#!/bin/bash

: ${OCF_FUNCTIONS=${OCF_ROOT}/resource.d/heartbeat/.ocf-shellfuncs}
. ${OCF_FUNCTIONS}
: ${__OCF_ACTION=$1}

OCF_RESKEY_dampening_delay_default=5
OCF_RESKEY_attr_name_prefix_default=drbd-promotion-score
OCF_RESKEY_record_event_details_default=false

# see pacemaker:include/crm/crm.h CRM_SCORE_INFINITY
MINUS_INFINITY=-1000000

meta-data() {
    cat <<___
<?xml version="1.0"?>
<!DOCTYPE resource-agent SYSTEM "ra-api-1.dtd">
<resource-agent name="drbd-attr" version="1.0">
<version>1.0</version>

<longdesc lang="en">
This listens for DRBD state change events, and sets or deletes transient node
attributes based on the "promotion_score" and "may_promote" values as presented
by the DRBD events2 interface.

Optionally using a dampening delay, see attrd_updater for details.

To be used as a clone on all DRBD nodes.  The idea is to start DRBD outside of
pacemaker, use DRBD auto-promote, and add location constraints for the
Filesystem or other resource agents which are using DRBD.
</longdesc>
<shortdesc lang="en">import DRBD state change events as transient node attributes</shortdesc>

<parameters>
<parameter name="dampening_delay">
<longdesc lang="en">
To be used as dampening delay in attrd_updater.
</longdesc>
<shortdesc lang="en">attrd_updater --delay</shortdesc>
<content type="integer" default="$OCF_RESKEY_dampening_delay_default" />
</parameter>

<parameter name="attr_name_prefix">
<longdesc lang="en">
The attributes will be named "*prefix*-drbd_resource_name".
You can chose that prefix here.
</longdesc>
<shortdesc lang="en">attrd_updater --name *prefix*-drbd_resource_name</shortdesc>
<content type="string" default="$OCF_RESKEY_attr_name_prefix_default" />
</parameter>

<parameter name="record_event_details">
<longdesc lang="en">
It may be convenient to know which event lead to the current score.
This setting toggles the recording of the event.
The attributes will be named "*prefix*:event-details-drbd_resource_name".
</longdesc>
<shortdesc lang="en"></shortdesc>
<content type="boolean" default="$OCF_RESKEY_record_event_details_default" />
</parameter>
</parameters>

<actions>
<action name="start"        timeout="20s" />
<action name="stop"         timeout="20s" />
<action name="monitor"      timeout="20s" interval="60s" depth="0"/>
<action name="validate-all" timeout="20s" />
<action name="meta-data"    timeout="5s" />
</actions>
</resource-agent>
___
}

validate-all()
{
	ocf_check_binary drbdadm
	ocf_check_binary drbdsetup

	# I think we can expect "coreutils"...
	# ocf_check_binary stat
	# ocf_check_binary sort
	# ocf_check_binary join

	# we need at least drbd utils 9.14.1
	if (( $(drbdadm -V | sed -ne 's/^DRBDADM_VERSION_CODE=\(0x[0-9a-f]*\)$/\1/p') < 0x090e00 )) ; then
		ocf_exit_reason "need at least drbd-utils 9.14.0"
		return $OCF_ERR_INSTALLED
	fi

	# Or do we allow empty prefix? Resulting attribute names would be "-$DRBD_RESOURCE_NAME" ...
	[[ -n $attr_name_prefix ]] || { ocf_exit_reason "attr_name_prefix must not be empty"; return $OCF_ERR_CONFIGURED; }
}

monitor()
{
	test -e $PIDFILE || return $OCF_NOT_RUNNING
	test -s $PIDFILE || return $OCF_ERR_GENERIC
	# could even add a: test $PIDFILE -ef "/proc/$pid/fd/9"
	read pid < $PIDFILE &> /dev/null && kill -0 "$pid" && return $OCF_SUCCESS
	if [[ $__OCF_ACTION == monitor ]]; then
		# try to report the "last words of the previous instance" as exit reason
		tmp=$(crm_attribute --lifetime reboot --query --name $attr_name_prefix --quiet)
		[[ $tmp ]] && ocf_exit_reason ":: $tmp"
	fi
	return $OCF_ERR_GENERIC
}

list_existing_attributes()
{
	cibadmin -Q --xpath "/cib/status/node_state[@id='$node_id']/transient_attributes/instance_attributes/nvpair[starts-with(@name,'${attr_name_prefix}-')]" |
		sed -ne 's,^.* name="'"${attr_name_prefix}"'-\([^"]*\)".*,\1,p'
}

re_init_daemon()
{
	resources_with_existing_attributes=$( list_existing_attributes )
	resources_known_to_drbdadm=$( drbadm sh-resources )
	resources_seen_in_initial_dump=''
	status_delay=''
	exec 0< <(exec drbdsetup events2 --timestamps)
}

delete_event_details()
{
	local name=$1
	attrd_updater -n $attr_name_prefix:event-details-$name --delete
}

record_event_details()
{
	$record_event_details || return 0
	local name=$1 ; shift # rest is "message"
	attrd_updater -n $attr_name_prefix:event-details-$name --delay $dampening_delay --update-both "$*"
}

delete_name()
{
	local name=$1
	# at the time of writing, there is only --update-both (delay and value),
	# but no --update-delay-and-delete...
	attrd_updater -n $attr_name_prefix-$name --delete
	attrd_updater -n $attr_name_prefix-$name --update-delay --delay 0
	delete_event_details $name
}

update_name_to_minus_inf()
{
	local name=$1
	attrd_updater -n $attr_name_prefix-$name --update-both $MINUS_INFINITY --delay 0
}

update_name_to_score()
{
	local name=$1 score=$2
	case $score in
	""|*[!0-9]*)
		delete_name $name ;;
	0)
		update_name_to_minus_inf $name ;;
	*)
		attrd_updater -n $attr_name_prefix-$name --update-both $score --delay $dampening_delay
	esac
}

handle_stale_attributes()
{
	local stale_and_unknown stale_but_known
	stale_but_known=$(
		join -v 2 <(printf "%s\n" $resources_seen_in_initial_dump | sort -u ) \
			<(printf "%s\n" $resources_known_to_drbdadm | sort -u )
		)
	stale_and_unknown=$(
		join -v 2 <(printf "%s\n" $resources_seen_in_initial_dump $resources_known_to_drbdadm | sort -u ) \
			<(printf "%s\n" $resources_with_existing_attributes | sort )
		)
	for r in $stale_and_unknown ; do delete_name $r;  done
	for r in $stale_but_known ; do update_name_to_minus_inf $r ; done
	unset resources_with_existing_attributes
	unset resources_seen_in_initial_dump
	unset resources_known_to_drbdadm
}

remove_all_event_details()
{
	# !! NOT --delete --force --xpath,
	# that would only manipulate the CIB,
	# but the ATTRD would still remember.
	local attr
	for attr in $(
		cibadmin -Q --xpath "/cib/status/node_state[@id='$node_id']/transient_attributes/instance_attributes/nvpair[starts-with(@name,'${attr_name_prefix}:event-details-')]" |
			sed -ne 's,^.* name="\([^"]*\)".*,\1,p'
		)
	do
		attrd_updater -n $attr --delete
	done
}

toggle_event_details()
{
	if $record_event_details ; then
		record_event_details=false
		remove_all_event_details
	else
		record_event_details=true
	fi
}

the-daemon()
{
	node_id=$(crm_node -i)
	event=()

	trap toggle_event_details USR1

	re_init_daemon
	trap "re_init_daemon" HUP
	while prev=( "${event[@]}" ); read -a event; do
		# Is this still *our* pidfile?
		test $PIDFILE -ef /proc/self/fd/9 || { event="lost PIDFILE"; break; }

		# handle end of initial state dump
		if [[ "${event[1]} ${event[2]}" == "exists -" ]]; then
			handle_stale_attributes
			attrd_updater -n $attr_name_prefix --update-both "(re)init completed at $(date "+%F %T") by $OCF_RESOURCE_INSTANCE [$BASHPID]" --delay 0
			[[ $dampening_delay != 0 ]] && status_delay=$dampening_delay || status_delay=$OCF_RESKEY_dampening_delay_default
			continue
		fi

		# ignore non-resource events for now
		[[ ${event[2]} == resource ]] || continue

		# parse the events line
		name=''
		promotion_score=''
		for f in ${event[@]:3}; do
			k=${f%%:*}; v=${f#*:}
			[[ $k == name ]]		&& name=$v
			[[ $k == promotion_score ]]	&& promotion_score=$v
		done

		# events lines must always contain the resource name
		[[ -n $name ]] || break

		# record name, if still in initial state dump
		[[ ${event[1]} == exists ]] && resources_seen_in_initial_dump+=" $name"

		# either this was a "destroy" event (and we remove the attribute),
		# or the event contained a promotion_score,
	        # or we ignore this line.
		[[ ${event[1]} == destroy ]] || [[ $promotion_score ]] || continue

		update_name_to_score $name $promotion_score || {
			event+=( "// attrd_updater failed with exit code $?" )
			break
		}
		record_event_details $name "${event[*]}"

		# let them know when we saw the last event, unless still within the initial state dump
		[[ $status_delay ]] &&
		attrd_updater -n $attr_name_prefix --update-both "Last event at $(date "+%F %T") by $OCF_RESOURCE_INSTANCE [$BASHPID]" --delay $status_delay
	done

	# some means of "error reporting" ...
	attrd_updater -n $attr_name_prefix --update-both "FAILED at $(date "+%F %T") as $OCF_RESOURCE_INSTANCE, inactive; last input: '${prev[*]}' // '${event[*]}'" --delay 0

	# Let them know we failed, but only do this if this script survived for
	# a minute already, using the bash magic SECONDS variable.
	# If we did not make it that long, don't immediately report that failure,
	# I want to avoid a potentially "tight" recovery loop.
	# Pacemaker will notice on the next monitoring interval,
	(( $SECONDS >= 60 )) && crm_resource --fail --resource $OCF_RESOURCE_INSTANCE
}

start()
{
	validate-all || return
	monitor && return $OCF_SUCCESS
	test -e $PIDFILE && return $OCF_GENERIC

	exec 9> $PIDFILE || return $OCF_GENERIC
	(
		echo $BASHPID >&9
		handle_SIGTERM() {
			attrd_updater -n $attr_name_prefix --update-both "STOPPED at $(date "+%F %T") as $OCF_RESOURCE_INSTANCE, inactive" --delay 0
			rm -f $PIDFILE
			exit
		}
		trap "handle_SIGTERM" TERM

		attrd_updater -n $attr_name_prefix --update-both "STARTED at $(date "+%F %T") as $OCF_RESOURCE_INSTANCE, initializing" --delay 0

		the-daemon
	) </dev/null >/dev/null 2>&1 &

	test -e $PIDFILE || return $OCF_ERR_GENERIC
	while ! monitor ; do sleep 1; done
	return $OCF_SUCCESS
}

stop()
{
	if monitor; then
		read pid < $PIDFILE && kill -TERM "$pid"
		while test -e $PIDFILE ; do sleep 1; done
	else
		rm -f $PIDFILE
	fi
	return $OCF_SUCCESS
}

dampening_delay=${OCF_RESKEY_dampening_delay:=$OCF_RESKEY_dampening_delay_default}
attr_name_prefix=${OCF_RESKEY_attr_name_prefix:=$OCF_RESKEY_attr_name_prefix_default}
if ocf_is_true ${OCF_RESKEY_record_event_details:=$OCF_RESKEY_record_event_details_default} ; then
	record_event_details=true
else
	record_event_details=false
	remove_all_event_details
fi

PIDFILE=${HA_VARRUN%%/}/drbd-attr-${OCF_RESOURCE_INSTANCE}.pid


case $__OCF_ACTION in
	validate-all|meta-data|start|stop|monitor)
		$__OCF_ACTION
		;;
	*)
		ocf_exit_reason "'$__OCF_ACTION' not implemented"
		exit $OCF_ERR_UNIMPLEMENTED
esac

exit $? # that would happen implicitly anyways