confdroid_prometheus/templates/prune_blocks.erb

#!/bin/bash
set -euo pipefail

# Load environment
source <%= @ps_env_file %>

LOG_FILE="/var/log/prometheus-prune.log"
echo "$(date '+%Y-%m-%d %H:%M:%S') Starting prune run" >> "$LOG_FILE"

# Find all TSDB blocks
TSDB_DIR="/var/lib/prometheus"
for block in "$TSDB_DIR"/*/; do
    [[ -d "$block" ]] || continue
    meta="$block/meta.json"
    [[ -f "$meta" ]] || continue

    # Get block min and max timestamps
    min_time=$(jq -r '.minTime' "$meta")
    max_time=$(jq -r '.maxTime' "$meta")

    # Skip very recent blocks
    block_age_hours=$(( ($(date +%s) - min_time/1000) / 3600 ))
    if (( block_age_hours < MIN_AGE_HOURS + GRACE_HOURS )); then
        echo "Skipping block $block (age ${block_age_hours}h)" >> "$LOG_FILE"
        continue
    fi

    # Verify block is in PostgreSQL
    check_sql="${CHECK_SQL_TEMPLATE//\{min\}/$min_time}"
    check_sql="${check_sql//\{max\}/$max_time}"

    exists=$(psql -h "$PGHOST" -p "$PGPORT" -U "$PGUSER" -d "$PGDATABASE" -t -c "$check_sql" | xargs)

    if [[ "$exists" != "1" ]]; then
        echo "Block $block not found in PostgreSQL, skipping deletion" >> "$LOG_FILE"
        continue
    fi

    # Dry-run or delete
    if [[ "$DRY_RUN" == "true" ]]; then
        echo "[DRY-RUN] Would delete block $block" >> "$LOG_FILE"
    else
        echo "Deleting block $block" >> "$LOG_FILE"
        rm -rf "$block"
    fi
done

echo "$(date '+%Y-%m-%d %H:%M:%S') Prune run finished" >> "$LOG_FILE"