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