|
+ if [ "$IS_BOOT_COMMAND" = 'yes' ]; then
|
|
|
+ COOKIE="$(call_hocon -s $SCHEMA_MOD -I "$CONFIGS_DIR/" -c "$RUNNER_ETC_DIR"/emqx.conf get node.cookie | tr -d \")"
|
|
|
+ else
|
|
|
+ vm_args_file="$(latest_vm_args 'EMQX_NODE__COOKIE')"
|
|
|
+ COOKIE="$(grep -E '^-setcookie' "${vm_args_file}" | awk '{print $2}')"
|
|
|
+ fi
|
|
|
+fi
|
|
|
+
|
|
|
+if [ -z "$COOKIE" ]; then
|
|
|
+ die "Please set node.cookie in $RUNNER_ETC_DIR/emqx.conf or override from environment variable EMQX_NODE__COOKIE"
|
|
|
+fi
|
|
|
|
|
|
cd "$ROOTDIR"
|
|
|
|
|
|
@@ -209,7 +556,6 @@ case "${COMMAND}" in
|
|
|
HEART_OPTION="start"
|
|
|
;;
|
|
|
esac
|
|
|
- export START_OPTION
|
|
|
RUN_PARAM="$*"
|
|
|
|
|
|
# Set arguments for the heart command
|