Use flock to avoid lock file stalling

containers-scripts/update-mainnet-chainstate.sh

@@ -11,20 +11,26 @@ fi
 
 # Shared volume path for snapshot storage
 SNAPSHOT_DIR="/shared_volume"
-
 # Paths for chainstate directory in each container
 CHAINSTATE_DIR="/root/.bitcoin/chainstate"
-
 # Paths for the timestamp file that marks when the snapshot was last downloaded
 TIMESTAMP_FILE="$SNAPSHOT_DIR/last_download_timestamp"
-
 # Backup URL and interval settings
 BACKUP_BASE_URL="http://75.119.150.111/backup"
 DOWNLOAD_INTERVAL_DAYS=1  # 1 day = 24 hours
-
 # Maximum retries for downloading
 MAX_RETRIES=30
 RETRY_INTERVAL=60  # 1 minute retry interval
+# Lock settings
+LOCK_FILE="$SNAPSHOT_DIR/download.lock"
+
+# Function to clean up the lock file
+cleanup() {
+    rm -f "$LOCK_FILE"
+}
+
+# Trap to ensure the lock file is removed on exit
+trap cleanup EXIT
 
 # Check if the local chainstate directory is updated for this container
 if [ -d "$CHAINSTATE_DIR" ]; then
@@ -46,23 +52,11 @@ CURRENT_UTC=$(date -u +"%Y-%m-%d_%H-UTC")
 BACKUP_FILE_NAME="backup_mainnet_blocks_chainstate_$CURRENT_UTC.tar.gz"
 BACKUP_URL="$BACKUP_BASE_URL/$BACKUP_FILE_NAME"
 BACKUP_HASH_URL="$BACKUP_URL.sha256"
-
 BACKUP_FILE="$SNAPSHOT_DIR/$BACKUP_FILE_NAME"
 BACKUP_HASH_FILE="$SNAPSHOT_DIR/$BACKUP_FILE_NAME.sha256"
 
-# Create a lock file to prevent concurrent downloads
-LOCK_FILE="$SNAPSHOT_DIR/download.lock"
-
-# Function to clean up the lock file
-cleanup() {
-    rm -rf "$LOCK_FILE"
-}
-
-# Trap to ensure the lock file is removed on exit
-trap cleanup EXIT
-
 # Check if there is an existing backup file in the shared volume
-LATEST_BACKUP_FILE=$(find /shared_volume -maxdepth 1 -name "backup_mainnet_blocks_chainstate_*.tar.gz" -type f -printf "%T@ %p\n" | sort -n | tail -1 | cut -d' ' -f2)
+LATEST_BACKUP_FILE=$(find "$SNAPSHOT_DIR" -maxdepth 1 -name "backup_mainnet_blocks_chainstate_*.tar.gz" -type f -printf "%T@ %p\n" | sort -n | tail -1 | cut -d' ' -f2)
 
 if [ -f "$TIMESTAMP_FILE" ]; then
     TIMESTAMP_MOD_TIME=$(stat -c %Y "$TIMESTAMP_FILE")
@@ -75,15 +69,27 @@ if [ -f "$TIMESTAMP_FILE" ]; then
             tar -xzvf "$LATEST_BACKUP_FILE" -C /root/.bitcoin
             echo "Extraction complete."
             exit 0
-        else
-            echo "Warning: No backup file found. Unable to proceed with extraction."
-            exit 1
         fi
     fi
 fi
 
-# Acquire the lock before downloading
-if mkdir "$LOCK_FILE" 2>/dev/null; then
+# Use flock for the lock file to prevent concurrent downloads
+{
+    flock -n 9 || {
+        echo "Another container is currently downloading the snapshot. Waiting for download to finish..."
+        # Wait for the download to complete
+        while [ -f "$LOCK_FILE" ]; do
+            sleep 5  # Wait for 5 seconds before checking again
+        done
+        echo "Download finished. Proceeding to extraction."
+        # Check again if the backup file exists after the wait
+        if [ -f "$LATEST_BACKUP_FILE" ]; then
+            tar -xzvf "$LATEST_BACKUP_FILE" -C /root/.bitcoin
+            echo "Extraction complete."
+            exit 0
+        fi
+    }
+
     echo "Lock acquired. Proceeding with snapshot download..."
 
     # Remove any old backup files in the shared volume
@@ -120,39 +126,19 @@ if mkdir "$LOCK_FILE" 2>/dev/null; then
 
     if [ $success -eq 0 ]; then
         echo "Failed to download the snapshot after $MAX_RETRIES attempts. Aborting."
-        cleanup  # Clean up the lock file
         exit 1
     fi
 
-    cleanup  # Clean up the lock file
-
-    # Extract the snapshot to the container's local chainstate directory
-    echo "Extracting the snapshot to the container's local chainstate..."
-    tar -xzvf "$BACKUP_FILE" -C /root/.bitcoin
-
     # Update the timestamp file to mark the download time
     touch "$TIMESTAMP_FILE"
 
-    echo "Update and extraction completed successfully."
-else
-    echo "Another container is currently downloading the snapshot. Waiting for it to complete..."
-
-    # Wait for the download to finish and the lock to be released
-    while [ -d "$LOCK_FILE" ]; do
-        sleep 5  # Check every 5 seconds
-    done
+    # Release the lock file to allow other containers to start extracting
+} 9>"$LOCK_FILE"  # Using file descriptor 9 for the lock file
 
-    # After the download, proceed with extraction
-    echo "Snapshot downloaded by another container. Checking for backup file..."
-
-    if [ -f "$BACKUP_HASH_FILE" ]; then
-        echo "Proceeding to extraction for this container."
-        tar -xzvf "$BACKUP_FILE" -C /root/.bitcoin
-        echo "Extraction complete."
-    else
-        echo "Expected backup file $BACKUP_FILE does not exist. Unable to proceed with extraction."
-        exit 1
-    fi
-fi
+# Extract the downloaded snapshot
+echo "Extracting the downloaded snapshot..."
+tar -xzvf "$BACKUP_FILE" -C /root/.bitcoin
+echo "Extraction complete."
 
-cleanup  # Clean up the lock file if it was acquired
+# If we reach here, the download was successful, and other containers can now extract the snapshot
+echo "Update and extraction completed successfully."