Are counters thread safe in Micrometer

Question

I am trying to publish number of events my App is processing. This is the code I have at the receiving end:

public void process(List<String> batch) {
    logger.info ("Processing batch of size " + batch.size())
    metrics.incrementCounter(MetricsType.CONSUMER_TOTAL_PROCESSED, batch.size)
}

The class Metrics is:

public class Metrics {
    private static final Map<MetricsType, Counter> COUNTER_MAP = new ConcurrentHashMap<>();

    public Metrics(
        @Autowired MeterRegistry meterRegistry
    ) {
        COUNTER_MAP.put(
            MetricsType.CONSUMER_TOTAL_PROCESSED,
            Counter.builder("CONSUMER_TOTAL_PROCESSED").register(meterRegistry)
        );

        COUNTER_MAP.put(
            MetricsType.CONSUMER_DUPLICATE_PROCESSED,
            Counter.builder("CONSUMER_DUPLICATE_PROCESSED").register(meterRegistry)
        );
    }

    public void increment(MetricsType metricsType, int size) {
        COUNTER_MAP.get(metricsType).increment(size);
    }
}

The enum MetricsType contains all type of counters.

The process method is be invoked by 16 threads at any time. The issue I am facing is the logger which prints the count and the total count reported in grafana are way off.

Do I have to synchonize everytime I am incrementing the counter?

---

Edit - What I mean by the counts are off is, if there are two logs with size 200, then grafana should report total counter 400. I am validating this by taking a time range of 2 hours, I extract all the sizes from logs and add them.

---

If you stumble upon this and see a difference between two sources please check what is your maximum number of data points in Grafana, that was the actual issue why I thought counters may not be threadsafe.

Answer 1

The counters are thread safe, internally they use some atomic data structure for counting. Given your code I would suspect that what the logger prints is quite different from what you see in grafana because the logger prints the size of the current batch while the counter prints the sum of all batch sizes.