mars/contrib/background-migration-scripts/hooks-cm3.sh

789 lines
22 KiB
Bash
Raw Normal View History

#!/bin/bash
#
# This file is part of MARS project: http://schoebel.github.io/mars/
#
# Copyright (C) 2017 Thomas Schoebel-Theuer
# Copyright (C) 2017 1&1 Internet AG
#
# This program is free software; you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation; either version 2 of the License, or
# (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License along
# with this program; if not, write to the Free Software Foundation, Inc.,
# 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
###########################################
# 1&1 specific plugin / hooks for working with Jessie icpu conventions and cm3
#
# This script must be sourced from the main script.
commands_installed "curl json_pp bc"
function hook_get_mountpoint
{
local res="$1"
echo "/vol/$res"
}
function hook_get_hyper
{
local res="$1"
remote "$res" "source /lib/ui-config-framework/bash-includes; __config_getConfigVar HWNODE_NAME; echo \$HWNODE_NAME | cut -d. -f1"
}
function hook_get_store
{
local host="$1"
try="$(remote "$host" "source /lib/ui-config-framework/bash-includes; __config_getConfigVar CLUSTER_STORAGEHOST; echo \$CLUSTER_STORAGEHOST | cut -d. -f1")"
if [[ "$try" != "" ]]; then
echo "$try"
return
fi
# fallback to nc over iscsi network
try="$(remote "$host" "nc \$(iscsiadm -m session -o show | grep -o '[0-9]\+\.[0-9]\+\.[0-9]\+\.[0-9]\+') 101 2>/dev/null | cut -d. -f1" 1)"
if [[ "$try" != "" ]]; then
echo "$try"
return
fi
# fallback to indirect retrieval
local hyper="$(hook_get_hyper "$host")"
if [[ "$hyper" != "" ]] && [[ "$hyper" != "$host" ]]; then
hook_get_store "$hyper"
fi
}
function hook_get_vg
{
local host="$1"
remote "$host" "vgs | awk '{ print \$1; }' | grep 'vginfong\|vg[0-9]\+[ab]'"
}
function hook_resource_stop
{
local host="$1"
local res="$2"
declare -g downtime_begin
[[ "$downtime_begin" = "" ]] && downtime_begin="$(date +%s)"
echo "DOWNTIME BEGIN $(date)"
# stop the whole stack
remote "$host" "cm3 --stop $res || cm3 --stop $res || { mountpoint /vol/$res && umount /vol/$res; } || false"
}
function hook_resource_stop_vm
{
local hyper="$1"
local res="$2"
declare -g downtime_begin
[[ "$downtime_begin" = "" ]] && downtime_begin="$(date +%s)"
echo "DOWNTIME BEGIN $(date)"
# stop only the vm, keep intermediate mounts etc
remote "$hyper" "nodeagent vmstop $res"
}
function hook_resource_stop_rest
{
local hyper="$1"
local primary="$2"
local res="$3"
# stop the rest of the stack
remote "$hyper" "nodeagent stop $res"
local mnt="$(hook_get_mountpoint "$res")"
remote "$hyper" "mountpoint $mnt && { umount -f $mnt ; exit \$?; } || true"
hook_resource_stop "$primary" "$res"
}
function hook_resource_start
{
local host="$1"
local res="$2"
remote "$host" "marsadm wait-cluster"
remote "$host" "service clustermanager restart"
remote "$host" "marsadm primary $res"
remote "$host" "cm3 --stop $res; cm3 --start $res || { cm3 --stop $res; cm3 --start $res; } || false"
echo "DOWNTIME END $(date)"
declare -g downtime_begin
declare -g downtime_end="$(date +%s)"
echo "DOWNTIME END $(date) ($(( downtime_end - downtime_begin )) s)"
remote "$host" "if [[ -x /usr/sbin/nodeagent ]]; then /usr/sbin/nodeagent status; fi"
}
function hook_resource_check
{
local res="$1"
local timeout="${2:-10}"
local host="$res"
echo "Checking whether $host is running...."
while ! ping -c1 $host; do
if (( timeout-- <= 0 )); then
echo "HOST $host DOES NOT PING!"
return
fi
sleep 3
done
echo "Checking $host via check_progs ...."
sleep 15
remote "$host" "check_progs -cvi" 1 || echo "ATTENTION SOMETHING DOES NOT WORK AT $host"
}
###########################################
# Workarounds for firewalling (transitional => TBD)
workaround_firewall="${workaround_firewall:-1}"
function hook_prepare_hosts
{
local host_list="$1"
if (( workaround_firewall )); then
local host
for host in $host_list; do
remote "$host" "systemctl disable ui-firewalling.service || echo IGNORE"
remote "$host" "service ui-firewalling stop || /etc/init.d/firewalling stop"
done
fi
}
function hook_finish_hosts
{
local host_list="$1"
if (( workaround_firewall )); then
local host
for host in $host_list; do
remote "$host" "systemctl enable ui-firewalling.service || echo IGNORE"
remote "$host" "service ui-firewalling restart || /etc/init.d/firewalling restart"
done
fi
}
###########################################
# Workarounds for ssh between different clusters
ip_magic="${ip_magic:-1}"
do_split_cluster="${do_split_cluster:-1}"
function hook_merge_cluster
{
local source="$1"
local target="$2"
[[ "$source" = "" ]] && return
[[ "$target" = "" ]] && return
if (( ip_magic )); then
# This MAGIC may be needed when mutual icpu / istore
# ssh connects via hostnames are disallowed
# by some network firewall rules.
# Workaround by going down to the replication IPs.
local source_ip="$(remote "$source" "marsadm lowlevel-ls-host-ips" | grep "$source" | awk '{ print $2; }')"
echo "Peer '$source' has IP '$source_ip'"
source="$source_ip"
fi
remote "$target" "marsadm merge-cluster --ssh-port=24 $source"
}
function hook_split_cluster
{
local host_list="$1"
if (( do_split_cluster )); then
local ok=0
local host
for host in $host_list; do
sleep 5
if remote "$host" "marsadm split-cluster --ssh-port=24" 1; then
ok=1
break
fi
remote "$host" "marsadm wait-cluster" 1
done
if (( !ok )); then
fail "Please run 'marsadm split-cluster --ssh-port=24' by hand"
fi
fi
}
function hook_join_resource
{
local source="$1"
local target="$2"
local res="$3"
local dev="$4"
[[ "$source" = "" ]] && return
[[ "$target" = "" ]] && return
[[ "$res" = "" ]] && return
remote "$target" "marsadm join-resource --ssh-port=24 $res $dev"
}
###########################################
# General checks
needed_marsadm="${needed_marsadm:-2.1 1.1}"
needed_mars="${needed_mars:-0.1stable49 0.1abeta0 mars0.1abeta0}"
max_cluster_size="${max_cluster_size:-4}"
max_syncs="${max_syncs:-2}"
function check_needed
{
local type="$1"
local skip_prefix="$2"
local actual="$3"
local needed_list="$4"
echo "$type actual version : $actual"
echo "$type needed versions: $needed_list"
local needed
for needed in $needed_list; do
local pa="$(echo "$actual" | grep -o "^$skip_prefix")"
local pb="$(echo "$needed" | grep -o "^$skip_prefix")"
#echo "pa='$pa' pb='$pb'"
if [[ "$pa" != "$pb" ]]; then
#echo "prefix '$pa' != '$pb'"
continue
fi
local a="$(echo "$actual" | sed "s:^$skip_prefix::" | grep -o '[0-9.]\+' | head -1)"
local b="$(echo "$needed" | sed "s:^$skip_prefix::" | grep -o '[0-9.]\+' | head -1)"
#echo "skip_prefix='$skip_prefix' needed='$needed' a='$a' b='$b'"
if [[ "$a" = "" ]] || [[ "$b" = "" ]]; then
continue
fi
if [[ "$b" =~ \. ]] && [[ "${a##*.}" != "${b##*.}" ]]; then
continue
fi
if (( $(echo "$a >= $b" | bc) )); then
echo "$type actual version '$actual' matches '$needed'"
return
fi
done
fail "$type actual version '$actual' does not match one of '$needed_list'"
}
function hook_check_host
{
local host_list="$1"
local host
for host in $host_list; do
local marsadm_version="$(remote "$host" "marsadm --version" | grep -o 'Version: [0-9.]*' | awk '{ print $2; }')"
echo "Installed marsadm version at $host: '$marsadm_version'"
check_needed "marsadm" "[0-9]\." "$marsadm_version" "$needed_marsadm"
local mars_version="$(remote "$host" "cat /sys/module/mars/version" | awk '{ print $1; }')"
if [[ "$mars_version" = "" ]]; then
fail "MARS kernel module is not loaded at $host"
fi
check_needed "mars kernel module" "[a-z]*[0-9.]*[a-z]*" "$mars_version" "$needed_mars"
done
echo "Checking that max_cluster_size=$max_cluster_size will not be exceeded at $host_list"
local new_cluster_size="$(
for host in $host_list; do
remote "$host" "marsadm lowlevel-ls-host-ips" 2>/dev/null
done | sort -u | wc -l)"
if (( new_cluster_size < 2 )); then
fail "Implausible new cluster size $new_cluster_size"
fi
echo "New cluster size: $new_cluster_size"
if (( new_cluster_size > max_cluster_size )); then
fail "Cluster size limit $max_cluster_size will be exceeded, aborting."
fi
# Check that not too much syncs are already running
local actual_syncs=0
for host in $host_list; do
local count="$(remote "$host" "marsadm view-sync-rest all" | grep '^[0-9]' | grep -v '^0' | wc -l)"
echo "There are $count syncs running at $host"
(( actual_syncs += count ))
done
echo "Total number of syncs: $actual_syncs"
echo "Max number of syncs: $max_syncs"
if (( max_syncs >= 0 && actual_syncs > max_syncs )); then
fail "There are more than $max_syncs syncs running."
fi
}
function hook_describe_plugin
{
cat <<EOF
PLUGIN hooks-cm3
1&1 specfic plugin for dealing with the cm3 cluster manager
and its concrete operating enviroment (singleton instance).
Current maximum cluster size limit: $max_cluster_size
Maximum #syncs running before migration can start: $max_syncs
Following marsadm --version must be installed: $needed_marsadm
Following mars kernel modules must be loaded: $needed_mars
EOF
}
###########################################
# Mini infrastucture for access to clustermw
clustertool_host="${clustertool_host:-http://clustermw:3042}"
clustertool_user="${clustertool_user:-$(shopt -u nullglob; ls *.password | head -1 | cut -d. -f1)}" || fail "cannot find a password file *.password for clustermw"
clustertool_passwd="${clustertool_passwd:-$(cat $clustertool_user.password)}"
echo "Using clustermw username: '$clustertool_user'"
function clustertool
{
local op="${1:-GET}"
local path="${2:-/clusters}"
local content="$3"
local cmd="curl -s -u \"$clustertool_user:$clustertool_passwd\" -X \"$op\" \"$clustertool_host$path\""
[[ "$content" != "" ]] && cmd+=" -d '${content//\'/\'}'"
echo "$cmd" | sed -u 's/\(curl .*\)-u *[^ ]*/\1/' >> /dev/stderr
eval "$cmd" || fail "failed REST command '$cmd'"
}
function _get_cluster_name
{
local host="$1"
local url="/vms/$host.schlund.de"
[[ "$host" =~ icpu ]] && url="/nodes/$host.schlund.de"
[[ "$host" =~ istore ]] && url="/storagehosts/$host.schlund.de"
clustertool GET "$url" |\
grep -o "cluster[0-9]\+" |\
sort -u
}
function _get_segment
{
local cluster="$1"
local url="/clusters/$cluster"
clustertool GET "$url" |\
json_pp |\
grep '"segment"' |\
cut -d: -f2 |\
sed 's/[ ",]//g'
}
function _get_members
{
local cluster="$1"
local url="/clusters/$cluster"
clustertool GET "$url" |\
json_pp |\
grep -o '[-a-z0-9]\+.schlund.de' |\
cut -d. -f1 |\
sort -u |\
grep -v infong
}
function hook_get_flavour
{
local host="$1"
clustertool GET "/nodes/$host.schlund.de" |\
json_pp |\
grep flavour |\
grep -o '".*"' |\
sed 's/"//g' |\
sed 's/^.*: *//'
}
###########################################
# Migration operation: move cm3 config from old cluster to a new cluster
do_migrate="${do_migrate:-1}" # must be enabled; disable for dry-run testing
always_migrate="${always_migrate:-0}" # only enable for testing
# 0 = disabled
# 1 = only display the segment names
# 2 = check for equality
check_segments="${check_segments:-1}"
backup_dir="${backup_dir:-.}"
function _check_migrate
{
local source="$1"
local target="$2"
local res="$3"
[[ "$source" = "" ]] && return
[[ "$target" = "" ]] && return
[[ "$res" = "" ]] && return
local source_cluster="$(_get_cluster_name "$source")" || fail "cannot get source_cluster"
local target_cluster="$(_get_cluster_name "$target")" || fail "cannot get target_cluster"
echo "source_cluster='$source_cluster'"
echo "target_cluster='$target_cluster'"
if [[ "$source_cluster" != "$target_cluster" ]]; then
if (( check_segments )); then
# At the moment, cross-segment migrations won't work.
# TBD.
local source_segment="$(_get_segment "$source_cluster")" || fail "cannot get source_segment"
local target_segment="$(_get_segment "$target_cluster")" || fail "cannot get target_segment"
echo "source_segment='$source_segment'"
echo "target_segment='$target_segment'"
if (( check_segments > 1 )); then
[[ "$source_segment" = "" ]] && fail "cannot determine source segment"
[[ "$target_segment" = "" ]] && fail "cannot determine target segment"
[[ "$source_segment" != "$target_segment" ]] && fail "source_segment '$source_segment' != target_segment '$target_segment'"
fi
fi
fi
}
function hook_update_cm3_config
{
local host_list="$1"
echo "UPDATE cm3 config on " $host_list
echo ""
local update_host_list="$host_list"
local host
local i
for (( i = 0; i < 3; i++ )); do
local new_host_list=""
local status=0
sleep 10
for host in $update_host_list; do
remote "$host" "service clustermanager restart"
done
sleep 5
for host in $update_host_list; do
timeout_cmd "remote '$host' 'cm3 --update --force'"
local rc=$?
(( status |= rc ))
(( rc )) && new_host_list+=" $host"
done
(( !status )) && break
echo "RESTARTING cm3 update on $new_host_list"
update_host_list="$new_host_list"
done
sleep 5
for host in $host_list; do
remote "$host" "service clustermanager restart"
done
sleep 3
for host in $host_list; do
remote "$host" "update-motd || echo IGNORE"
done
}
function hook_migrate_cm3_config
{
local source="$1"
local target="$2"
local res="$3"
[[ "$source" = "" ]] && return
[[ "$target" = "" ]] && return
[[ "$res" = "" ]] && return
local source_cluster="$(_get_cluster_name "$source")" || fail "cannot get source_cluster"
local target_cluster="$(_get_cluster_name "$target")" || fail "cannot get target_cluster"
if (( always_migrate )) || [[ "$source_cluster" != "$target_cluster" ]]; then
echo "Moving config from cluster '$source_cluster' to cluster '$target_cluster'"
local backup=""
if [[ "$backup_dir" != "" ]]; then
local backup="$backup_dir/json-backup.$start_stamp"
mkdir -p $backup
fi
local status_url="/vms/$res.schlund.de"
clustertool GET "$status_url" |\
log "$backup" "$res.old.raw.json" |\
json_pp |\
log "$backup" "$res.old.pp.json"
local old_url="/clusters/$source_cluster/vms/$res.schlund.de"
local new_url="/clusters/$target_cluster/vms/$res.schlund.de"
echo clustertool DELETE "$old_url"
(( do_migrate )) && clustertool DELETE "$old_url"
echo clustertool PUT "$new_url"
(( do_migrate )) && clustertool PUT "$new_url"
clustertool GET "$status_url" |\
log "$backup" "$res.new.raw.json" |\
json_pp |\
log "$backup" "$res.new.pp.json"
echo "--------------------- diff old => new ------------------"
diff -ui $backup/$res.old.pp.json $backup/$res.new.pp.json
echo ""
clustertool PUT "/clusters/$source_cluster/properties/CLUSTERCONF_SERIAL"
clustertool PUT "/clusters/$target_cluster/properties/CLUSTERCONF_SERIAL"
echo ""
section "Update cm3 configs"
# Determine all hosts where a clustermanagerd _should_ be running.
# Unfortunately, parsing the "storagehost" part will not work for standalone icpus.
# Use a provisionary heuristics here, based on naming conventions.
local total_list="$(
{
echo "$source"
echo "$target"
for cluster in $source_cluster $target_cluster; do
if ! _get_members $cluster | grep istore; then
_get_members $cluster
fi
done
} |\
sort -u)"
hook_update_cm3_config "$total_list"
else
echo "Source and target clusters are equal: '$source_cluster'"
echo "Nothing to do."
fi
}
function hook_check_migrate
{
local source="$1"
local target="$2"
local res="$3"
_check_migrate "$source" "$target" "$res"
}
function hook_determine_old_replicas
{
local primary="$1"
local res="$2"
local primary_cluster="$(_get_cluster_name "$primary")"
local secondary_list="$(remote "$primary" "marsadm view-resource-members $res" | { grep -v "^$primary$" || true; })" || fail "cannot determine secondary_list"
local host
for host in $secondary_list; do
local cluster="$(_get_cluster_name "$host")"
if [[ "$cluster" != "$primary_cluster" ]]; then
echo "FOREIGN:$host"
fi
done
}
function hook_determine_new_replicas
{
local primary="$1"
local res="$2"
local primary_cluster="$(_get_cluster_name "$primary")"
local secondary_list="$(remote "$primary" "marsadm view-resource-members $res" | { grep -v "^$primary$" || true; })" || fail "cannot determine secondary_list"
local host
for host in $primary $secondary_list; do
local cluster="$(_get_cluster_name "$host")"
if [[ "$cluster" = "$primary_cluster" ]]; then
echo "FOREIGN:$host"
fi
done
}
###########################################
# Local quota transfer for LXC containers.
# Problem: kuid_t implies that dump and restore is USER_NS specific,
# at least with currently used kernel + userspace tool versions.
function _get_kunden_dev
{
local lv_name="$1"
remote "$lv_name" "df /kunden/homepages/ | grep '^/dev' | head -1 | awk '{ print \$1; }'"
}
function hook_save_local_quota
{
local hyper="$1"
local lv_name="$2"
if (( do_quota >= 2 )); then
section "Local xfs quota dump"
local dumpfile="$xfs_dump_dir/xfs_dump.local.$hyper.$lv_name"
local local_dev="/dev/mars/$lv_name"
local store="$(get_store "$lv_name")"
echo "store='$store'"
if [[ "$store" != "$hyper" ]]; then
local_dev="$(_get_kunden_dev "$lv_name")"
if ! [[ "$local_dev" =~ /dev/ ]]; then
# last resort fallback, try to preserve the data for hand repair
echo "WARNING: local device is '$local_dev'" >> /dev/stderr
echo "WARNING: cannot determine local device for /kunden/homepages on $lv_name" >> /dev/stderr
local_dev="/kunden/homepages"
fi
fi
echo "local_dev='$local_dev'"
remote "$lv_name" "$xfs_dump $local_dev" > $dumpfile
ls -l $dumpfile
wc -l $dumpfile
fi
}
function hook_restore_local_quota
{
local hyper="$1"
local lv_name="$2"
if (( do_quota >= 2 )); then
local dumpfile="$xfs_dump_dir/xfs_dump.local.$hyper.$lv_name"
section "Local xfs quota restore"
if [[ -s "$dumpfile" ]]; then
local max_rounds=10
while ! ping -c 1 "$lv_name"; do
(( max_rounds-- < 0 )) && fail "host $lv_name does not ping"
sleep 10
done
sleep 10
local local_dev="/dev/mars/$lv_name"
local store="$(get_store "$lv_name")"
echo "store='$store'"
if [[ "$store" != "$hyper" ]]; then
local_dev="$(_get_kunden_dev "$lv_name")"
if ! [[ "$local_dev" =~ /dev/ ]]; then
echo "Sorry, cannot determine local device for /kunden/homepages on $lv_name" >> /dev/stderr
return 0
fi
fi
echo "local_dev='$local_dev'"
{
echo "fs = $local_dev"
grep -v "^fs =" < $dumpfile
} > $dumpfile.new
ls -l $dumpfile.new
wc -l $dumpfile.new
local max_rounds=10
while ! remote "$lv_name" "$xfs_restore $local_dev" 1 < $dumpfile.new; do
(( max_rounds-- < 0 )) && fail "host $lv_name does not ping"
sleep 10
done
else
echo "LOCAL $lv_name QUOTA IS EMPTY"
fi
fi
}
###########################################
# Hooks for shrinking
iqn_base="${iqn_base:-iqn.2000-01.info.test:test}"
iet_type="${iet_type:-blockio}"
iscsi_eth="${iscsi_eth:-eth1}"
iscsi_tid="${iscsi_tid:-4711}"
function new_tid
{
local iqn="$1"
local store="$2"
declare -g iscsi_tid
local old_tids="$(remote "$store" "cat /proc/net/iet/volume /proc/net/iet/session" | grep -o 'tid:[0-9]\+' | cut -d: -f2 | sort -u)"
echo "old tids: " $old_tids >> /dev/stderr
while echo $old_tids | grep "$iscsi_tid" 1>&2; do
(( iscsi_tid++ ))
done
echo "iSCSI IQN '$iqn' has new tid '$iscsi_tid'" >> /dev/stderr
echo "$iscsi_tid"
}
function hook_disconnect
{
local store="$1"
local res="$2"
local iqn="$iqn_base.$res.tmp"
# safeguarding: retrieve any matching runtime session
local hyper
for hyper in $(remote "$store" "grep -A1 'name:$iqn' < /proc/net/iet/session | grep 'initiator:' | grep -o 'icpu[0-9]\+'"); do
remote "$hyper" "iscsiadm -m node -T $iqn -u || echo IGNORE iSCSI initiator logout"
done
# safeguarding: retrieve any matching tid
local tid
for tid in $(remote "$store" "grep 'name:$iqn' < /proc/net/iet/volume | cut -d' ' -f1 | cut -d: -f2"); do
echo "KILLING old tid '$tid' for iqn '$iqn' on '$store'"
remote "$store" "ietadm --op delete --tid=$tid || echo IGNORE iSCSI target deletion"
done
}
function hook_connect
{
local store="$1"
local hyper="$2"
local res="$3"
# for safety, kill any old session
hook_disconnect "$store" "$res"
local vg_name="$(get_vg "$store")" || fail "cannot determine VG for host '$store'"
local dev="/dev/$vg_name/$res"
local iqn="$iqn_base.$res.tmp"
local iscsi_ip="$(remote "$store" "ifconfig $iscsi_eth" | grep "inet addr:" | cut -d: -f2 | awk '{print $1;}')"
echo "using iscsi IP '$iscsi_ip'"
# saftey check
remote "$hyper" "ping -c1 $iscsi_ip"
# step 1: setup stone-aged IET on storage node
local tid="$(new_tid "$iqn" "$store")"
remote "$store" "ietadm --op new --tid=$tid --params=Name=$iqn"
remote "$store" "ietadm --op new --tid=$tid --lun=0 --params=Path=$dev"
sleep 2
# step2: import iSCSI on hypervisor
remote "$hyper" "iscsiadm -m discovery -p $iscsi_ip --type sendtargets"
tmp_list="/tmp/devlist.$$"
remote "$hyper" "ls /dev/sd?" > $tmp_list
remote "$hyper" "iscsiadm -m node -p $iscsi_ip -T $iqn -l"
while true; do
sleep 3
local new_dev="$(remote "$hyper" "ls /dev/sd?" | diff -u $tmp_list - | grep '^+/' | cut -c2-)"
[[ -n "$new_dev" ]] && break
done
rm -f $tmp_list
echo "NEW_DEV:$new_dev"
}
###########################################
# Hooks for extending of XFS
function hook_extend_iscsi
{
local hyper="$1"
remote "$hyper" "iscsiadm -m session -R"
}