Erich Blume
ff557b0437
- New borgmatic_metrics ansible role: - Collects borg repository and archive stats hourly - Exports to Prometheus via textfile collector - Metrics: repo size, dedup ratio, archive count, last backup stats - New Grafana dashboard "Borgmatic Backups": - Repository status and disk usage - Backup size trends over time - New data per backup (bar chart) - Deduplication ratio - Time since last backup with alerting thresholds Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
131 lines
5.1 KiB
Django/Jinja
131 lines
5.1 KiB
Django/Jinja
#!/bin/bash
|
|
# {{ ansible_managed }}
|
|
# Collects borg backup metrics for node_exporter textfile collector
|
|
|
|
set -euo pipefail
|
|
|
|
export BORG_PASSCOMMAND="{{ borgmatic_metrics_passcommand }}"
|
|
BORG_REPO="{{ borgmatic_metrics_repo }}"
|
|
OUTPUT_FILE="{{ borgmatic_metrics_dir }}/borgmatic.prom"
|
|
TEMP_FILE="${OUTPUT_FILE}.tmp"
|
|
|
|
# Check if borg is available via mise
|
|
BORG_CMD="mise x -- borg"
|
|
|
|
# Get repository info
|
|
repo_json=$($BORG_CMD info --json "$BORG_REPO" 2>/dev/null) || {
|
|
echo "Failed to get borg repo info" >&2
|
|
# Write down metric
|
|
cat > "$TEMP_FILE" << 'EOF'
|
|
# HELP borgmatic_up Borg backup repository is accessible
|
|
# TYPE borgmatic_up gauge
|
|
borgmatic_up 0
|
|
EOF
|
|
mv "$TEMP_FILE" "$OUTPUT_FILE"
|
|
exit 0
|
|
}
|
|
|
|
# Get archive list
|
|
archives_json=$($BORG_CMD list --json "$BORG_REPO" 2>/dev/null) || {
|
|
echo "Failed to list borg archives" >&2
|
|
exit 1
|
|
}
|
|
|
|
# Extract repository stats
|
|
total_size=$(echo "$repo_json" | jq -r '.cache.stats.total_size')
|
|
total_csize=$(echo "$repo_json" | jq -r '.cache.stats.total_csize')
|
|
unique_size=$(echo "$repo_json" | jq -r '.cache.stats.unique_size')
|
|
unique_csize=$(echo "$repo_json" | jq -r '.cache.stats.unique_csize')
|
|
total_chunks=$(echo "$repo_json" | jq -r '.cache.stats.total_chunks')
|
|
unique_chunks=$(echo "$repo_json" | jq -r '.cache.stats.total_unique_chunks')
|
|
|
|
# Count archives
|
|
archive_count=$(echo "$archives_json" | jq -r '.archives | length')
|
|
|
|
# Get last archive info
|
|
last_archive_name=$(echo "$archives_json" | jq -r '.archives[-1].name // empty')
|
|
|
|
if [ -n "$last_archive_name" ]; then
|
|
# Get detailed info for the last archive
|
|
last_archive_json=$($BORG_CMD info --json "${BORG_REPO}::${last_archive_name}" 2>/dev/null) || {
|
|
echo "Failed to get last archive info" >&2
|
|
last_archive_json=""
|
|
}
|
|
|
|
if [ -n "$last_archive_json" ]; then
|
|
last_original_size=$(echo "$last_archive_json" | jq -r '.archives[0].stats.original_size')
|
|
last_compressed_size=$(echo "$last_archive_json" | jq -r '.archives[0].stats.compressed_size')
|
|
last_deduplicated_size=$(echo "$last_archive_json" | jq -r '.archives[0].stats.deduplicated_size')
|
|
last_nfiles=$(echo "$last_archive_json" | jq -r '.archives[0].stats.nfiles')
|
|
last_start=$(echo "$last_archive_json" | jq -r '.archives[0].start')
|
|
last_end=$(echo "$last_archive_json" | jq -r '.archives[0].end')
|
|
last_duration=$(echo "$last_archive_json" | jq -r '.archives[0].duration')
|
|
|
|
# Convert timestamp to unix epoch
|
|
last_timestamp=$(date -j -f "%Y-%m-%dT%H:%M:%S" "${last_start%.*}" "+%s" 2>/dev/null || echo "0")
|
|
fi
|
|
fi
|
|
|
|
# Write metrics
|
|
cat > "$TEMP_FILE" << EOF
|
|
# HELP borgmatic_up Borg backup repository is accessible
|
|
# TYPE borgmatic_up gauge
|
|
borgmatic_up 1
|
|
|
|
# HELP borgmatic_repo_original_size_bytes Total original size of all archives (sum of what each backup contains)
|
|
# TYPE borgmatic_repo_original_size_bytes gauge
|
|
borgmatic_repo_original_size_bytes $total_size
|
|
|
|
# HELP borgmatic_repo_compressed_size_bytes Total compressed size of all archives
|
|
# TYPE borgmatic_repo_compressed_size_bytes gauge
|
|
borgmatic_repo_compressed_size_bytes $total_csize
|
|
|
|
# HELP borgmatic_repo_deduplicated_size_bytes Actual disk usage after deduplication (unique data)
|
|
# TYPE borgmatic_repo_deduplicated_size_bytes gauge
|
|
borgmatic_repo_deduplicated_size_bytes $unique_csize
|
|
|
|
# HELP borgmatic_repo_total_chunks Total number of chunks across all archives
|
|
# TYPE borgmatic_repo_total_chunks gauge
|
|
borgmatic_repo_total_chunks $total_chunks
|
|
|
|
# HELP borgmatic_repo_unique_chunks Number of unique chunks (after deduplication)
|
|
# TYPE borgmatic_repo_unique_chunks gauge
|
|
borgmatic_repo_unique_chunks $unique_chunks
|
|
|
|
# HELP borgmatic_archive_count Number of archives in the repository
|
|
# TYPE borgmatic_archive_count gauge
|
|
borgmatic_archive_count $archive_count
|
|
EOF
|
|
|
|
# Add last archive metrics if available
|
|
if [ -n "${last_original_size:-}" ]; then
|
|
cat >> "$TEMP_FILE" << EOF
|
|
|
|
# HELP borgmatic_last_archive_original_size_bytes Original size of the last archive (data being backed up)
|
|
# TYPE borgmatic_last_archive_original_size_bytes gauge
|
|
borgmatic_last_archive_original_size_bytes $last_original_size
|
|
|
|
# HELP borgmatic_last_archive_compressed_size_bytes Compressed size of the last archive
|
|
# TYPE borgmatic_last_archive_compressed_size_bytes gauge
|
|
borgmatic_last_archive_compressed_size_bytes $last_compressed_size
|
|
|
|
# HELP borgmatic_last_archive_deduplicated_size_bytes Deduplicated size of last archive (new data added)
|
|
# TYPE borgmatic_last_archive_deduplicated_size_bytes gauge
|
|
borgmatic_last_archive_deduplicated_size_bytes $last_deduplicated_size
|
|
|
|
# HELP borgmatic_last_archive_files Number of files in the last archive
|
|
# TYPE borgmatic_last_archive_files gauge
|
|
borgmatic_last_archive_files $last_nfiles
|
|
|
|
# HELP borgmatic_last_archive_timestamp Unix timestamp of the last backup
|
|
# TYPE borgmatic_last_archive_timestamp gauge
|
|
borgmatic_last_archive_timestamp $last_timestamp
|
|
|
|
# HELP borgmatic_last_archive_duration_seconds Duration of the last backup in seconds
|
|
# TYPE borgmatic_last_archive_duration_seconds gauge
|
|
borgmatic_last_archive_duration_seconds ${last_duration:-0}
|
|
EOF
|
|
fi
|
|
|
|
# Atomic move
|
|
mv "$TEMP_FILE" "$OUTPUT_FILE"
|