Skip to content

Commit

Permalink
wait-for-bricks: Use ps instead of pidof, improve performance
Browse files Browse the repository at this point in the history 
1) 'pidof' command may not exist in all setups, so changing that
to use 'ps' command which is available more often. If 'ps' command is
not found in the setup it prints that the command is not available
in the setup in the error log.

2) For finding if 10 block volumes are up and running it was taking 9 seconds
when all the bricks are up. I changed the script to reduce the number of
gluster commands that are executed to get this time to 1 seconds.

Tested-by: Prasanna Kumar Kalever <prasanna.kalever@redhat.com>
Reviewed-by: Prasanna Kumar Kalever <prasanna.kalever@redhat.com>
Reviewed-by: Niels Devos <ndevos@redhat.com>
Signed-off-by: Pranith Kumar K <pkarampu@redhat.com>
  • Loading branch information
@pkalever
Pranith Kumar K authored and pkalever committed Aug 9, 2018
1 parent 3541885 commit e596c3a
Showing 1 changed file with 43 additions and 77 deletions.
120 changes: 43 additions & 77 deletions extras/wait-for-bricks.sh
View file
Original file line number Diff line number Diff line change
Expand Up @@ -22,68 +22,24 @@ function printLog()
echo "[$(date -u +'%Y-%m-%d %I:%M:%S')] ${msg}" >> "${LOGDIR}/gluster-block-bricks-start.log"
}

function volinfo_field()
{
local vol=$1;
local field=$2;

gluster volume info "$vol" --xml | grep -oPm1 "(?<=<$field>)[^<]+"
}

function is_enabled()
{
case "$1" in
0|off|no|false|disable)
echo "N"
;;
1|on|yes|true|enable)
echo "Y"
;;
*)
echo "N/A"
;;
esac
}

function volinfo_option()
function volume_online_brick_count()
{
local vol=$1;
local key=$2;

gluster volume get "$vol" "$key"| awk '{print $NF}' | tail -1
}

function check_brick_status() {
local volname=$1
local daemon=$2

cmd="gluster --xml volume status $volname"
if [[ -z "$daemon" ]]
then
$cmd | grep -c '<status>1'
else
$cmd | sed -n -e "/${daemon}/,/<status>/ p" | grep -c '<status>1'
fi
local brick_count=$2
gluster --xml volume status "$volname" | grep '<status>' | head -n "$brick_count" | grep -c '<status>1'
}

function volume_online_brick_count()
function get_block_volnames_brick_count()
{
local volname=$1
local v1=0
local v2=0
local v3=0
local v4=0
local v5=0
local tot=0

#First count total Number of bricks and then subtract daemon status
v1=$(check_brick_status "$volname")
v2=$(check_brick_status "$volname" "Self-heal")
v3=$(check_brick_status "$volname" "Quota")
v4=$(check_brick_status "$volname" "Snapshot")
v5=$(check_brick_status "$volname" "Tier")
tot=$((v1-v2-v3-v4-v5))
echo $tot
gluster v info | \
# get volume name, status, shard-status, number-of-bricks
grep -P "(^Volume Name:|^Status:|^features.shard:|^Number of Bricks:)" | \
# get the 'started' volumes
sed '/^Volume Name:/{x;p;x;}' | sed -e '/./{H;$!d;}' -e 'x;/Status: Started/!d;' | \
# get shard enabled volumes
sed -e '/./{H;$!d;}' -e 'x;/features.shard: on/b' -e '/features.shard: yes/b' -e '/features.shard: enable/b' -e '/features.shard: true/b' -e '/features.shard: 1/b' -e d | \
# Print volume names
grep -P "(^Volume Name:|^Number of Bricks:)" | awk '{print $NF}'
}

function volumes_waiting_for_bricks_up()
Expand All @@ -92,34 +48,43 @@ function volumes_waiting_for_bricks_up()
local vol_down_info
while read -r volname
do
if [[ "Started" == "$(volinfo_field "$volname" 'statusStr')" ]]
read -r total_brick_count
brick_count="$(volume_online_brick_count "$volname" "$total_brick_count")"
if [[ $brick_count -ne $total_brick_count ]]
then
if [[ "Y" == "$(is_enabled "$(volinfo_option "$volname" features.shard)")" ]]
vol_down_info="$volname ($((total_brick_count - brick_count))/$total_brick_count)"
if [[ -z "$wait_info" ]]
then
brick_count="$(volume_online_brick_count "$volname")"
total_brick_count=$(volinfo_field "$volname" 'brickCount')
if [[ $brick_count -ne $total_brick_count ]]
then
vol_down_info="$volname ($((total_brick_count - brick_count))/$total_brick_count)"
if [[ -z "$wait_info" ]]
then
wait_info="$vol_down_info"
else
wait_info="$wait_info, $vol_down_info"
fi
fi
wait_info="$vol_down_info"
else
wait_info="$wait_info, $vol_down_info"
fi
fi
done < <(gluster volume list)
done < <(get_block_volnames_brick_count)
if [[ ! -z "$wait_info" ]]; then echo "$wait_info"; fi
}

function check_dependent_commands_exist()
{
declare -a arr=("ps" "sed" "grep" "awk" "head" "echo")
for i in "${arr[@]}"
do
if ! command -v "$i" > /dev/null 2>&1
then
printLog "ERROR: \'$i\' command is not found"
exit 1
fi
done

}
function check_glusterd_running()
{
if ! pidof glusterd > /dev/null 2>&1
# Explanation of why ps command is used this way:
# https://stackoverflow.com/questions/9117507/linux-unix-command-to-determine-if-process-is-running
if ! ps cax | grep -w '[g]lusterd' > /dev/null 2>&1
then
printLog "ERROR: Glusterd is not running";
exit 1;
printLog "ERROR: Glusterd is not running"
exit 1
fi
}

Expand All @@ -136,7 +101,7 @@ function wait_for_bricks_up()
printLog "WARNING: Timeout Expired, bricks of volumes:\"$(volumes_waiting_for_bricks_up)\" are yet to come online"
exit 1
fi
sleep 1;
sleep 1
done
}

Expand All @@ -151,7 +116,8 @@ case $1 in
*) ;;
esac

check_glusterd_running;
check_dependent_commands_exist
check_glusterd_running
timeout=$1
start=$(date +%s%N)
#Convert timeout to Nano Seconds as 'start' is in Nano seconds
Expand Down

0 comments on commit e596c3a

Please sign in to comment.