ctdb/tools/ctdb_diagnostics

   1 #!/bin/sh
   2 # a script to test the basic setup of a CTDB/Samba install
   3 # tridge@samba.org September 2007
   4 # martin@meltin.net August 2010
   5
   6 usage ()
   7 {
   8     cat >&2 <<EOF
   9 Usage: ctdb_diagnostics [OPTION] ...
  10   options:
  11     -n <nodes>  Comma separated list of nodes to operate on
  12     -c          Ignore comment lines (starting with '#') in file comparisons
  13     -w          Ignore whitespace in file comparisons
  14     --no-ads    Do not use commands that assume an Active Directory Server
  15 EOF
  16     exit 1
  17
  18 }
  19
  20 nodes=$(ctdb listnodes -X | cut -d'|' -f2)
  21 bad_nodes=""
  22 diff_opts=
  23 no_ads=false
  24
  25 parse_options ()
  26 {
  27     temp=$(getopt -n "ctdb_diagnostics" -o "n:cwh" -l no-ads,help -- "$@")
  28
  29     # No! Checking the exit code afterwards is actually clearer...
  30     # shellcheck disable=SC2181
  31     [ $? -eq 0 ] || usage
  32
  33     eval set -- "$temp"
  34
  35     while true ; do
  36         case "$1" in
  37             -n) nodes=$(echo "$2" | sed -e 's@,@ @g') ; shift 2 ;;
  38             -c) diff_opts="${diff_opts} -I ^#.*" ; shift ;;
  39             -w) diff_opts="${diff_opts} -w" ; shift ;;
  40             --no-ads) no_ads=true ; shift ;;
  41             --) shift ; break ;;
  42             -h|--help|*) usage ;;
  43         esac
  44     done
  45
  46     [ $# -ne 0 ] && usage
  47 }
  48
  49 parse_options "$@"
  50
  51 # Use 5s ssh timeout if EXTRA_SSH_OPTS doesn't set a timeout.
  52 case "$EXTRA_SSH_OPTS" in
  53     *ConnectTimeout=*) : ;;
  54     *)
  55         export EXTRA_SSH_OPTS="${EXTRA_SSH_OPTS} -o ConnectTimeout=5"
  56 esac
  57
  58 # Filter nodes.  Remove any nodes we can't contact from $node and add
  59 # them to $bad_nodes.
  60 _nodes=""
  61 for _i in $nodes ; do
  62     if onnode "$_i" true >/dev/null 2>&1 ; then
  63         _nodes="${_nodes}${_nodes:+ }${_i}"
  64     else
  65         bad_nodes="${bad_nodes}${bad_nodes:+,}${_i}"
  66     fi
  67 done
  68 nodes="$_nodes"
  69
  70 nodes_comma=$(echo "$nodes" | sed -e 's@[[:space:]]@,@g')
  71
  72 PATH="$PATH:/sbin:/usr/sbin:/usr/lpp/mmfs/bin"
  73
  74 # list of config files that must exist and that we check are the same
  75 # on the nodes
  76 if [ -d /etc/sysconfig ] ; then
  77     CONFIG_FILES_MUST="/etc/krb5.conf /etc/hosts /usr/local/etc/ctdb/nodes /etc/sysconfig/ctdb /etc/resolv.conf /etc/nsswitch.conf /etc/sysctl.conf /etc/samba/smb.conf /etc/fstab /etc/multipath.conf /etc/pam.d/system-auth /etc/sysconfig/nfs /etc/exports /etc/vsftpd/vsftpd.conf"
  78 else
  79     CONFIG_FILES_MUST="/etc/krb5.conf /etc/hosts /usr/local/etc/ctdb/nodes /etc/default/ctdb /etc/resolv.conf /etc/nsswitch.conf /etc/sysctl.conf /etc/samba/smb.conf /etc/fstab /etc/multipath.conf /etc/pam.d/system-auth /etc/default/nfs /etc/exports /etc/vsftpd/vsftpd.conf"
  80 fi
  81
  82 # list of config files that may exist and should be checked that they
  83 # are the same on the nodes
  84 CONFIG_FILES_MAY="/usr/local/etc/ctdb/public_addresses /usr/local/etc/ctdb/static-routes"
  85
  86 exec 2>&1
  87
  88 cat <<EOF
  89 --------------------------------------------------------------------
  90 ctdb_diagnostics starting. This script will gather information about
  91 your ctdb cluster. You should send the output of this script along
  92 with any ctdb or clustered Samba bug reports.
  93 --------------------------------------------------------------------
  94 EOF
  95
  96 date
  97
  98 error() {
  99     msg="$1"
 100     echo "ERROR: $msg"
 101     NUM_ERRORS=$((NUM_ERRORS + 1))
 102     echo " ERROR[$NUM_ERRORS]: $msg" >> "$ERRORS"
 103 }
 104
 105 show_file() {
 106     fname="$1"
 107     _fdetails=$(ls -l "$fname" 2>&1)
 108     echo "  ================================"
 109     echo "  File: $fname"
 110     echo "  $_fdetails"
 111     sed 's/^/  /' "$fname" 2>&1
 112     echo "  ================================"
 113 }
 114
 115 show_all() {
 116     echo "running $1 on nodes $nodes_comma"
 117     onnode "$nodes_comma" "hostname; date; $1 2>&1 | sed 's/^/  /'" 2>&1
 118 }
 119
 120 show_and_compare_files () {
 121
 122     fmt="$1" ; shift
 123
 124     for f ; do
 125         _bf=$(basename "$f")
 126         first=true
 127
 128         for n in $nodes ; do
 129
 130             if $first ; then
 131                 onnode "$n" [ -r "$f" ] || {
 132                     # This function takes a format string
 133                     # shellcheck disable=SC2059
 134                     msg=$(printf "$fmt" "$f" "$n")
 135                     error "$msg"
 136                     continue 2;
 137                 }
 138
 139                 fstf="${tmpdir}/${_bf}.node${n}"
 140                 onnode "$n" cat "$f" >"$fstf" 2>&1
 141
 142                 _fdetails=$(onnode "$n" ls -l "$f" 2>&1)
 143                 echo "  ================================"
 144                 echo "  File (on node $n): $f"
 145                 echo "  $_fdetails"
 146                 sed 's/^/  /' "$fstf"
 147                 echo "  ================================"
 148                 first=false
 149             else
 150                 echo "Testing for same config file $f on node $n"
 151                 tmpf="${tmpdir}/${_bf}.node${n}"
 152                 onnode "$n" cat "$f" >"$tmpf" 2>&1
 153                 # Intentional multi-word splitting on diff_opts
 154                 # shellcheck disable=SC2086
 155                 diff $diff_opts "$fstf" "$tmpf" >/dev/null 2>&1 || {
 156                     error "File $f is different on node $n"
 157                     diff -u $diff_opts "$fstf" "$tmpf"
 158                 }
 159                 rm -f "$tmpf"
 160             fi
 161         done
 162
 163         rm -f "$fstf"
 164     done
 165 }
 166
 167 if ! tmpdir=$(mktemp -d) ; then
 168     echo "Unable to create a temporary directory"
 169     exit 1
 170 fi
 171 ERRORS="${tmpdir}/diag_err"
 172 NUM_ERRORS=0
 173
 174 cat <<EOF
 175 Diagnosis started on these nodes:
 176 $nodes_comma
 177 EOF
 178
 179 if [ -n "$bad_nodes" ] ; then
 180     cat <<EOF
 181
 182 NOT RUNNING DIAGNOSTICS on these uncontactable nodes:
 183 $bad_nodes
 184 EOF
 185
 186 fi
 187
 188 cat <<EOF
 189
 190 For reference, here is the nodes file on the current node...
 191 EOF
 192
 193 show_file /usr/local/etc/ctdb/nodes
 194
 195 cat <<EOF
 196 --------------------------------------------------------------------
 197 Comping critical config files on nodes $nodes_comma
 198 EOF
 199
 200 # Intentional multi-word splitting on CONFIG_FILES_MUST
 201 # shellcheck disable=SC2086
 202 show_and_compare_files \
 203     "%s is missing on node %d" \
 204     $CONFIG_FILES_MUST
 205
 206 # Intentional multi-word splitting on CONFIG_FILES_MAY
 207 # shellcheck disable=SC2086
 208 show_and_compare_files \
 209     "Optional file %s is not present on node %d" \
 210     $CONFIG_FILES_MAY
 211
 212 cat <<EOF
 213 --------------------------------------------------------------------
 214 Checking for clock drift
 215 EOF
 216 t=$(date +%s)
 217 for i in $nodes; do
 218     t2=$(onnode "$i" date +%s)
 219     d=$((t2 - t))
 220     if [ "$d" -gt 30 ] || [ "$d" -lt -30 ]; then
 221         error "time on node $i differs by $d seconds"
 222     fi
 223 done
 224
 225 cat <<EOF
 226 --------------------------------------------------------------------
 227 Showing software versions
 228 EOF
 229 show_all "uname -a"
 230 [ -x /bin/rpm ] && {
 231     show_all "rpm -qa | egrep 'samba|ctdb|gpfs'"
 232 }
 233 [ -x /usr/bin/dpkg-query ] && {
 234     show_all "/usr/bin/dpkg-query --show 'ctdb'"
 235     show_all "/usr/bin/dpkg-query --show 'samba'"
 236     #show_all "/usr/bin/dpkg-query --show 'gpfs'"
 237 }
 238
 239
 240 cat <<EOF
 241 --------------------------------------------------------------------
 242 Showing ctdb status and recent log entries
 243 EOF
 244 show_all "ctdb status; ctdb ip"
 245 show_all "ctdb statistics"
 246 show_all "ctdb uptime"
 247 show_all "ctdb listvars"
 248 show_all "ctdb getdbmap"
 249 show_all "ctdb -X getdbmap | awk -F'|' 'NR > 1 {print \$3}' | sort | xargs -n 1 ctdb dbstatistics"
 250
 251 echo "Showing log.ctdb"
 252 show_all "test -f /usr/local/var/log/log.ctdb && tail -100 /usr/local/var/log/log.ctdb"
 253
 254 show_all "tail -200 /var/log/messages"
 255 show_all "ls -lRs /usr/local/var/lib/ctdb"
 256 show_all "ls -lRs /usr/local/etc/ctdb"
 257
 258
 259 cat <<EOF
 260 --------------------------------------------------------------------
 261 Showing system and process status
 262 EOF
 263 show_all "df"
 264 show_all "df -i"
 265 show_all "mount"
 266 show_all "w"
 267 show_all "ps axfwu"
 268 show_all "dmesg"
 269 show_all "/sbin/lspci"
 270 show_all "dmidecode"
 271 show_all "cat /proc/partitions"
 272 show_all "cat /proc/cpuinfo"
 273 show_all "cat /proc/scsi/scsi"
 274 show_all "/sbin/ifconfig -a"
 275 show_all "/sbin/ifconfig -a"
 276 show_all "/sbin/ip addr list"
 277 show_all "/sbin/route -n"
 278 show_all "netstat -s"
 279 show_all "free"
 280 show_all "crontab -l"
 281 show_all "sysctl -a"
 282 show_all "iptables -L -n"
 283 show_all "iptables -L -n -t nat"
 284 show_all "/usr/sbin/rpcinfo -p"
 285 show_all "/usr/sbin/showmount -a"
 286 show_all "/usr/sbin/showmount -e"
 287 show_all "/usr/sbin/nfsstat -v"
 288 [ -x /sbin/multipath ] && {
 289     show_all "/sbin/multipath -ll"
 290 }
 291 [ -x /sbin/chkconfig ] && {
 292     show_all "/sbin/chkconfig --list"
 293 }
 294 [ -x /usr/sbin/getenforce ] && {
 295     show_all "/usr/sbin/getenforce"
 296 }
 297 [ -d /proc/net/bonding ] && {
 298     for f in /proc/net/bonding/*; do
 299         show_all "cat $f"
 300     done
 301 }
 302
 303 cat <<EOF
 304 --------------------------------------------------------------------
 305 Showing Samba status
 306 EOF
 307 show_all "smbstatus -n -B"
 308 if $no_ads ; then
 309     echo
 310     echo "Skipping \"net ads testjoin\" as requested"
 311     echo
 312 else
 313     show_all "net ads testjoin"
 314 fi
 315 show_all "net conf list"
 316 show_all "lsof -n | grep smbd"
 317 show_all "lsof -n | grep ctdbd"
 318 show_all "netstat -tan"
 319 if $no_ads ; then
 320     echo
 321     echo "Skipping \"net ads info\" as requested"
 322     echo
 323 else
 324     show_all "net ads info"
 325 fi
 326 show_all "date"
 327 show_all "smbclient -U% -L 127.0.0.1"
 328 WORKGROUP=$(testparm -s --parameter-name=WORKGROUP 2> /dev/null)
 329 show_all id "$WORKGROUP/Administrator"
 330 show_all "wbinfo -p"
 331 show_all "wbinfo --online-status"
 332 show_all "smbd -b"
 333
 334 date
 335 echo "Diagnostics finished with $NUM_ERRORS errors"
 336
 337 [ -r "$ERRORS" ] && {
 338     cat "$ERRORS"
 339     rm -f "$ERRORS"
 340 }
 341
 342 rm -rf "$tmpdir"
 343
 344 exit $NUM_ERRORS
 345