#!/bin/bash
+# Copyright (C) The Arvados Authors. All rights reserved.
+#
+# SPDX-License-Identifier: AGPL-3.0
+
+set -e
+
DEBUG=0
+UNMANAGED=0
+SSH_PORT=22
+PUPPET_CONCURRENCY=5
+
+read -d] -r SCOPES <<EOF
+--scopes
+'["GET /arvados/v1/virtual_machines",\n
+"GET /arvados/v1/keep_services",\n
+"GET /arvados/v1/keep_services/",\n
+"GET /arvados/v1/groups",\n
+"GET /arvados/v1/groups/",\n
+"GET /arvados/v1/links",\n
+"GET /arvados/v1/collections",\n
+"POST /arvados/v1/collections",\n
+"POST /arvados/v1/links",\n
+"GET /arvados/v1/users/current",\n
+"POST /arvados/v1/users/current",\n
+"GET /arvados/v1/jobs",\n
+"POST /arvados/v1/jobs",\n
+"GET /arvados/v1/pipeline_instances",\n
+"POST /arvados/v1/pipeline_instances",\n
+"PUT /arvados/v1/pipeline_instances/",\n
+"GET /arvados/v1/collections/",\n
+"POST /arvados/v1/collections/",\n
+"GET /arvados/v1/logs"]'
+EOF
function usage {
echo >&2
echo >&2 " <identifier> Arvados cluster name"
echo >&2
echo >&2 "$0 options:"
+ echo >&2 " -n, --node <node> Single machine to deploy, use fqdn, optional"
+ echo >&2 " -p, --port <ssh port> SSH port to use (default 22)"
+ echo >&2 " -c, --concurrency <max> Maximum concurrency for puppet runs (default 5)"
+ echo >&2 " -u, --unmanaged Deploy to unmanaged node/cluster"
echo >&2 " -d, --debug Enable debug output"
echo >&2 " -h, --help Display this help and exit"
echo >&2
echo >&2 "Note: this script requires an arvados token created with these permissions:"
echo >&2 ' arv api_client_authorization create_system_auth \'
- echo >&2 ' --scopes "[\"GET /arvados/v1/virtual_machines\",'
- echo >&2 ' \"GET /arvados/v1/keep_services\",'
- echo >&2 ' \"GET /arvados/v1/keep_services/\",'
- echo >&2 ' \"GET /arvados/v1/groups\",'
- echo >&2 ' \"GET /arvados/v1/groups/\",'
- echo >&2 ' \"GET /arvados/v1/links\",'
- echo >&2 ' \"GET /arvados/v1/collections\",'
- echo >&2 ' \"POST /arvados/v1/collections\",'
- echo >&2 ' \"POST /arvados/v1/links\"]"'
+ echo -e $SCOPES"]'" >&2
echo >&2
}
+
# NOTE: This requires GNU getopt (part of the util-linux package on Debian-based distros).
-TEMP=`getopt -o hd \
- --long help,debug \
+TEMP=`getopt -o hudp:c:n: \
+ --long help,unmanaged,debug,port:,concurrency:,node: \
-n "$0" -- "$@"`
if [ $? != 0 ] ; then echo "Use -h for help"; exit 1 ; fi
while [ $# -ge 1 ]
do
case $1 in
+ -n | --node)
+ NODE="$2"; shift 2
+ ;;
+ -p | --port)
+ SSH_PORT="$2"; shift 2
+ ;;
+ -c | --concurrency)
+ PUPPET_CONCURRENCY="$2"; shift 2
+ ;;
+ -u | --unmanaged)
+ UNMANAGED=1
+ shift
+ ;;
-d | --debug)
DEBUG=1
+ set -x
shift
;;
--)
COLUMNS=80
+PUPPET_AGENT='
+if [[ -e "/usr/local/rvm/scripts/rvm" ]]; then
+ source /usr/local/rvm/scripts/rvm
+ __rvm_unload
+fi
+now() { date +%s; }
+let endtime="$(now) + 600"
+while [ "$endtime" -gt "$(now)" ]; do
+ puppet agent --test --detailed-exitcodes
+ agent_exitcode=$?
+ if [ 0 = "$agent_exitcode" ] || [ 2 = "$agent_exitcode" ]; then
+ break
+ else
+ sleep 10s
+ fi
+done
+exit ${agent_exitcode:-99}
+'
+
+APT_AGENT='
+now() { date +%s; }
+let endtime="$(now) + 600"
+while [ "$endtime" -gt "$(now)" ]; do
+ apt-get update
+ DEBIAN_FRONTEND=noninteractive apt-get -y upgrade
+ apt_exitcode=$?
+ if [ 0 = "$apt_exitcode" ]; then
+ break
+ else
+ sleep 10s
+ fi
+done
+exit ${apt_exitcode:-99}
+'
+
title () {
date=`date +'%Y-%m-%d %H:%M:%S'`
printf "$date $1\n"
}
-function run_puppet() {
+function update_node() {
+ if [[ $UNMANAGED -ne 0 ]]; then
+ run_apt $@
+ else
+ run_puppet $@
+ fi
+}
+
+function run_apt() {
node=$1
- return_var=$2
- title "Running puppet on $node"
+ title "Running apt on $node"
+ sleep $[ $RANDOM / 6000 ].$[ $RANDOM / 1000 ]
TMP_FILE=`mktemp`
if [[ "$DEBUG" != "0" ]]; then
- ssh -t -p2222 -o "StrictHostKeyChecking no" -o "ConnectTimeout 5" root@$node -C "/usr/bin/puppet agent -t" | tee $TMP_FILE
+ ssh -t -p$SSH_PORT -o "StrictHostKeyChecking no" -o "ConnectTimeout 5" root@$node -C bash -c "'$APT_AGENT'" 2>&1 | sed 's/^/['"${node}"'] /' | tee $TMP_FILE
else
- ssh -t -p2222 -o "StrictHostKeyChecking no" -o "ConnectTimeout 5" root@$node -C "/usr/bin/puppet agent -t" > $TMP_FILE 2>&1
+ ssh -t -p$SSH_PORT -o "StrictHostKeyChecking no" -o "ConnectTimeout 5" root@$node -C bash -c "'$APT_AGENT'" 2>&1 | sed 's/^/['"${node}"'] /' > $TMP_FILE 2>&1
fi
- ECODE=$?
+ ECODE=${PIPESTATUS[0]}
RESULT=$(cat $TMP_FILE)
- if [[ "$ECODE" != "255" && ! ("$RESULT" =~ 'already in progress') && "$ECODE" != "2" && "$ECODE" != "0" ]]; then
+ if [[ "$ECODE" != "255" && "$ECODE" != "0" ]]; then
# Ssh exits 255 if the connection timed out. Just ignore that.
- # Puppet exits 2 if there are changes. For real!
- # Puppet prints 'Notice: Run of Puppet configuration client already in progress' if another puppet process
- # was already running
- echo "ERROR running puppet on $node: exit code $ECODE"
+ echo "ERROR running apt on $node: exit code $ECODE"
if [[ "$DEBUG" == "0" ]]; then
title "Command output follows:"
echo $RESULT
title "Connection timed out"
ECODE=0
fi
- if [[ "$ECODE" == "2" ]]; then
- ECODE=0
+
+ if [[ "$ECODE" == "0" ]]; then
+ rm -f $TMP_FILE
+ title "$node successfully updated"
+ else
+ title "$node exit code: $ECODE see $TMP_FILE for details"
fi
- rm -f $TMP_FILE
- eval "$return_var=$ECODE"
}
-function run_command() {
+function run_puppet() {
node=$1
- return_var=$2
- command=$3
- title "Running '$command' on $node"
+ title "Running puppet on $node"
+ sleep $[ $RANDOM / 6000 ].$[ $RANDOM / 1000 ]
TMP_FILE=`mktemp`
if [[ "$DEBUG" != "0" ]]; then
- ssh -t -p2222 -o "StrictHostKeyChecking no" -o "ConnectTimeout 5" root@$node -C "$command" | tee $TMP_FILE
+ ssh -t -p$SSH_PORT -o "StrictHostKeyChecking no" -o "ConnectTimeout 5" root@$node -C bash -c "'$PUPPET_AGENT'" 2>&1 | sed 's/^/['"${node}"'] /' | tee $TMP_FILE
else
- ssh -t -p2222 -o "StrictHostKeyChecking no" -o "ConnectTimeout 5" root@$node -C "$command" > $TMP_FILE 2>&1
+ ssh -t -p$SSH_PORT -o "StrictHostKeyChecking no" -o "ConnectTimeout 5" root@$node -C bash -c "'$PUPPET_AGENT'" 2>&1 | sed 's/^/['"${node}"'] /' > $TMP_FILE 2>&1
fi
- ECODE=$?
+ ECODE=${PIPESTATUS[0]}
RESULT=$(cat $TMP_FILE)
- if [[ "$ECODE" != "255" && "$ECODE" != "0" ]]; then
- # Ssh exists 255 if the connection timed out. Just ignore that, it's possible that this node is
- # a shell node that is down.
- title "ERROR running command on $node: exit code $ECODE"
+ if [[ "$ECODE" != "255" && ! ("$RESULT" =~ 'already in progress') && "$ECODE" != "2" && "$ECODE" != "0" ]]; then
+ # Ssh exits 255 if the connection timed out. Just ignore that.
+ # Puppet exits 2 if there are changes. For real!
+ # Puppet prints 'Notice: Run of Puppet configuration client already in progress' if another puppet process
+ # was already running
+ echo "ERROR running puppet on $node: exit code $ECODE"
if [[ "$DEBUG" == "0" ]]; then
title "Command output follows:"
echo $RESULT
title "Connection timed out"
ECODE=0
fi
- rm -f $TMP_FILE
- eval "$return_var=$ECODE"
+ if [[ "$ECODE" == "2" ]]; then
+ ECODE=0
+ fi
+
+ if [[ "$ECODE" == "0" ]]; then
+ rm -f $TMP_FILE
+ echo $node successfully updated
+ else
+ echo $node exit code: $ECODE see $TMP_FILE for details
+ fi
}
-title "Updating API server"
-SUM_ECODE=0
-run_puppet $IDENTIFIER.arvadosapi.com ECODE
-SUM_ECODE=$(($SUM_ECODE + $ECODE))
-run_command $IDENTIFIER.arvadosapi.com ECODE "/usr/local/bin/arvados-api-server-upgrade.sh"
-SUM_ECODE=$(($SUM_ECODE + $ECODE))
-run_command $IDENTIFIER.arvadosapi.com ECODE "dpkg -L arvados-mailchimp-plugin 2>/dev/null && apt-get install arvados-mailchimp-plugin --reinstall || echo"
-SUM_ECODE=$(($SUM_ECODE + $ECODE))
-
-if [[ "$SUM_ECODE" != "0" ]]; then
- title "ERROR: Updating API server FAILED"
- EXITCODE=$(($EXITCODE + $SUM_ECODE))
- exit $EXITCODE
+if [[ "$NODE" == "" ]] || [[ "$NODE" == "$IDENTIFIER.arvadosapi.com" ]]; then
+ title "Updating API server"
+ SUM_ECODE=0
+ update_node $IDENTIFIER.arvadosapi.com ECODE
+ SUM_ECODE=$(($SUM_ECODE + $ECODE))
+
+ if [[ "$SUM_ECODE" != "0" ]]; then
+ title "ERROR: Updating API server FAILED"
+ EXITCODE=$(($EXITCODE + $SUM_ECODE))
+ exit $EXITCODE
+ fi
+fi
+
+if [[ "$NODE" == "$IDENTIFIER.arvadosapi.com" ]]; then
+ # we are done
+ exit 0
fi
title "Loading ARVADOS_API_HOST and ARVADOS_API_TOKEN"
exit 1
fi
-title "Locating Arvados Standard Docker images project"
-
-JSON_FILTER="[[\"name\", \"=\", \"Arvados Standard Docker Images\"], [\"owner_uuid\", \"=\", \"$IDENTIFIER-tpzed-000000000000000\"]]"
-DOCKER_IMAGES_PROJECT=`ARVADOS_API_HOST=$ARVADOS_API_HOST ARVADOS_API_TOKEN=$ARVADOS_API_TOKEN arv --format=uuid group list --filters="$JSON_FILTER"`
-
-if [[ "$DOCKER_IMAGES_PROJECT" == "" ]]; then
- title "Warning: Arvados Standard Docker Images project not found. Creating it."
-
- DOCKER_IMAGES_PROJECT=`ARVADOS_API_HOST=$ARVADOS_API_HOST ARVADOS_API_TOKEN=$ARVADOS_API_TOKEN arv --format=uuid group create --group "{\"owner_uuid\":\"$IDENTIFIER-tpzed-000000000000000\", \"name\":\"Arvados Standard Docker Images\", \"group_class\":\"project\"}"`
- ARVADOS_API_HOST=$ARVADOS_API_HOST ARVADOS_API_TOKEN=$ARVADOS_API_TOKEN arv link create --link "{\"tail_uuid\":\"$IDENTIFIER-j7d0g-fffffffffffffff\", \"head_uuid\":\"$DOCKER_IMAGES_PROJECT\", \"link_class\":\"permission\", \"name\":\"can_read\" }"
- if [[ "$?" != "0" ]]; then
- title "ERROR: could not create standard Docker images project Please create it, cf. http://doc.arvados.org/install/create-standard-objects.html"
- exit 1
- fi
+title "Gathering list of nodes"
+start_nodes="workbench"
+if [[ "$IDENTIFIER" != "ce8i5" ]] && [[ "$IDENTIFIER" != "tordo" ]]; then
+ start_nodes="$start_nodes manage switchyard"
fi
-
-title "Found Arvados Standard Docker Images project with uuid $DOCKER_IMAGES_PROJECT"
-GIT_COMMIT=`ssh -o "StrictHostKeyChecking no" $IDENTIFIER cat /usr/local/arvados/src/git-commit.version`
-
-if [[ "$?" != "0" ]] || [[ "$GIT_COMMIT" == "" ]]; then
- title "ERROR: unable to get arvados/jobs Docker image git revision"
- exit 1
-else
- title "Found git commit for arvados/jobs Docker image: $GIT_COMMIT"
-fi
-
-run_command shell.$IDENTIFIER ECODE "ARVADOS_API_HOST=$ARVADOS_API_HOST ARVADOS_API_TOKEN=$ARVADOS_API_TOKEN /usr/local/rvm/bin/rvm-exec default arv keep docker" |grep -q $GIT_COMMIT
-
-if [[ "$?" == "0" ]]; then
- title "Found latest arvados/jobs Docker image, nothing to upload"
-else
- title "Installing latest arvados/jobs Docker image"
- ssh -o "StrictHostKeyChecking no" shell.$IDENTIFIER "ARVADOS_API_HOST=$ARVADOS_API_HOST ARVADOS_API_TOKEN=$ARVADOS_API_TOKEN /usr/local/rvm/bin/rvm-exec default arv keep docker --pull --project-uuid=$DOCKER_IMAGES_PROJECT arvados/jobs $GIT_COMMIT"
- if [[ "$?" -ne 0 ]]; then
- title "'git pull' failed exiting..."
- exit 1
- fi
-fi
-
-title "Gathering list of shell and Keep nodes"
SHELL_NODES=`ARVADOS_API_HOST=$ARVADOS_API_HOST ARVADOS_API_TOKEN=$ARVADOS_API_TOKEN arv virtual_machine list |jq .items[].hostname -r`
KEEP_NODES=`ARVADOS_API_HOST=$ARVADOS_API_HOST ARVADOS_API_TOKEN=$ARVADOS_API_TOKEN arv keep_service list |jq .items[].service_host -r`
+SHELL_NODE_FOR_ARV_KEEPDOCKER="shell.$IDENTIFIER"
+start_nodes="$start_nodes $SHELL_NODES $KEEP_NODES"
-title "Updating workbench"
-SUM_ECODE=0
-if [[ `host workbench.$ARVADOS_API_HOST |cut -f4 -d' '` != `host $ARVADOS_API_HOST |cut -f4 -d' '` ]]; then
- # Workbench runs on a separate host. We need to run puppet there too.
- run_puppet workbench.$IDENTIFIER ECODE
- SUM_ECODE=$(($SUM_ECODE + $ECODE))
-fi
-
-run_command workbench.$IDENTIFIER ECODE "/usr/local/bin/arvados-workbench-upgrade.sh"
-SUM_ECODE=$(($SUM_ECODE + $ECODE))
-
-if [[ "$SUM_ECODE" != "0" ]]; then
- title "ERROR: Updating workbench FAILED"
- EXITCODE=$(($EXITCODE + $SUM_ECODE))
- exit $EXITCODE
-fi
-
-for n in manage switchyard $SHELL_NODES $KEEP_NODES; do
+nodes=""
+for n in $start_nodes; do
ECODE=0
if [[ $n =~ $ARVADOS_API_HOST$ ]]; then
# e.g. keep.qr1hi.arvadosapi.com
# e.g. shell
node=$n.$ARVADOS_API_HOST
fi
+ if [[ "$NODE" == "" ]] || [[ "$NODE" == "$node" ]]; then
+ # e.g. keep.qr1hi
+ nodes="$nodes ${node%.arvadosapi.com}"
+ fi
+done
- # e.g. keep.qr1hi
- node=${node%.arvadosapi.com}
+if [[ "$nodes" != "" ]]; then
+ ## at this point nodes should be an array containing
+ ## manage.qr1hi, keep.qr1hi, etc
+ ## that should be defined in the .ssh/config file
+ title "Updating in parallel:$nodes"
+ export -f update_node
+ export -f run_puppet
+ export -f run_apt
+ export -f title
+ export SSH_PORT
+ export PUPPET_AGENT
+ export APT_AGENT
+ export UNMANAGED
+ echo $nodes|xargs -d " " -n 1 -P $PUPPET_CONCURRENCY -I {} bash -c "update_node {}"
+fi
- title "Updating $node"
- run_puppet $node ECODE
- if [[ "$ECODE" != "0" ]]; then
- title "ERROR: Updating $node node FAILED: exit code $ECODE"
- EXITCODE=$(($EXITCODE + $ECODE))
- exit $EXITCODE
+if [[ "$NODE" == "" ]]; then
+ title "Locating Arvados Standard Docker images project"
+
+ JSON_FILTER="[[\"name\", \"=\", \"Arvados Standard Docker Images\"], [\"owner_uuid\", \"=\", \"$IDENTIFIER-tpzed-000000000000000\"]]"
+ DOCKER_IMAGES_PROJECT=`ARVADOS_API_HOST=$ARVADOS_API_HOST ARVADOS_API_TOKEN=$ARVADOS_API_TOKEN arv --format=uuid group list --filters="$JSON_FILTER"`
+
+ if [[ "$DOCKER_IMAGES_PROJECT" == "" ]]; then
+ title "Warning: Arvados Standard Docker Images project not found. Creating it."
+
+ DOCKER_IMAGES_PROJECT=`ARVADOS_API_HOST=$ARVADOS_API_HOST ARVADOS_API_TOKEN=$ARVADOS_API_TOKEN arv --format=uuid group create --group "{\"owner_uuid\":\"$IDENTIFIER-tpzed-000000000000000\", \"name\":\"Arvados Standard Docker Images\", \"group_class\":\"project\"}"`
+ ARVADOS_API_HOST=$ARVADOS_API_HOST ARVADOS_API_TOKEN=$ARVADOS_API_TOKEN arv link create --link "{\"tail_uuid\":\"$IDENTIFIER-j7d0g-fffffffffffffff\", \"head_uuid\":\"$DOCKER_IMAGES_PROJECT\", \"link_class\":\"permission\", \"name\":\"can_read\" }"
+ if [[ "$?" != "0" ]]; then
+ title "ERROR: could not create standard Docker images project Please create it, cf. http://doc.arvados.org/install/create-standard-objects.html"
+ exit 1
+ fi
fi
-done
+
+ title "Found Arvados Standard Docker Images project with uuid $DOCKER_IMAGES_PROJECT"
+
+ if [[ "$SHELL_NODE_FOR_ARV_KEEPDOCKER" == "" ]]; then
+ VERSION=`ssh -t -p$SSH_PORT -o "StrictHostKeyChecking no" -o "ConnectTimeout 125" -o "LogLevel QUIET" $IDENTIFIER apt-cache policy python3-arvados-cwl-runner|grep Candidate`
+ VERSION=`echo $VERSION|cut -f2 -d' '|cut -f1 -d-`
+
+ if [[ "$?" != "0" ]] || [[ "$VERSION" == "" ]]; then
+ title "ERROR: unable to get python3-arvados-cwl-runner version"
+ exit 1
+ else
+ title "Found version for python3-arvados-cwl-runner: $VERSION"
+ fi
+
+ set +e
+ CLEAN_VERSION=`echo $VERSION | sed s/~dev/.dev/g | sed s/~rc/rc/g`
+ ARVADOS_API_HOST=$ARVADOS_API_HOST ARVADOS_API_TOKEN=$ARVADOS_API_TOKEN arv-keepdocker |grep -qP "arvados/jobs +$CLEAN_VERSION "
+ if [[ $? -eq 0 ]]; then
+ set -e
+ title "Found arvados/jobs Docker image version $CLEAN_VERSION, nothing to upload"
+ else
+ set -e
+ title "Installing arvados/jobs Docker image version $CLEAN_VERSION"
+ ARVADOS_API_HOST=$ARVADOS_API_HOST ARVADOS_API_TOKEN=$ARVADOS_API_TOKEN arv-keepdocker --pull --project-uuid=$DOCKER_IMAGES_PROJECT arvados/jobs $CLEAN_VERSION
+ if [[ $? -ne 0 ]]; then
+ title "'arv-keepdocker' failed..."
+ exit 1
+ fi
+ fi
+ else
+ VERSION=`ssh -t -p$SSH_PORT -o "StrictHostKeyChecking no" -o "ConnectTimeout 125" -o "LogLevel QUIET" $SHELL_NODE_FOR_ARV_KEEPDOCKER apt-cache policy python3-arvados-cwl-runner|grep Candidate`
+ VERSION=`echo $VERSION|cut -f2 -d' '|cut -f1 -d-`
+
+ if [[ "$?" != "0" ]] || [[ "$VERSION" == "" ]]; then
+ title "ERROR: unable to get python3-arvados-cwl-runner version"
+ exit 1
+ else
+ title "Found version for python3-arvados-cwl-runner: $VERSION"
+ fi
+
+ set +e
+ CLEAN_VERSION=`echo $VERSION | sed s/~dev/.dev/g | sed s/~rc/rc/g`
+ ssh -t -p$SSH_PORT -o "StrictHostKeyChecking no" -o "ConnectTimeout 125" -o "LogLevel QUIET" $SHELL_NODE_FOR_ARV_KEEPDOCKER "ARVADOS_API_HOST=$ARVADOS_API_HOST ARVADOS_API_TOKEN=$ARVADOS_API_TOKEN arv-keepdocker" |grep -qP "arvados/jobs +$CLEAN_VERSION "
+ if [[ $? -eq 0 ]]; then
+ set -e
+ title "Found arvados/jobs Docker image version $CLEAN_VERSION, nothing to upload"
+ else
+ set -e
+ title "Installing arvados/jobs Docker image version $CLEAN_VERSION"
+ ssh -t -p$SSH_PORT -o "StrictHostKeyChecking no" -o "ConnectTimeout 125" -o "LogLevel QUIET" $SHELL_NODE_FOR_ARV_KEEPDOCKER "ARVADOS_API_HOST=$ARVADOS_API_HOST ARVADOS_API_TOKEN=$ARVADOS_API_TOKEN arv-keepdocker --pull --project-uuid=$DOCKER_IMAGES_PROJECT arvados/jobs $CLEAN_VERSION"
+ if [[ $? -ne 0 ]]; then
+ title "'arv-keepdocker' failed..."
+ exit 1
+ fi
+ fi
+ fi
+fi
projects / arvados-dev.git / blobdiff