avm/hassio-addons
1
0
Fork 0
mirror of https://github.com/alexbelgium/hassio-addons.git synced 2026-06-03 22:34:11 +02:00
Code Issues Packages Projects Releases Wiki Activity

Improve services monitoring

Browse Source
This commit is contained in:
Alexandre
2025-02-04 21:00:42 +01:00
committed by GitHub
parent e8f9009b51
commit 884370083b
1 changed files with 34 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

41
birdnet-pi/rootfs/custom-services.d/30-monitoring.sh
View File

@@ -43,12 +43,13 @@ chown -R pi:pi "$INGEST_DIR" || log_yellow "Could not change ownership for $INGE
chmod -R 755 "$INGEST_DIR" || log_yellow "Could not set permissions for $INGEST_DIR" chmod -R 755 "$INGEST_DIR" || log_yellow "Could not set permissions for $INGEST_DIR"
# Services to monitor # Services to monitor
SERVICES=(birdnet_analysis chart_viewer spectrogram_viewer icecast2 birdnet_recording birdnet_log birdnet_stats) SERVICES=(birdnet_analysis chart_viewer spectrogram_viewer birdnet_recording birdnet_log birdnet_stats)
# Notification settings # Notification settings
NOTIFICATION_INTERVAL=1800 # seconds (30 minutes) NOTIFICATION_INTERVAL=1800 # seconds (30 minutes)
last_notification_time=0 last_notification_time=0
issue_reported=0 # 1 = an issue was reported, 0 = system is normal issue_reported=0 # 1 = an issue was reported, 0 = system is normal
declare -A SERVICE_INACTIVE_COUNT=()
# Disk usage threshold (percentage) # Disk usage threshold (percentage)
DISK_USAGE_THRESHOLD=95 DISK_USAGE_THRESHOLD=95
@@ -191,20 +192,46 @@ check_queue() {
} }
check_services() { check_services() {
local inactive_services=() local any_inactive=0
for service in "${SERVICES[@]}"; do for service in "${SERVICES[@]}"; do
if [[ "$(systemctl is-active "$service")" != "active" ]]; then if [[ "$(systemctl is-active "$service")" != "active" ]]; then
inactive_services+=("$service")
# Increment the service's inactive counter
SERVICE_INACTIVE_COUNT["$service"]=$(( SERVICE_INACTIVE_COUNT["$service"] + 1 ))
if (( SERVICE_INACTIVE_COUNT["$service"] == 1 )); then
# First time we see it inactive in a row => Try restarting (silent recovery attempt)
log_yellow "$(date) INFO: Service '$service' is inactive. Attempting to start..."
systemctl start "$service"
any_inactive=1
elif (( SERVICE_INACTIVE_COUNT["$service"] == 2 )); then
# Second consecutive time => Send an alert
log_red "$(date) INFO: Service '$service' is still inactive after restart attempt."
apprisealert "Service '$service' remains inactive after restart attempt."
any_inactive=1
else
# If it is still inactive beyond 2 checks, keep trying or do advanced actions
log_red "$(date) INFO: Service '$service' has been inactive for ${SERVICE_INACTIVE_COUNT["$service"]} checks in a row."
any_inactive=1
fi
else
# Service is active => reset the inactive counter
if (( SERVICE_INACTIVE_COUNT["$service"] > 0 )); then
log_green "$(date) INFO: Service '$service' is back to active. Resetting counter."
fi
SERVICE_INACTIVE_COUNT["$service"]=0
fi fi
done done
if (( ${#inactive_services[@]} == 0 )); then if (( any_inactive == 0 )); then
# Example: "Tue Feb 4 20:18:50 CET 2025 INFO: All services are active"
log_green "$(date) INFO: All services are active" log_green "$(date) INFO: All services are active"
return 0 return 0
else else
log_red "$(date) INFO: Some services are NOT active: ${inactive_services[*]}" log_red "$(date) INFO: One or more services are inactive"
apprisealert "One or more services inactive: ${inactive_services[*]}"
return 1 return 1
fi fi
} }