Write a Python function using Apache Spark Structured Streaming that reads messages from a Kafka topic, counts occurrences of each 'event_type' in real time, and writes the result to a BigQuery table. Handle connection and streaming errors gracefully.

To solve this, create a SparkSession with the BigQuery connector and Kafka packages. Read the Kafka stream with spark.readStream.format("kafka"), specifying the topic, bootstrap servers, and starting offsets. The Kafka value is a binary column; cast it to string and parse the JSON payload to extract the event_type field. Use groupBy("event_type").count() to aggregate per micro‑batch. Write the aggregated DataFrame to BigQuery using the writeStream API with the BigQuery connector, specifying a checkpoint location for fault tolerance. Wrap the streaming query in a try/except block and use foreachBatch to handle errors per batch: inside the batch function, attempt to write to BigQuery and catch exceptions, logging them and optionally retrying with exponential back‑off. Use spark.conf.set("spark.sql.streaming.checkpointLocation", "/tmp/checkpoints") to persist state. Complexity is O(n) per micro‑batch, where n is the number of events in that batch. The solution is robust because checkpointing and error handling ensure the stream can recover from transient network or BigQuery outages without data loss.