|
| 1 | +# OSC NHC - slabinfo checks |
| 2 | +# |
| 3 | +# Trey Dockendorf < [email protected]> |
| 4 | +# 21 September 2018 |
| 5 | +# |
| 6 | + |
| 7 | +SLAB_NAME=() |
| 8 | +SLAB_ACTIVE_OBJS=() |
| 9 | +SLAB_NUM_OBJS=() |
| 10 | +SLAB_ACTIVE_SLABS=() |
| 11 | +SLAB_NUM_SLABS=() |
| 12 | +export SLAB_NAME SLAB_ACTIVE_OBJS SLAB_NUM_OBJS SLAB_ACTIVE_SLABS SLAB_NUM_SLABS |
| 13 | + |
| 14 | +function nhc_slabinfo_gather_data() { |
| 15 | + local LINE_CNT |
| 16 | + local -a FIELD |
| 17 | + |
| 18 | + SLAB_NAME=() SLAB_ACTIVE_OBJS=() SLAB_NUM_OBJS=() SLAB_ACTIVE_SLABS=() SLAB_NUM_SLABS=() |
| 19 | + |
| 20 | + ((LINE_CNT=0)) |
| 21 | + while read -a FIELD; do |
| 22 | + if [[ "${FIELD[0]}" == "#" ]]; then |
| 23 | + continue |
| 24 | + fi |
| 25 | + if [[ "${FIELD[0]}" == "slabinfo" ]]; then |
| 26 | + continue |
| 27 | + fi |
| 28 | + SLAB_NAME[$LINE_CNT]="${FIELD[0]}" |
| 29 | + SLAB_ACTIVE_OBJS[$LINE_CNT]="${FIELD[1]}" |
| 30 | + SLAB_NUM_OBJS[$LINE_CNT]="${FIELD[2]}" |
| 31 | + SLAB_ACTIVE_SLABS[$LINE_CNT]="${FIELD[13]}" |
| 32 | + SLAB_NUM_SLABS[$LINE_CNT]="${FIELD[14]}" |
| 33 | + dbg "Got slabinfo ${SLAB_NAME[$LINE_CNT]} ${SLAB_ACTIVE_OBJS[$LINE_CNT]} ${SLAB_NUM_OBJS[$LINE_CNT]} ${SLAB_ACTIVE_SLABS[$LINE_CNT]} ${SLAB_NUM_SLABS[$LINE_CNT]}" |
| 34 | + ((LINE_CNT++)) |
| 35 | + done < /proc/slabinfo |
| 36 | + |
| 37 | + export SLAB_NAME SLAB_ACTIVE_OBJS SLAB_NUM_OBJS SLAB_ACTIVE_SLABS SLAB_NUM_SLABS |
| 38 | +} |
| 39 | + |
| 40 | +# Do the actual work of looking at slabinfo. |
| 41 | +# Used by check_slabinfo_{active_objs,num_objs,active_slabs,num_slabs}. |
| 42 | +# Additional parameters are the check name and the type of resource to look at. |
| 43 | +function nhc_slabinfo_check() { |
| 44 | + local NONFATAL=0 ALL=0 LOG=0 SYSLOG=0 MATCH="" ACTION="" |
| 45 | + local CHECK="$1" COL="$2" THRESHOLD |
| 46 | + local THIS_NAME MSG i |
| 47 | + |
| 48 | + if [[ ${#SLAB_NAME[*]} -eq 0 ]]; then |
| 49 | + nhc_slabinfo_gather_data |
| 50 | + fi |
| 51 | + |
| 52 | + shift 2 |
| 53 | + OPTIND=1 |
| 54 | + while getopts ":0alsm:e:" OPTION ; do |
| 55 | + case "$OPTION" in |
| 56 | + 0) NONFATAL=1 ;; |
| 57 | + a) ALL=1 ;; |
| 58 | + l) LOG=1 ;; |
| 59 | + s) SYSLOG=1 ;; |
| 60 | + m) MATCH="$OPTARG" ;; |
| 61 | + e) ACTION="$OPTARG" ;; |
| 62 | + :) die 1 "$CHECK: Option -$OPTARG requires an argument." ; return 1 ;; |
| 63 | + \?) die 1 "$CHECK: Invalid option: -$OPTARG" ; return 1 ;; |
| 64 | + esac |
| 65 | + done |
| 66 | + shift $((OPTIND-1)) |
| 67 | + THRESHOLD="$1" |
| 68 | + if [[ -z "$THRESHOLD" || "${THRESHOLD//[^0-9]}" != "$THRESHOLD" ]]; then |
| 69 | + die 1 "$CHECK: Syntax error: threshold must be an integer." |
| 70 | + fi |
| 71 | + if [[ -z "$MATCH" ]]; then |
| 72 | + MATCH="*" |
| 73 | + fi |
| 74 | + dbg "Looking for slabinfo matching \"$MATCH\"" |
| 75 | + for ((i=0; i < ${#SLAB_NAME[*]}; i++)); do |
| 76 | + THIS_NAME="${SLAB_NAME[$i]}" |
| 77 | + dbg "CHECKING \"$THIS_NAME\" vs. \"$MATCH\"" |
| 78 | + if ! mcheck "$THIS_NAME" "$MATCH"; then |
| 79 | + continue |
| 80 | + fi |
| 81 | + dbg "Matching slabinfo found: $THIS_NAME: active_objs=${SLAB_ACTIVE_OBJS[$i]} num_objs=${SLAB_NUM_OBJS[$i]} active_slabs=${SLAB_ACTIVE_SLABS[$i]} num_slabs=${SLAB_NUM_SLABS[$i]}" |
| 82 | + case $COL in |
| 83 | + 1) if [[ ${SLAB_ACTIVE_OBJS[$i]} -lt $THRESHOLD ]]; then continue ; fi |
| 84 | + MSG="$CHECK: slabinfo \"$THIS_NAME\" using ${SLAB_ACTIVE_OBJS[$i]} active_objs (limit $THRESHOLD)" |
| 85 | + ;; |
| 86 | + 2) if [[ ${SLAB_NUM_OBJS[$i]} -lt $THRESHOLD ]]; then continue ; fi |
| 87 | + MSG="$CHECK: slabinfo \"$THIS_NAME\" using ${SLAB_NUM_OBJS[$i]} num_objs (limit $THRESHOLD)" |
| 88 | + ;; |
| 89 | + 3) if [[ ${SLAB_ACTIVE_SLABS[$i]} -lt $THRESHOLD ]]; then continue ; fi |
| 90 | + MSG="$CHECK: slabinfo \"$THIS_NAME\" using ${SLAB_ACTIVE_SLABS[$i]} active_slabs (limit $THRESHOLD)" |
| 91 | + ;; |
| 92 | + 4) if [[ ${SLAB_NUM_SLABS[$i]} -lt $THRESHOLD ]]; then continue ; fi |
| 93 | + MSG="$CHECK: slabinfo \"$THIS_NAME\" using ${SLAB_NUM_SLABS[$i]} num_slabs (limit $THRESHOLD)" |
| 94 | + ;; |
| 95 | + esac |
| 96 | + # We have a winner. Or loser, as the case may be. |
| 97 | + if [[ "$LOG" == "1" ]]; then |
| 98 | + log $MSG |
| 99 | + fi |
| 100 | + if [[ "$SYSLOG" == "1" ]]; then |
| 101 | + syslog $MSG |
| 102 | + fi |
| 103 | + if [[ "$ACTION" != "" ]]; then |
| 104 | + ${SHELL:-/bin/bash} -c "$ACTION" & |
| 105 | + fi |
| 106 | + if [[ $ALL -ge 1 ]]; then |
| 107 | + if [[ -n "$MSG" ]]; then |
| 108 | + log "$MSG ($ALL)" |
| 109 | + fi |
| 110 | + ((ALL++)) |
| 111 | + continue |
| 112 | + elif [[ $NONFATAL == 1 ]]; then |
| 113 | + if [[ -n "$MSG" ]]; then |
| 114 | + log "$MSG (non-fatal)" |
| 115 | + fi |
| 116 | + return 0 |
| 117 | + fi |
| 118 | + die 1 "$MSG" |
| 119 | + return 1 |
| 120 | + done |
| 121 | + # -a (all) does not necessarily imply -0 (non-fatal). A value of 1 for $ALL |
| 122 | + # means -a was passed in but no errors were found. 2 or above is an error. |
| 123 | + if [[ $ALL -gt 1 ]]; then |
| 124 | + # We had at least 1 flagged process. Fail unless we're also non-fatal. |
| 125 | + if [[ $NONFATAL == 1 ]]; then |
| 126 | + if [[ -n "$MSG" ]]; then |
| 127 | + log "$MSG (non-fatal)" |
| 128 | + fi |
| 129 | + return 0 |
| 130 | + fi |
| 131 | + ((ALL--)) |
| 132 | + die $ALL "$MSG (last of $ALL)" |
| 133 | + return $ALL |
| 134 | + fi |
| 135 | + return 0 |
| 136 | +} |
| 137 | + |
| 138 | +# Checks slabinfo active_objs is above a threshold |
| 139 | +# check_slabinfo_active_objs [-0] [-a] [-l] [-s] [-m <match>] [-e <action>] <threshold> |
| 140 | +function check_slabinfo_active_objs() { |
| 141 | + nhc_slabinfo_check $FUNCNAME 1 "$@" |
| 142 | +} |
| 143 | + |
| 144 | +# Checks slabinfo num_objs is above a threshold |
| 145 | +# check_slabinfo_num_objs [-0] [-a] [-l] [-s] [-m <match>] [-e <action>] <threshold> |
| 146 | +function check_slabinfo_num_objs() { |
| 147 | + nhc_slabinfo_check $FUNCNAME 2 "$@" |
| 148 | +} |
| 149 | + |
| 150 | +# Checks slabinfo active_slabs is above a threshold |
| 151 | +# check_slabinfo_active_slabs [-0] [-a] [-l] [-s] [-m <match>] [-e <action>] <threshold> |
| 152 | +function check_slabinfo_active_slabs() { |
| 153 | + nhc_slabinfo_check $FUNCNAME 3 "$@" |
| 154 | +} |
| 155 | + |
| 156 | +# Checks slabinfo num_slabs is above a threshold |
| 157 | +# check_slabinfo_num_slabs [-0] [-a] [-l] [-s] [-m <match>] [-e <action>] <threshold> |
| 158 | +function check_slabinfo_num_slabs() { |
| 159 | + nhc_slabinfo_check $FUNCNAME 4 "$@" |
| 160 | +} |
0 commit comments