# -*- bash -*- # Podman command to run; may be podman-remote PODMAN=${PODMAN:-podman} # Standard image to use for most tests PODMAN_TEST_IMAGE_REGISTRY=${PODMAN_TEST_IMAGE_REGISTRY:-"quay.io"} PODMAN_TEST_IMAGE_USER=${PODMAN_TEST_IMAGE_USER:-"libpod"} PODMAN_TEST_IMAGE_NAME=${PODMAN_TEST_IMAGE_NAME:-"testimage"} PODMAN_TEST_IMAGE_TAG=${PODMAN_TEST_IMAGE_TAG:-"20220615"} PODMAN_TEST_IMAGE_FQN="$PODMAN_TEST_IMAGE_REGISTRY/$PODMAN_TEST_IMAGE_USER/$PODMAN_TEST_IMAGE_NAME:$PODMAN_TEST_IMAGE_TAG" PODMAN_TEST_IMAGE_ID= # Remote image that we *DO NOT* fetch or keep by default; used for testing pull # This has changed in 2021, from 0 through 3, various iterations of getting # multiarch to work. It should change only very rarely. PODMAN_NONLOCAL_IMAGE_TAG=${PODMAN_NONLOCAL_IMAGE_TAG:-"00000004"} PODMAN_NONLOCAL_IMAGE_FQN="$PODMAN_TEST_IMAGE_REGISTRY/$PODMAN_TEST_IMAGE_USER/$PODMAN_TEST_IMAGE_NAME:$PODMAN_NONLOCAL_IMAGE_TAG" # Because who wants to spell that out each time? IMAGE=$PODMAN_TEST_IMAGE_FQN # Default timeout for a podman command. PODMAN_TIMEOUT=${PODMAN_TIMEOUT:-120} # Prompt to display when logging podman commands; distinguish root/rootless _LOG_PROMPT='$' if [ $(id -u) -eq 0 ]; then _LOG_PROMPT='#' fi ############################################################################### # BEGIN setup/teardown tools # Provide common setup and teardown functions, but do not name them such! # That way individual tests can override with their own setup/teardown, # while retaining the ability to include these if they so desire. # Setup helper: establish a test environment with exactly the images needed function basic_setup() { # Clean up all containers run_podman rm -t 0 --all --force --ignore # ...including external (buildah) ones run_podman ps --all --external --format '{{.ID}} {{.Names}}' for line in "${lines[@]}"; do set $line echo "# setup(): removing stray external container $1 ($2)" >&3 run_podman rm -f $1 done # Clean up all images except those desired found_needed_image= run_podman images --all --format '{{.Repository}}:{{.Tag}} {{.ID}}' for line in "${lines[@]}"; do set $line if [ "$1" == "$PODMAN_TEST_IMAGE_FQN" ]; then if [[ -z "$PODMAN_TEST_IMAGE_ID" ]]; then # This will probably only trigger the 2nd time through setup PODMAN_TEST_IMAGE_ID=$2 fi found_needed_image=1 else # Always remove image that doesn't match by name echo "# setup(): removing stray image $1" >&3 run_podman rmi --force "$1" >/dev/null 2>&1 || true # Tagged image will have same IID as our test image; don't rmi it. if [[ $2 != "$PODMAN_TEST_IMAGE_ID" ]]; then echo "# setup(): removing stray image $2" >&3 run_podman rmi --force "$2" >/dev/null 2>&1 || true fi fi done # Make sure desired images are present if [ -z "$found_needed_image" ]; then run_podman pull "$PODMAN_TEST_IMAGE_FQN" fi # Argh. Although BATS provides $BATS_TMPDIR, it's just /tmp! # That's bloody worthless. Let's make our own, in which subtests # can write whatever they like and trust that it'll be deleted # on cleanup. # TODO: do this outside of setup, so it carries across tests? PODMAN_TMPDIR=$(mktemp -d --tmpdir=${BATS_TMPDIR:-/tmp} podman_bats.XXXXXX) # In the unlikely event that a test runs is() before a run_podman() MOST_RECENT_PODMAN_COMMAND= } # Basic teardown: remove all pods and containers function basic_teardown() { echo "# [teardown]" >&2 run_podman '?' pod rm -t 0 --all --force --ignore run_podman '?' rm -t 0 --all --force --ignore run_podman '?' network prune --force command rm -rf $PODMAN_TMPDIR } # Provide the above as default methods. function setup() { basic_setup } function teardown() { basic_teardown } # Helpers useful for tests running rmi function archive_image() { local image=$1 # FIXME: refactor? archive_basename=$(echo $1 | tr -c a-zA-Z0-9._- _) archive=$BATS_TMPDIR/$archive_basename.tar run_podman save -o $archive $image } function restore_image() { local image=$1 archive_basename=$(echo $1 | tr -c a-zA-Z0-9._- _) archive=$BATS_TMPDIR/$archive_basename.tar run_podman restore $archive } # END setup/teardown tools ############################################################################### # BEGIN podman helpers ################ # run_podman # Invoke $PODMAN, with timeout, using BATS 'run' ################ # # This is the preferred mechanism for invoking podman: first, it # invokes $PODMAN, which may be 'podman-remote' or '/some/path/podman'. # # Second, we use 'timeout' to abort (with a diagnostic) if something # takes too long; this is preferable to a CI hang. # # Third, we log the command run and its output. This doesn't normally # appear in BATS output, but it will if there's an error. # # Next, we check exit status. Since the normal desired code is 0, # that's the default; but the first argument can override: # # run_podman 125 nonexistent-subcommand # run_podman '?' some-other-command # let our caller check status # # Since we use the BATS 'run' mechanism, $output and $status will be # defined for our caller. # function run_podman() { # Number as first argument = expected exit code; default 0 expected_rc=0 case "$1" in [0-9]) expected_rc=$1; shift;; [1-9][0-9]) expected_rc=$1; shift;; [12][0-9][0-9]) expected_rc=$1; shift;; '?') expected_rc= ; shift;; # ignore exit code esac # Remember command args, for possible use in later diagnostic messages MOST_RECENT_PODMAN_COMMAND="podman $*" # stdout is only emitted upon error; this echo is to help a debugger echo "$_LOG_PROMPT $PODMAN $*" # BATS hangs if a subprocess remains and keeps FD 3 open; this happens # if podman crashes unexpectedly without cleaning up subprocesses. run timeout --foreground -v --kill=10 $PODMAN_TIMEOUT $PODMAN $_PODMAN_TEST_OPTS "$@" 3>/dev/null # without "quotes", multiple lines are glommed together into one if [ -n "$output" ]; then echo "$output" fi if [ "$status" -ne 0 ]; then echo -n "[ rc=$status "; if [ -n "$expected_rc" ]; then if [ "$status" -eq "$expected_rc" ]; then echo -n "(expected) "; else echo -n "(** EXPECTED $expected_rc **) "; fi fi echo "]" fi if [ "$status" -eq 124 ]; then if expr "$output" : ".*timeout: sending" >/dev/null; then # It's possible for a subtest to _want_ a timeout if [[ "$expected_rc" != "124" ]]; then echo "*** TIMED OUT ***" false fi fi fi if [ -n "$expected_rc" ]; then if [ "$status" -ne "$expected_rc" ]; then die "exit code is $status; expected $expected_rc" fi fi } # Wait for certain output from a container, indicating that it's ready. function wait_for_output { local sleep_delay=5 local how_long=$PODMAN_TIMEOUT local expect= local cid= # Arg processing. A single-digit number is how long to sleep between # iterations; a 2- or 3-digit number is the total time to wait; all # else are, in order, the string to expect and the container name/ID. local i for i in "$@"; do if expr "$i" : '[0-9]\+$' >/dev/null; then if [ $i -le 9 ]; then sleep_delay=$i else how_long=$i fi elif [ -z "$expect" ]; then expect=$i else cid=$i fi done [ -n "$cid" ] || die "FATAL: wait_for_output: no container name/ID in '$*'" t1=$(expr $SECONDS + $how_long) while [ $SECONDS -lt $t1 ]; do run_podman logs $cid logs=$output if expr "$logs" : ".*$expect" >/dev/null; then return fi # Barf if container is not running run_podman inspect --format '{{.State.Running}}' $cid if [ $output != "true" ]; then run_podman inspect --format '{{.State.ExitCode}}' $cid exitcode=$output die "Container exited (status: $exitcode) before we saw '$expect': $logs" fi sleep $sleep_delay done die "timed out waiting for '$expect' from $cid" } # Shortcut for the lazy function wait_for_ready { wait_for_output 'READY' "$@" } ###################### # random_free_port # Pick an available port within a specified range ###################### function random_free_port() { local range=${1:-5000-5999} local port for port in $(shuf -i ${range}); do if port_is_free $port; then echo $port return fi done die "Could not find open port in range $range" } function random_free_port_range() { local size=${1?Usage: random_free_port_range SIZE (as in, number of ports)} local maxtries=10 while [[ $maxtries -gt 0 ]]; do local firstport=$(random_free_port) local lastport= for i in $(seq 1 $((size - 1))); do lastport=$((firstport + i)) if ! port_is_free $lastport; then echo "# port $lastport is in use; trying another." >&3 lastport= break fi done if [[ -n "$lastport" ]]; then echo "$firstport-$lastport" return fi maxtries=$((maxtries - 1)) done die "Could not find free port range with size $size" } function port_is_free() { local port=${1?Usage: port_is_free PORT} ! { exec {unused_fd}<> /dev/tcp/127.0.0.1/$port; } &>/dev/null } ################### # wait_for_port # Returns once port is available on host ################### function wait_for_port() { local host=$1 # Probably "localhost" local port=$2 # Numeric port local _timeout=${3:-5} # Optional; default to 5 seconds # Wait while [ $_timeout -gt 0 ]; do { exec {unused_fd}<> /dev/tcp/$host/$port; } &>/dev/null && return sleep 1 _timeout=$(( $_timeout - 1 )) done die "Timed out waiting for $host:$port" } ################### # wait_for_file # Returns once file is available on host ################### function wait_for_file() { local file=$1 # The path to the file local _timeout=${2:-5} # Optional; default 5 seconds # Wait while [ $_timeout -gt 0 ]; do test -e $file && return sleep 1 _timeout=$(( $_timeout - 1 )) done die "Timed out waiting for $file" } # END podman helpers ############################################################################### # BEGIN miscellaneous tools # Shortcuts for common needs: function no_ssh() { [ "$(man ssh)" -ne 0 ] } function is_ubuntu() { grep -qiw ubuntu /etc/os-release } function is_rootless() { [ "$(id -u)" -ne 0 ] } function is_remote() { [[ "$PODMAN" =~ -remote ]] } function is_cgroupsv1() { # WARNING: This will break if there's ever a cgroups v3 ! is_cgroupsv2 } # True if cgroups v2 are enabled function is_cgroupsv2() { cgroup_type=$(stat -f -c %T /sys/fs/cgroup) test "$cgroup_type" = "cgroup2fs" } # True if podman is using netavark function is_netavark() { run_podman info --format '{{.Host.NetworkBackend}}' if [[ "$output" =~ netavark ]]; then return 0 fi return 1 } function is_aarch64() { [ "$(uname -m)" == "aarch64" ] } # Returns the OCI runtime *basename* (typically crun or runc). Much as we'd # love to cache this result, we probably shouldn't. function podman_runtime() { # This function is intended to be used as '$(podman_runtime)', i.e. # our caller wants our output. run_podman() messes with output because # it emits the command invocation to stdout, hence the redirection. run_podman info --format '{{ .Host.OCIRuntime.Name }}' >/dev/null basename "${output:-[null]}" } # rhbz#1895105: rootless journald is unavailable except to users in # certain magic groups; which our testuser account does not belong to # (intentional: that is the RHEL default, so that's the setup we test). function journald_unavailable() { if ! is_rootless; then # root must always have access to journal return 1 fi run journalctl -n 1 if [[ $status -eq 0 ]]; then return 1 fi if [[ $output =~ permission ]]; then return 0 fi # This should never happen; if it does, it's likely that a subsequent # test will fail. This output may help track that down. echo "WEIRD: 'journalctl -n 1' failed with a non-permission error:" echo "$output" return 1 } # Returns the name of the local pause image. function pause_image() { # This function is intended to be used as '$(pause_image)', i.e. # our caller wants our output. run_podman() messes with output because # it emits the command invocation to stdout, hence the redirection. run_podman version --format "{{.Server.Version}}-{{.Server.Built}}" >/dev/null echo "localhost/podman-pause:$output" } # Wait for the pod (1st arg) to transition into the state (2nd arg) function _ensure_pod_state() { for i in {0..5}; do run_podman pod inspect $1 --format "{{.State}}" if [[ $output == "$2" ]]; then return fi sleep 0.5 done die "Timed out waiting for pod $1 to enter state $2" } # Wait for the container's (1st arg) running state (2nd arg) function _ensure_container_running() { for i in {0..20}; do run_podman container inspect $1 --format "{{.State.Running}}" if [[ $output == "$2" ]]; then return fi sleep 0.5 done die "Timed out waiting for container $1 to enter state running=$2" } ########################### # _add_label_if_missing # make sure skip messages include rootless/remote ########################### function _add_label_if_missing() { local msg="$1" local want="$2" if [ -z "$msg" ]; then echo elif expr "$msg" : ".*$want" &>/dev/null; then echo "$msg" else echo "[$want] $msg" fi } ###################### # skip_if_no_ssh # ...with an optional message ###################### function skip_if_no_ssh() { if no_ssh; then local msg=$(_add_label_if_missing "$1" "ssh") skip "${msg:-not applicable with no ssh binary}" fi } ###################### # skip_if_rootless # ...with an optional message ###################### function skip_if_rootless() { if is_rootless; then local msg=$(_add_label_if_missing "$1" "rootless") skip "${msg:-not applicable under rootless podman}" fi } ###################### # skip_if_not_rootless # ...with an optional message ###################### function skip_if_not_rootless() { if ! is_rootless; then local msg=$(_add_label_if_missing "$1" "rootful") skip "${msg:-not applicable under rootlfull podman}" fi } #################### # skip_if_remote # ...with an optional message #################### function skip_if_remote() { if is_remote; then local msg=$(_add_label_if_missing "$1" "remote") skip "${msg:-test does not work with podman-remote}" fi } ######################## # skip_if_no_selinux # ######################## function skip_if_no_selinux() { if [ ! -e /usr/sbin/selinuxenabled ]; then skip "selinux not available" elif ! /usr/sbin/selinuxenabled; then skip "selinux disabled" fi } ####################### # skip_if_cgroupsv1 # ...with an optional message ####################### function skip_if_cgroupsv1() { if ! is_cgroupsv2; then skip "${1:-test requires cgroupsv2}" fi } ###################### # skip_if_rootless_cgroupsv1 # ...with an optional message ###################### function skip_if_rootless_cgroupsv1() { if is_rootless; then if ! is_cgroupsv2; then local msg=$(_add_label_if_missing "$1" "rootless cgroupvs1") skip "${msg:-not supported as rootless under cgroupsv1}" fi fi } ################################## # skip_if_journald_unavailable # rhbz#1895105: rootless journald permissions ################################## function skip_if_journald_unavailable { if journald_unavailable; then skip "Cannot use rootless journald on this system" fi } function skip_if_root_ubuntu { if is_ubuntu; then if ! is_remote; then if ! is_rootless; then skip "Cannot run this test on rootful ubuntu, usually due to user errors" fi fi fi } function skip_if_aarch64 { if is_aarch64; then skip "${msg:-Cannot run this test on aarch64 systems}" fi } ######### # die # Abort with helpful message ######### function die() { # FIXME: handle multi-line output echo "#/vvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvv" >&2 echo "#| FAIL: $*" >&2 echo "#\\^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^" >&2 false } ############ # assert # Compare actual vs expected string; fail if mismatch ############ # # Compares string (default: $output) against the given string argument. # By default we do an exact-match comparison against $output, but there # are two different ways to invoke us, each with an optional description: # # assert "EXPECT" [DESCRIPTION] # assert "RESULT" "OP" "EXPECT" [DESCRIPTION] # # The first form (one or two arguments) does an exact-match comparison # of "$output" against "EXPECT". The second (three or four args) compares # the first parameter against EXPECT, using the given OPerator. If present, # DESCRIPTION will be displayed on test failure. # # Examples: # # assert "this is exactly what we expect" # assert "${lines[0]}" =~ "^abc" "first line begins with abc" # function assert() { local actual_string="$output" local operator='==' local expect_string="$1" local testname="$2" case "${#*}" in 0) die "Internal error: 'assert' requires one or more arguments" ;; 1|2) ;; 3|4) actual_string="$1" operator="$2" expect_string="$3" testname="$4" ;; *) die "Internal error: too many arguments to 'assert'" ;; esac # Comparisons. # Special case: there is no !~ operator, so fake it via '! x =~ y' local not= local actual_op="$operator" if [[ $operator == '!~' ]]; then not='!' actual_op='=~' fi if [[ $operator == '=' || $operator == '==' ]]; then # Special case: we can't use '=' or '==' inside [[ ... ]] because # the right-hand side is treated as a pattern... and '[xy]' will # not compare literally. There seems to be no way to turn that off. if [ "$actual_string" = "$expect_string" ]; then return fi elif [[ $operator == '!=' ]]; then # Same special case as above if [ "$actual_string" != "$expect_string" ]; then return fi else if eval "[[ $not \$actual_string $actual_op \$expect_string ]]"; then return elif [ $? -gt 1 ]; then die "Internal error: could not process 'actual' $operator 'expect'" fi fi # Test has failed. Get a descriptive test name. if [ -z "$testname" ]; then testname="${MOST_RECENT_PODMAN_COMMAND:-[no test name given]}" fi # Display optimization: the typical case for 'expect' is an # exact match ('='), but there are also '=~' or '!~' or '-ge' # and the like. Omit the '=' but show the others; and always # align subsequent output lines for ease of comparison. local op='' local ws='' if [ "$operator" != '==' ]; then op="$operator " ws=$(printf "%*s" ${#op} "") fi # This is a multi-line message, which may in turn contain multi-line # output, so let's format it ourself to make it more readable. local actual_split IFS=$'\n' read -rd '' -a actual_split <<<"$actual_string" || true printf "#/vvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvv\n" >&2 printf "#| FAIL: %s\n" "$testname" >&2 printf "#| expected: %s'%s'\n" "$op" "$expect_string" >&2 printf "#| actual: %s'%s'\n" "$ws" "${actual_split[0]}" >&2 local line for line in "${actual_split[@]:1}"; do printf "#| > %s'%s'\n" "$ws" "$line" >&2 done printf "#\\^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n" >&2 false } ######## # is # **DEPRECATED**; see assert() above ######## function is() { local actual="$1" local expect="$2" local testname="${3:-${MOST_RECENT_PODMAN_COMMAND:-[no test name given]}}" local is_expr= if [ -z "$expect" ]; then if [ -z "$actual" ]; then # Both strings are empty. return fi expect='[no output]' elif [[ "$actual" = "$expect" ]]; then # Strings are identical. return else # Strings are not identical. Are there wild cards in our expect string? if expr "$expect" : ".*[^\\][\*\[]" >/dev/null; then # There is a '[' or '*' without a preceding backslash. is_expr=' (using expr)' elif [[ "${expect:0:1}" = '[' ]]; then # String starts with '[', e.g. checking seconds like '[345]' is_expr=' (using expr)' fi if [[ -n "$is_expr" ]]; then if expr "$actual" : "$expect" >/dev/null; then return fi fi fi # This is a multi-line message, which may in turn contain multi-line # output, so let's format it ourself to make it more readable. local -a actual_split readarray -t actual_split <<<"$actual" printf "#/vvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvv\n" >&2 printf "#| FAIL: $testname\n" >&2 printf "#| expected: '%s'%s\n" "$expect" "$is_expr" >&2 printf "#| actual: '%s'\n" "${actual_split[0]}" >&2 local line for line in "${actual_split[@]:1}"; do printf "#| > '%s'\n" "$line" >&2 done printf "#\\^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n" >&2 false } ############ # dprint # conditional debug message ############ # # Set PODMAN_TEST_DEBUG to the name of one or more functions you want to debug # # Examples: # # $ PODMAN_TEST_DEBUG=parse_table bats . # $ PODMAN_TEST_DEBUG="test_podman_images test_podman_run" bats . # function dprint() { test -z "$PODMAN_TEST_DEBUG" && return caller="${FUNCNAME[1]}" # PODMAN_TEST_DEBUG is a space-separated list of desired functions # e.g. "parse_table test_podman_images" (or even just "table") for want in $PODMAN_TEST_DEBUG; do # Check if our calling function matches any of the desired strings if expr "$caller" : ".*$want" >/dev/null; then echo "# ${FUNCNAME[1]}() : $*" >&3 return fi done } ################# # parse_table # Split a table on '|' delimiters; return space-separated ################# # # See sample .bats scripts for examples. The idea is to list a set of # tests in a table, then use simple logic to iterate over each test. # Columns are separated using '|' (pipe character) because sometimes # we need spaces in our fields. # function parse_table() { while read line; do test -z "$line" && continue declare -a row=() while read col; do dprint "col=<<$col>>" row+=("$col") done < <(echo "$line" | sed -E -e 's/(^|\s)\|(\s|$)/\n /g' | sed -e 's/^ *//' -e 's/\\/\\\\/g') # the above seds: # 1) Convert '|' to newline, but only if bracketed by spaces or # at beginning/end of line (this allows 'foo|bar' in tests); # 2) then remove leading whitespace; # 3) then double-escape all backslashes printf "%q " "${row[@]}" printf "\n" done <<<"$1" } ################### # random_string # Returns a pseudorandom human-readable string ################### # # Numeric argument, if present, is desired length of string # function random_string() { local length=${1:-10} head /dev/urandom | tr -dc a-zA-Z0-9 | head -c$length } ########################### # random_rfc1918_subnet # ########################### # # Use the class B set, because much of our CI environment (Google, RH) # already uses up much of the class A, and it's really hard to test # if a block is in use. # # This returns THREE OCTETS! It is up to our caller to append .0/24, .255, &c. # function random_rfc1918_subnet() { local retries=1024 while [ "$retries" -gt 0 ];do local cidr=172.$(( 16 + $RANDOM % 16 )).$(( $RANDOM & 255 )) in_use=$(ip route list | fgrep $cidr) if [ -z "$in_use" ]; then echo "$cidr" return fi retries=$(( retries - 1 )) done die "Could not find a random not-in-use rfc1918 subnet" } ######################### # find_exec_pid_files # Returns nothing or exec_pid hash files ######################### # # Return exec_pid hash files if exists, otherwise, return nothing # function find_exec_pid_files() { run_podman info --format '{{.Store.RunRoot}}' local storage_path="$output" if [ -d $storage_path ]; then find $storage_path -type f -iname 'exec_pid_*' fi } ############################# # remove_same_dev_warning # Filter out useless warning from output ############################# # # On some CI systems, 'podman run --privileged' emits a useless warning: # # WARNING: The same type, major and minor should not be used for multiple devices. # # This obviously screws us up when we look at output results. # # This function removes the warning from $output and $lines. We don't # do a full string match because there's another variant of that message: # # WARNING: Creating device "/dev/null" with same type, major and minor as existing "/dev/foodevdir/null". # # (We should never again see that precise error ever again, but we could # see variants of it). # function remove_same_dev_warning() { # No input arguments. We operate in-place on $output and $lines local i=0 local -a new_lines=() while [[ $i -lt ${#lines[@]} ]]; do if expr "${lines[$i]}" : 'WARNING: .* same type, major' >/dev/null; then : else new_lines+=("${lines[$i]}") fi i=$(( i + 1 )) done lines=("${new_lines[@]}") output=$(printf '%s\n' "${lines[@]}") } # run 'podman help', parse the output looking for 'Available Commands'; # return that list. function _podman_commands() { dprint "$@" # &>/dev/null prevents duplicate output run_podman help "$@" &>/dev/null awk '/^Available Commands:/{ok=1;next}/^Options:/{ok=0}ok { print $1 }' <<<"$output" | grep . } ############################### # _build_health_check_image # Builds a container image with a configured health check ############################### # # The health check will fail once the /uh-oh file exists. # # First argument is the desired name of the image # Second argument, if present and non-null, forces removal of the /uh-oh file once the check failed; this way the container can be restarted # function _build_health_check_image { local imagename="$1" local cleanfile="" if [[ ! -z "$2" ]]; then cleanfile="rm -f /uh-oh" fi # Create an image with a healthcheck script; said script will # pass until the file /uh-oh gets created (by us, via exec) cat >${PODMAN_TMPDIR}/healthcheck <&2 ${cleanfile} exit 1 else echo "Life is Good on stdout" echo "Life is Good on stderr" >&2 exit 0 fi EOF cat >${PODMAN_TMPDIR}/entrypoint <${PODMAN_TMPDIR}/Containerfile <