You've already forked nrpe
Ludovic Cartier
138 lines
4.1 KiB
Bash
Executable File
138 lines
4.1 KiB
Bash
Executable File
#!/usr/bin/env bash
|
|
# check_k8s_deployments
|
|
# Vérifie les Deployments Kubernetes: availableReplicas < spec.replicas
|
|
# Retour: 0=OK, 1=WARNING, 2=CRITICAL, 3=UNKNOWN
|
|
#
|
|
# Usage:
|
|
# sudo /usr/lib/nagios/plugins/check_k8s_deployments [--warn N] [--crit M] [--ignore-ns ns1,ns2] [--namespaces ns1,ns2] [--age-min MINUTES]
|
|
#
|
|
# Exemples:
|
|
# sudo /usr/lib/nagios/plugins/check_k8s_deployments --crit 1
|
|
# sudo /usr/lib/nagios/plugins/check_k8s_deployments --ignore-ns kube-system,monitoring
|
|
#
|
|
set -euo pipefail
|
|
|
|
WARN=${WARN:-0} # nombre de deploys en erreur pour WARNING
|
|
CRIT=${CRIT:-1} # nombre de deploys en erreur pour CRITICAL par défaut (1 => tout problème -> CRITICAL)
|
|
IGNORE_NS=""
|
|
INCLUDE_NS=""
|
|
AGE_MIN=0
|
|
|
|
print_usage() {
|
|
cat <<EOF
|
|
Usage: $0 [--warn N] [--crit M] [--ignore-ns ns1,ns2] [--namespaces ns1,ns2] [--age-min MINUTES]
|
|
--warn N : seuil warn si >=N déploiements en erreur (default 0)
|
|
--crit M : seuil crit si >=M déploiements en erreur (default 1)
|
|
--ignore-ns LIST : comma separated namespaces to ignore (default none)
|
|
--namespaces LIST: comma separated namespaces to check only (default all)
|
|
--age-min N : ignore deployments created less than N minutes ago (avoid flapping during rollout)
|
|
EOF
|
|
}
|
|
|
|
# parse args
|
|
while [[ $# -gt 0 ]]; do
|
|
case "$1" in
|
|
--warn) WARN="$2"; shift 2;;
|
|
--crit) CRIT="$2"; shift 2;;
|
|
--ignore-ns) IGNORE_NS="$2"; shift 2;;
|
|
--namespaces) INCLUDE_NS="$2"; shift 2;;
|
|
--age-min) AGE_MIN="$2"; shift 2;;
|
|
-h|--help) print_usage; exit 3;;
|
|
*) echo "Unknown arg: $1"; print_usage; exit 3;;
|
|
esac
|
|
done
|
|
|
|
if ! command -v kubectl >/dev/null 2>&1; then
|
|
echo "UNKNOWN - kubectl not found"
|
|
exit 3
|
|
fi
|
|
|
|
# Build filter for namespace inclusion/exclusion
|
|
ignore_pattern=""
|
|
if [[ -n "$IGNORE_NS" ]]; then
|
|
IFS=',' read -ra arr <<< "$IGNORE_NS"
|
|
for ns in "${arr[@]}"; do
|
|
ignore_pattern="${ignore_pattern}|^${ns}\$"
|
|
done
|
|
# remove leading |
|
|
ignore_pattern="${ignore_pattern#|}"
|
|
fi
|
|
|
|
include_pattern=""
|
|
if [[ -n "$INCLUDE_NS" ]]; then
|
|
IFS=',' read -ra arr2 <<< "$INCLUDE_NS"
|
|
for ns in "${arr2[@]}"; do
|
|
include_pattern="${include_pattern}|^${ns}\$"
|
|
done
|
|
include_pattern="${include_pattern#|}"
|
|
fi
|
|
|
|
# result collection
|
|
# Initialize failures array to avoid "variable sans liaison" when running with set -u
|
|
failures=()
|
|
|
|
# get list: namespace, name, desired, available, creationTimestamp
|
|
mapfile -t lines < <(kubectl get deploy -A -o jsonpath='{range .items[*]}{.metadata.namespace}{"\t"}{.metadata.name}{"\t"}{.spec.replicas}{"\t"}{.status.availableReplicas}{"\t"}{.metadata.creationTimestamp}{"\n"}{end}' 2>/dev/null || true)
|
|
|
|
now_s=$(date +%s)
|
|
|
|
for line in "${lines[@]}"; do
|
|
# skip empty lines
|
|
[[ -z "${line}" ]] && continue
|
|
|
|
ns=$(echo "$line" | awk -F'\t' '{print $1}')
|
|
name=$(echo "$line" | awk -F'\t' '{print $2}')
|
|
desired=$(echo "$line" | awk -F'\t' '{print $3}')
|
|
available=$(echo "$line" | awk -F'\t' '{print $4}')
|
|
created=$(echo "$line" | awk -F'\t' '{print $5}')
|
|
|
|
# normalize
|
|
desired=${desired:-0}
|
|
available=${available:-0}
|
|
|
|
# namespace filtering
|
|
if [[ -n "$include_pattern" ]]; then
|
|
if ! echo "$ns" | egrep -q "$include_pattern"; then
|
|
continue
|
|
fi
|
|
fi
|
|
if [[ -n "$ignore_pattern" ]]; then
|
|
if echo "$ns" | egrep -q "$ignore_pattern"; then
|
|
continue
|
|
fi
|
|
fi
|
|
|
|
# age filtering
|
|
if [[ -n "$created" && "$AGE_MIN" -gt 0 ]]; then
|
|
# convert to epoch
|
|
created_s=$(date -d "$created" +%s 2>/dev/null || echo 0)
|
|
age_min=$(( (now_s - created_s) / 60 ))
|
|
if (( age_min < AGE_MIN )); then
|
|
# skip new deployments (they might be still rolling out)
|
|
continue
|
|
fi
|
|
fi
|
|
|
|
if (( available < desired )); then
|
|
failures+=("${ns}/${name} (desired=${desired},available=${available})")
|
|
fi
|
|
done
|
|
|
|
count=${#failures[@]}
|
|
|
|
if (( count == 0 )); then
|
|
echo "OK - all deployments report desired==available"
|
|
exit 0
|
|
fi
|
|
|
|
# Decide severity
|
|
if (( count >= CRIT )); then
|
|
echo "CRITICAL - ${count} deployments not available: ${failures[*]}"
|
|
exit 2
|
|
elif (( count >= WARN )); then
|
|
echo "WARNING - ${count} deployments not available: ${failures[*]}"
|
|
exit 1
|
|
else
|
|
echo "OK - ${count} deployments not available but below thresholds"
|
|
exit 0
|
|
fi |