|
| 1 | +#!/bin/bash -e |
| 2 | + |
| 3 | +# tests-wrapper.sh -- wrap execution of CTest scripts by adding retry |
| 4 | +# |
| 5 | +# Usage (through CMake): |
| 6 | +# |
| 7 | +# tests-wrapper.sh --name <name> [--max-attempts N] [--timeout T] \ |
| 8 | +# --dont-fail-on-timeout] -- prog [arg1 [arg2...]] |
| 9 | + |
| 10 | +MAX_ATTEMPTS=1 |
| 11 | +TIMEOUT= # no timeout |
| 12 | +DONT_FAIL_ON_TIMEOUT= # if it times out, it fails |
| 13 | +ARGS=("$@") |
| 14 | + |
| 15 | +while [[ $# -gt 0 ]]; do |
| 16 | + case "$1" in |
| 17 | + --) |
| 18 | + shift |
| 19 | + break |
| 20 | + ;; |
| 21 | + --name) |
| 22 | + TEST_NAME="$2" |
| 23 | + shift 2 |
| 24 | + ;; |
| 25 | + --max-attempts) |
| 26 | + MAX_ATTEMPTS="$2" |
| 27 | + shift 2 |
| 28 | + ;; |
| 29 | + --timeout) |
| 30 | + TIMEOUT="$2" |
| 31 | + shift 2 |
| 32 | + ;; |
| 33 | + --dont-fail-on-timeout) |
| 34 | + DONT_FAIL_ON_TIMEOUT=1 |
| 35 | + shift |
| 36 | + ;; |
| 37 | + *) |
| 38 | + echo "Parameter unknown: $1" >&2 |
| 39 | + exit 1 |
| 40 | + ;; |
| 41 | + esac |
| 42 | +done |
| 43 | + |
| 44 | +# Check mandatory parmeters |
| 45 | +if [[ ! $TEST_NAME ]]; then |
| 46 | + echo "Test name is mandatory" >&2 |
| 47 | + exit 1 |
| 48 | +fi |
| 49 | + |
| 50 | +LOG="@CMAKE_BINARY_DIR@/test_logs/${TEST_NAME//\//_}.log" |
| 51 | +mkdir -p "$(dirname "$LOG")" |
| 52 | +rm -f "$LOG"* &> /dev/null |
| 53 | +exec &> >(tee "$LOG") |
| 54 | + |
| 55 | +function banner() { |
| 56 | + echo "=== $TEST_NAME - $1 ===" >&2 |
| 57 | +} |
| 58 | + |
| 59 | +banner "Starting test. Max attempts: $MAX_ATTEMPTS.${TIMEOUT:+" Timeout per attempt: $TIMEOUT."}${DONT_FAIL_ON_TIMEOUT:+" Timeouts are non-fatal."}" |
| 60 | +for A in "${ARGS[@]}"; do |
| 61 | + banner "Argument: $A" |
| 62 | +done |
| 63 | +banner "Current working directory: $PWD" |
| 64 | +banner "Environment" |
| 65 | +env |
| 66 | +banner "/Environment" |
| 67 | + |
| 68 | +# Do we have timeout? |
| 69 | +TIMEOUT_EXEC=timeout |
| 70 | +TIMEOUT_CMD= |
| 71 | +TIMEOUT_PSTACK= |
| 72 | +type $TIMEOUT_EXEC &> /dev/null || TIMEOUT_EXEC=gtimeout |
| 73 | +type $TIMEOUT_EXEC &> /dev/null || TIMEOUT_EXEC= |
| 74 | +if [[ $TIMEOUT_EXEC && $TIMEOUT ]]; then |
| 75 | + # Kill with 15; if after 10 seconds it's still alive, send 9 |
| 76 | + TIMEOUT_CMD="$TIMEOUT_EXEC --signal=SIGTERM --kill-after=10s ${TIMEOUT}s" |
| 77 | + # Get a stack trace, if possible, shortly before sending the first SIGTERM |
| 78 | + if type pstack &> /dev/null; then |
| 79 | + export GDB=$(which gdb 2> /dev/null) |
| 80 | + TIMEOUT_PSTACK=$(( TIMEOUT - 10 )) |
| 81 | + if [[ $TIMEOUT_PSTACK -lt 10 ]]; then |
| 82 | + TIMEOUT_PSTACK=10 |
| 83 | + fi |
| 84 | + else |
| 85 | + banner "Will not get stack trace of processes timing out: pstack not found" |
| 86 | + fi |
| 87 | +fi |
| 88 | + |
| 89 | +banner "Timeout prefix: $TIMEOUT_CMD" |
| 90 | + |
| 91 | +CMD="$1" |
| 92 | +shift |
| 93 | +type "$CMD" &> /dev/null || CMD="@CMAKE_BINARY_DIR@/bin/$CMD" |
| 94 | + |
| 95 | +for ((ATTEMPT=1; ATTEMPT<=MAX_ATTEMPTS; ATTEMPT++)); do |
| 96 | + banner "Running $CMD with args $* (attempt $ATTEMPT/$MAX_ATTEMPTS)" |
| 97 | + ERR=0 |
| 98 | + rm -f "${LOG}.bt" |
| 99 | + $TIMEOUT_CMD "$CMD" "$@" & |
| 100 | + REAL_PID=$(pgrep -P$! 2> /dev/null || true) |
| 101 | + if [[ $REAL_PID && $TIMEOUT_PSTACK ]]; then |
| 102 | + banner "Process running as $REAL_PID for attempt $ATTEMPT" |
| 103 | + FINISHED= |
| 104 | + for ((I=0; I<TIMEOUT_PSTACK; I++)); do |
| 105 | + if ! kill -0 $REAL_PID &> /dev/null; then |
| 106 | + FINISHED=1 |
| 107 | + break |
| 108 | + fi |
| 109 | + sleep 1 |
| 110 | + done |
| 111 | + if [[ ! $FINISHED ]]; then |
| 112 | + pstack $REAL_PID &> "${LOG}.bt" || true |
| 113 | + fi |
| 114 | + fi |
| 115 | + wait $! || ERR=$? # wait timeout process, not real PID |
| 116 | + if [[ $ERR == 0 ]]; then |
| 117 | + banner "Test finished with success after $ATTEMPT attempts, exiting" |
| 118 | + mv "$LOG" "${LOG}.0" |
| 119 | + exit 0 |
| 120 | + else |
| 121 | + banner "Test attempt $ATTEMPT/$MAX_ATTEMPTS failed with exit code $ERR" |
| 122 | + if [[ -e "${LOG}.bt" ]]; then |
| 123 | + banner "Stack trace follows for attempt $ATTEMPT" |
| 124 | + cat "${LOG}.bt" |
| 125 | + rm -f "${LOG}.bt" |
| 126 | + banner "End of stack trace for attempt $ATTEMPT" |
| 127 | + fi |
| 128 | + fi |
| 129 | +done |
| 130 | + |
| 131 | +mv "$LOG" "${LOG}.${ERR}" # log file will contain exitcode in name |
| 132 | +banner "Test failed after $MAX_ATTEMPTS attempts with $ERR" |
| 133 | +if [[ $DONT_FAIL_ON_TIMEOUT && $ERR == 124 ]]; then |
| 134 | + # man timeout --> 124 is for "timed out" |
| 135 | + banner "Reason for failure: timeout, explicitly set as non-fatal. Exiting with 0" |
| 136 | + exit 0 |
| 137 | +fi |
| 138 | +exit 1 |
0 commit comments