Improve error handling and reporting in run-cwl-test.sh script. no issue #
[arvados-dev.git] / jenkins / run-cwl-test.sh
1 #!/bin/bash
2
3 set -o pipefail
4
5 DEBUG=0
6 SSH_PORT=22
7 ACCT=ci
8
9 function usage {
10     echo >&2
11     echo >&2 "usage: $0 [options] <identifier>"
12     echo >&2
13     echo >&2 "   <identifier>                 Arvados cluster name"
14     echo >&2
15     echo >&2 "$0 options:"
16     echo >&2 "  -p, --port <ssh port>         SSH port to use (default 22)"
17     echo >&2 "      --acct <username>         Account to log in with"
18     echo >&2 "  -d, --debug                   Enable debug output"
19     echo >&2 "  -h, --help                    Display this help and exit"
20     echo >&2
21     echo >&2 " Required scope for the token used to run the tests:"
22     echo >&2
23     echo >&2 " arv api_client_authorization create_system_auth     --scopes "
24     echo >&2 "[\"GET /arvados/v1/virtual_machines\","
25     echo >&2 "\"GET /arvados/v1/keep_services\","
26     echo >&2 "\"GET /arvados/v1/keep_services/\","
27     echo >&2 "\"GET /arvados/v1/groups\","
28     echo >&2 "\"GET /arvados/v1/groups/\","
29     echo >&2 "\"GET /arvados/v1/links\","
30     echo >&2 "\"GET /arvados/v1/collections\","
31     echo >&2 "\"POST /arvados/v1/collections\","
32     echo >&2 "\"POST /arvados/v1/links\","
33     echo >&2 "\"GET /arvados/v1/users/current\","
34     echo >&2 "\"POST /arvados/v1/users/current\","
35     echo >&2 "\"GET /arvados/v1/jobs\","
36     echo >&2 "\"POST /arvados/v1/jobs\","
37     echo >&2 "\"GET /arvados/v1/pipeline_instances\","
38     echo >&2 "\"GET /arvados/v1/pipeline_instances/\","
39     echo >&2 "\"POST /arvados/v1/pipeline_instances\","
40     echo >&2 "\"GET /arvados/v1/collections/\","
41     echo >&2 "\"POST /arvados/v1/collections/\","
42     echo >&2 "\"GET /arvados/v1/logs\" ]"
43     echo >&2
44 }
45
46 # NOTE: This requires GNU getopt (part of the util-linux package on Debian-based distros).
47 TEMP=`getopt -o hdp: \
48     --long help,debug,port:,acct: \
49     -n "$0" -- "$@"`
50
51 if [ $? != 0 ] ; then echo "Use -h for help"; exit 1 ; fi
52 # Note the quotes around `$TEMP': they are essential!
53 eval set -- "$TEMP"
54
55 while [ $# -ge 1 ]
56 do
57     case $1 in
58         -p | --port)
59             SSH_PORT="$2"; shift 2
60             ;;
61         --acct)
62             ACCT="$2"; shift 2
63             ;;
64         -d | --debug)
65             DEBUG=1
66             shift
67             ;;
68         --)
69             shift
70             break
71             ;;
72         *)
73             usage
74             exit 1
75             ;;
76     esac
77 done
78
79 IDENTIFIER=$1
80
81 if [[ "$IDENTIFIER" == '' ]]; then
82   usage
83   exit 1
84 fi
85
86 EXITCODE=0
87
88 COLUMNS=80
89
90 PUPPET_AGENT='
91 now() { date +%s; }
92 let endtime="$(now) + 600"
93 while [ "$endtime" -gt "$(now)" ]; do
94     puppet agent --test --detailed-exitcodes
95     agent_exitcode=$?
96     if [ 0 = "$agent_exitcode" ] || [ 2 = "$agent_exitcode" ]; then
97         break
98     else
99         sleep 10s
100     fi
101 done
102 exit ${agent_exitcode:-99}
103 '
104
105 title () {
106   date=`date +'%Y-%m-%d %H:%M:%S'`
107   printf "%s\n" "$date $1"
108 }
109
110 function run_command() {
111   node=$1
112   return_var=$2
113   command=$3
114
115   title "Running '${command/ARVADOS_API_TOKEN=*/ARVADOS_API_TOKEN=suppressed}' on $node"
116   TMP_FILE=`mktemp`
117   if [[ "$DEBUG" != "0" ]]; then
118     ssh -t -p$SSH_PORT -o "StrictHostKeyChecking no" -o "ConnectTimeout 125" $ACCT@$node -C "$command" | tee $TMP_FILE
119     ECODE=$?
120   else
121     ssh -t -p$SSH_PORT -o "StrictHostKeyChecking no" -o "ConnectTimeout 125" $ACCT@$node -C "$command" > $TMP_FILE 2>&1
122     ECODE=$?
123   fi
124
125   if [[ "$ECODE" != "255" && "$ECODE" != "0"  ]]; then
126     # Ssh exists 255 if the connection timed out. Just ignore that, it's possible that this node is
127     #   a shell node that is down.
128     title "ERROR running command on $node: exit code $ECODE"
129     if [[ "$DEBUG" == "0" ]]; then
130       title "Command output follows:"
131       cat $TMP_FILE
132     fi
133   fi
134   if [[ "$ECODE" == "255" ]]; then
135     title "Connection denied or timed out"
136   fi
137   rm -f $TMP_FILE
138   eval "$return_var=$ECODE"
139 }
140
141 title "Loading ARVADOS_API_HOST and ARVADOS_API_TOKEN"
142 if [[ -f "$HOME/.config/arvados/$IDENTIFIER.arvadosapi.com.conf" ]]; then
143   . $HOME/.config/arvados/$IDENTIFIER.arvadosapi.com.conf
144 else
145   title "WARNING: $HOME/.config/arvados/$IDENTIFIER.arvadosapi.com.conf not found."
146 fi
147 if [[ "$ARVADOS_API_HOST" == "" ]] || [[ "$ARVADOS_API_TOKEN" == "" ]]; then
148   title "ERROR: ARVADOS_API_HOST and/or ARVADOS_API_TOKEN environment variables are not set."
149   exit 1
150 fi
151
152 run_command shell.$IDENTIFIER ECODE "if [[ ! -e common-workflow-language ]]; then git clone https://github.com/common-workflow-language/common-workflow-language.git; fi"
153
154 if [[ "$ECODE" != "0" ]]; then
155   echo "Failed to git clone https://github.com/common-workflow-language/common-workflow-language.git"
156   exit $ECODE
157 fi
158
159 run_command shell.$IDENTIFIER ECODE "if [[ ! -e arvados-cwl-runner-with-checksum.sh ]]; then printf \"%s\n%s\n\" '#!/bin/sh' 'exec arvados-cwl-runner --compute-checksum \"\$@\"' > arvados-cwl-runner-with-checksum.sh; chmod 755 arvados-cwl-runner-with-checksum.sh; fi"
160
161 if [[ "$ECODE" != "0" ]]; then
162   echo "Failed to create ~$ACCT/arvados-cwl-runner-with-checksum.sh"
163   exit $ECODE
164 fi
165
166 run_command shell.$IDENTIFIER ECODE "cd common-workflow-language; git pull; ARVADOS_API_HOST=$ARVADOS_API_HOST ARVADOS_API_TOKEN=$ARVADOS_API_TOKEN ./run_test.sh RUNNER=/home/$ACCT/arvados-cwl-runner-with-checksum.sh "
167
168 exit $ECODE