prole/tests/etc/test_init_cloudnative_pg_instances.sh

393 lines
12 KiB
Bash

#!/usr/bin/env bash
# Unit test for etc/init_cloudnative_pg.sh
# Verifies CNPG placement policy is label-based (no hostname pinning) and that
# bootstrap reconciles instances safely under degraded startup.
set -euo pipefail
SCRIPT_DIR=$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)
KNOE_HOME=$(cd "$SCRIPT_DIR/../.." && pwd)
ETC_DIR="$KNOE_HOME/etc"
SCRIPT_UNDER_TEST="$ETC_DIR/init_cloudnative_pg.sh"
TMP_DIR=$(mktemp -d)
trap 'rm -rf "$TMP_DIR"' EXIT
BIN_DIR="$TMP_DIR/bin"
mkdir -p "$BIN_DIR"
mock_tool() {
cat <<M_EOF > "$BIN_DIR/$1"
#!/usr/bin/env bash
echo "Mocked $1 called with \$@" >> "$TMP_DIR/mock_calls.log"
exit 0
M_EOF
chmod +x "$BIN_DIR/$1"
}
cat <<'K_EOF' > "$BIN_DIR/kubectl"
#!/usr/bin/env bash
set -euo pipefail
_state_file="${TMP_DIR}/mock_cluster_instances"
_log_file="${TMP_DIR}/mock_calls.log"
echo "Mocked kubectl called with $*" >> "${_log_file}"
args="$*"
# API server readiness probes used by wait_for_apiserver_ready()
if [[ "${args}" == *"get --raw=/readyz"* || "${args}" == *"get --raw='/readyz'"* || "${args}" == *"get --raw=\"/readyz\""* ]]; then
echo "ok"
exit 0
fi
if [[ "${args}" == *"version --short"* ]]; then
echo "Client Version: v0.0.0"
echo "Server Version: v0.0.0"
exit 0
fi
# Short-circuit CNPG operator + webhook waits
if [[ "${args}" == *"get deployment"*"-n cnpg-system"*"cnpg-controller-manager"* ]]; then
exit 0
fi
if [[ "${args}" == *"get endpoints"*"cnpg-webhook-service"* ]]; then
echo "10.42.0.10"
exit 0
fi
# Avoid cert-manager installation waits
if [[ "${args}" == *"get crd"*"certificates.cert-manager.io"* ]]; then
exit 0
fi
if [[ "${args}" == *"-n cert-manager"*"get deploy"* ]]; then
exit 0
fi
# Avoid barman plugin waits
if [[ "${args}" == *"get crd"*"objectstores.barmancloud.cnpg.io"* ]]; then
exit 0
fi
if [[ "${args}" == *"-n cnpg-system"*"get secret"*"barman-cloud-client-tls"* ]]; then
exit 0
fi
if [[ "${args}" == *"-n cnpg-system"*"get secret"*"barman-cloud-server-tls"* ]]; then
exit 0
fi
if [[ "${args}" == *"-n cnpg-system"*"get deploy"*"barman-cloud"* ]]; then
exit 0
fi
if [[ "${args}" == *"-n cnpg-system"*"rollout status"*"deploy/barman-cloud"* ]]; then
exit 0
fi
if [[ "${args}" == *"-n cnpg-system"*"get svc"*"-l cnpg.io/pluginName=barman-cloud.cloudnative-pg.io"*"-o jsonpath="*"metadata.name"* ]]; then
echo "barman-cloud"
exit 0
fi
if [[ "${args}" == *"-n cnpg-system"*"get svc barman-cloud"*"-o jsonpath="*"metadata.name"* ]]; then
echo "barman-cloud"
exit 0
fi
if [[ "${args}" == *"-n cnpg-system"*"get svc barman-cloud"*"pluginClientSecret"* ]]; then
echo "barman-cloud-client-tls"
exit 0
fi
if [[ "${args}" == *"-n cnpg-system"*"get svc barman-cloud"*"pluginServerSecret"* ]]; then
echo "barman-cloud-server-tls"
exit 0
fi
if [[ "${args}" == *"-n cnpg-system"*"get svc barman-cloud"*"pluginPort"* ]]; then
echo "9090"
exit 0
fi
if [[ "${args}" == *"-n cnpg-system"*"get endpoints barman-cloud"* ]]; then
echo "10.42.0.11"
exit 0
fi
# Namespace check in ensure_namespace()
if [[ "${args}" == "get namespace test-ns" ]]; then
exit 0
fi
# Simulate kubectl local merge patch used for selector injection.
if [[ "${args}" == patch*" --local "*" -o yaml"* ]]; then
src=""
if [[ "$#" -ge 2 ]]; then
for ((i=1; i<=$#; i++)); do
if [[ "${!i}" == "-f" ]]; then
j=$((i+1))
src="${!j:-}"
break
fi
done
fi
if [[ -n "${src}" && -f "${src}" ]]; then
awk '
{
print
if ($0 ~ /^[[:space:]]*synology\.storage\/role:[[:space:]]*data[[:space:]]*$/) {
match($0, /^[[:space:]]*/)
indent = substr($0, RSTART, RLENGTH)
print indent "knoe.io/namespace: test-ns"
print indent "knoe.io/cluster: knoe-db"
}
if ($0 ~ /^[[:space:]]*synology\.storage\/role:[[:space:]]*wal[[:space:]]*$/) {
match($0, /^[[:space:]]*/)
indent = substr($0, RSTART, RLENGTH)
print indent "knoe.io/namespace: test-ns"
print indent "knoe.io/cluster: knoe-db"
}
}
' "${src}"
fi
exit 0
fi
# Capture applied CNPG manifest (stdin or file)
if [[ "${args}" == apply* && "${args}" == *"-n"*"test-ns"* && "${args}" == *" -f "* ]]; then
f=""
if [[ "$#" -ge 2 ]]; then
for ((i=1; i<=$#; i++)); do
if [[ "${!i}" == "-f" ]]; then
j=$((i+1))
f="${!j:-}"
break
fi
done
fi
if [[ "${f}" == "-" ]]; then
cat > "${TMP_DIR}/applied.yaml"
elif [[ -n "${f}" && -f "${f}" ]]; then
cat "${f}" > "${TMP_DIR}/applied.yaml"
else
: > "${TMP_DIR}/applied.yaml"
fi
echo "cluster.postgresql.cnpg.io/knoe-db configured"
exit 0
fi
# Runtime storage validation (k3s) expects live Cluster + PVC/PV objects.
if [[ "${args}" == *"-n test-ns"*"get cluster"*"knoe-db"*"-o json"* ]]; then
cat <<'JSON'
{"apiVersion":"postgresql.cnpg.io/v1","kind":"Cluster","metadata":{"name":"knoe-db"},"spec":{"storage":{"pvcTemplate":{"storageClassName":"synology-iscsi","selector":{"matchLabels":{"synology.storage/role":"data"}}}},"walStorage":{"pvcTemplate":{"storageClassName":"synology-iscsi","selector":{"matchLabels":{"synology.storage/role":"wal"}}}}}}
JSON
exit 0
fi
if [[ "${args}" == *"-n test-ns"*"get pvc"*"cnpg.io/cluster=knoe-db"*"-o json"* ]]; then
cat <<'JSON'
{"items":[
{"metadata":{"name":"knoe-db-1"},"spec":{"storageClassName":"synology-iscsi","volumeName":"pv-data"}},
{"metadata":{"name":"knoe-db-1-wal"},"spec":{"storageClassName":"synology-iscsi","volumeName":"pv-wal"}}
]}
JSON
exit 0
fi
if [[ "${args}" == "get pv pv-data -o json"* ]]; then
echo '{"spec":{"storageClassName":"synology-iscsi","local":{"path":"/synology/d001/data"}}}'
exit 0
fi
if [[ "${args}" == "get pv pv-wal -o json"* ]]; then
echo '{"spec":{"storageClassName":"synology-iscsi","local":{"path":"/synology/d001/wal"}}}'
exit 0
fi
# Node inventory used by reconcile_cnpg_instances
if [[ "${args}" == "get nodes --no-headers" ]]; then
n="${MOCK_READY_NODES:-1}"
for i in $(seq 1 "$n"); do
echo "node-${i} Ready <none> 0d v0"
done
exit 0
fi
if [[ "${args}" == "get nodes -l knoe.org/node-role=db --no-headers" ]]; then
n="${MOCK_DB_NODES:-1}"
for i in $(seq 1 "$n"); do
echo "dbnode-${i} Ready <none> 0d v0"
done
exit 0
fi
# Cluster instance state used by reconcile_cnpg_instances
if [[ "${args}" == *"-n test-ns"*"get cluster"*"knoe-db"*"-o jsonpath={.spec.instances}"* ]]; then
if [[ -f "${_state_file}" ]]; then
cat "${_state_file}"
else
echo "3"
fi
exit 0
fi
if [[ "${args}" == *"-n test-ns"*"patch cluster"*"knoe-db"*"\"instances\""* ]]; then
inst=$(printf '%s' "${args}" | sed -n 's/.*"instances":\([0-9][0-9]*\).*/\1/p' | head -n 1)
if [[ -n "${inst:-}" ]]; then
printf '%s' "${inst}" > "${_state_file}"
fi
echo "patched"
exit 0
fi
exit 0
K_EOF
chmod +x "$BIN_DIR/kubectl"
mock_tool curl
mock_tool docker
mock_tool k3d
mock_tool skopeo
mock_tool ansible-playbook
mock_tool tofu
mock_tool terraform
mock_tool ollama
export PATH="$BIN_DIR:$PATH"
export TMP_DIR
mkdir -p "$TMP_DIR/service/secrets"
printf '%s' "dummy-private-key" > "$TMP_DIR/service/secrets/admin.key"
printf '%s' "dummy-public-key" > "$TMP_DIR/service/secrets/admin.pub"
run_case() {
local label="$1"
local cnpg_instances_cfg="${2:-}"
local mock_ready_nodes="${3:-1}"
local mock_db_nodes="${4:-1}"
local selector_mode="${5:-provided}"
local data_selector_env='{"synology.storage/role":"data","knoe.io/namespace":"test-ns","knoe.io/cluster":"knoe-db"}'
local wal_selector_env='{"synology.storage/role":"wal","knoe.io/namespace":"test-ns","knoe.io/cluster":"knoe-db"}'
if [[ "$selector_mode" == "missing" ]]; then
data_selector_env=""
wal_selector_env=""
fi
local conf_dir="$TMP_DIR/conf-${label}"
local env_home="$TMP_DIR/env-${label}"
mkdir -p "$conf_dir"
mkdir -p "$env_home"
cat <<C_EOF > "$conf_dir/knoe.cfg"
[User]
NAMESPACE = test-ns
SERVICE_NAMESPACE = test-system
KNOE_HOME = $KNOE_HOME
KNOE_SERVICE = $TMP_DIR/service
[Global]
DEPLOYMENT_MODE = k3s
${cnpg_instances_cfg}
C_EOF
# `etc/knoe_cfg.sh` sources `$KNOE_HOME/env.sh` unconditionally if present.
# The repo's generated `env.sh` hard-codes KNOE_CONF to the real conf dir,
# so for tests we provide a minimal env.sh that points KNOE_CONF at our temp config.
cat <<E_EOF > "$env_home/env.sh"
#!/usr/bin/env bash
export KNOE_HOME="$KNOE_HOME"
export KNOE_CONF="$conf_dir"
export KNOE_SERVICE="$TMP_DIR/service"
E_EOF
chmod +x "$env_home/env.sh"
rm -f "$TMP_DIR/applied.yaml"
rm -f "$TMP_DIR/mock_cluster_instances"
rm -f "$TMP_DIR/mock_calls.log"
set +e
MOCK_READY_NODES="$mock_ready_nodes" MOCK_DB_NODES="$mock_db_nodes" \
CNPG_DATA_SELECTOR_JSON="$data_selector_env" \
CNPG_WAL_SELECTOR_JSON="$wal_selector_env" \
BARMAN_CRD_TIMEOUT=5 BARMAN_TLS_TIMEOUT=5 BARMAN_DEPLOY_TIMEOUT=5 BARMAN_SERVICE_TIMEOUT=5 \
CNPG_STORAGE_VALIDATE_TIMEOUT=5 \
KNOE_HOME="$env_home" \
bash "$SCRIPT_UNDER_TEST" --mode k3s start >/dev/null 2>"$TMP_DIR/stderr-${label}"
local rc=$?
set -e
if [[ $rc -ne 0 ]]; then
echo "FAILURE: init_cloudnative_pg.sh returned rc=$rc for case '${label}'"
sed -n '1,200p' "$TMP_DIR/mock_calls.log" || true
sed -n '1,200p' "$TMP_DIR/stderr-${label}" || true
exit 1
fi
if [[ ! -f "$TMP_DIR/applied.yaml" ]]; then
echo "FAILURE: did not capture applied manifest for case '${label}'"
sed -n '1,200p' "$TMP_DIR/mock_calls.log" || true
sed -n '1,200p' "$TMP_DIR/stderr-${label}" || true
exit 1
fi
}
assert_applied_has_namespace_cluster_selectors() {
local label="$1"
local ns_count cluster_count
ns_count=$(grep -cE '^\s*knoe\.io/namespace:\s*test-ns\s*$' "$TMP_DIR/applied.yaml" || true)
cluster_count=$(grep -cE '^\s*knoe\.io/cluster:\s*knoe-db\s*$' "$TMP_DIR/applied.yaml" || true)
if (( ns_count < 2 )); then
echo "FAILURE: expected namespace selectors for both data and wal in case '${label}'"
sed -n '1,200p' "$TMP_DIR/applied.yaml" || true
exit 1
fi
if (( cluster_count < 2 )); then
echo "FAILURE: expected cluster selectors for both data and wal in case '${label}'"
sed -n '1,200p' "$TMP_DIR/applied.yaml" || true
exit 1
fi
}
# Default: manifest must be label-based (no hostname pinning) and reconcile must scale down to 1
# when only one node is Ready/schedulable.
run_case "default" "" 1 1
assert_applied_has_namespace_cluster_selectors "default"
if grep -qE '^\s*- key:\s*kubernetes\\.io/hostname\s*$|myrddin\\.knoe\\.org' "$TMP_DIR/applied.yaml"; then
echo "FAILURE: expected no hostname pinning in applied manifest (default)"
sed -n '1,160p' "$TMP_DIR/applied.yaml" || true
exit 1
fi
if ! grep -qE '^\s*- key:\s*node\.kubernetes\.io/instance-type\s*$' "$TMP_DIR/applied.yaml"; then
echo "FAILURE: expected k3s instance-type node affinity in applied manifest (default)"
sed -n '1,160p' "$TMP_DIR/applied.yaml" || true
exit 1
fi
if ! grep -qE '^\s*- k3s\s*$' "$TMP_DIR/applied.yaml"; then
echo "FAILURE: expected node affinity value k3s in applied manifest (default)"
sed -n '1,160p' "$TMP_DIR/applied.yaml" || true
exit 1
fi
if ! grep -qE '^\s*podAntiAffinityType:\s*preferred\s*$' "$TMP_DIR/applied.yaml"; then
echo "FAILURE: expected podAntiAffinityType=preferred in applied manifest (default)"
sed -n '1,160p' "$TMP_DIR/applied.yaml" || true
exit 1
fi
if ! grep -qE '^\s*topologyKey:\s*kubernetes\.io/hostname\s*$' "$TMP_DIR/applied.yaml"; then
echo "FAILURE: expected topologyKey=kubernetes.io/hostname in applied manifest (default)"
sed -n '1,160p' "$TMP_DIR/applied.yaml" || true
exit 1
fi
if [[ "$(cat "$TMP_DIR/mock_cluster_instances" 2>/dev/null || true)" != "1" ]]; then
echo "FAILURE: expected reconcile to patch instances to 1 under degraded startup"
sed -n '1,200p' "$TMP_DIR/mock_calls.log" || true
exit 1
fi
# Regression: selector JSON env vars may be absent in direct shell invocation, but
# manifest must still be patched with namespace+cluster selector labels before validation.
run_case "no-selector-env" "" 1 1 "missing"
assert_applied_has_namespace_cluster_selectors "no-selector-env"
if ! grep -q "patch --local" "$TMP_DIR/mock_calls.log"; then
echo "FAILURE: expected local CNPG selector patch to run in case 'no-selector-env'"
sed -n '1,200p' "$TMP_DIR/mock_calls.log" || true
exit 1
fi
# Override: when capacity allows (2 ready db nodes), CNPG_INSTANCES must be honored (cap=2)
run_case "override" "CNPG_INSTANCES = 2" 2 2
if [[ "$(cat "$TMP_DIR/mock_cluster_instances" 2>/dev/null || true)" != "2" ]]; then
echo "FAILURE: expected reconcile to patch instances to 2 when 2 db nodes are available"
sed -n '1,200p' "$TMP_DIR/mock_calls.log" || true
exit 1
fi
echo "SUCCESS"