[ISSUE #494] support prometheus

[Slideee]

What is the purpose of the change

#494

Brief changelog

XX

Verifying this change

XXXX

Follow this checklist to help us incorporate your contribution quickly and easily. Notice, it would be helpful if you could finish the following 5 checklist(the last one is not necessary)before request the community to review your PR.

• Make sure there is a Github issue filed for the change (usually before you start working on it). Trivial changes like typos do not require a Github issue. Your pull request should address just this issue, without pulling in other changes - one PR resolves one issue.
• Format the pull request title like [ISSUE #123] Fix UnknownException when host config not exist. Each commit in the pull request should have a meaningful subject line and body.
• Write a pull request description that is detailed enough to understand what the pull request does, how, and why.
• Write necessary unit-test(over 80% coverage) to verify your logic correction, more mock a little better when cross module dependency exist. If the new feature or significant change is committed, please remember to add integration-test in test module.
• Run mvn -B clean apache-rat:check findbugs:findbugs checkstyle:checkstyle to make sure basic checks pass. Run mvn clean install -DskipITs to make sure unit-test pass. Run mvn clean test-compile failsafe:integration-test to make sure integration-test pass.
• If this contribution is large, please file an Apache Individual Contributor License Agreement.

[RockteMQ-AI]

Summary

This PR modifies 9 file(s) with 607 lines of diff. No test changes detected — consider adding test coverage.

---

Automated review by github-manager-bot

Additional notes (not anchored to a changed line)

• [INFO] metric-exporter/pom.xml:1 — Large diff (607 lines). Consider breaking into smaller, focused PRs for easier review. (line outside diff)

[RockteMQ-AI]

No test changes detected alongside source modifications. Consider adding tests to cover the changes.

[RockteMQ-AI]

Summary

PR received and logged for review. This PR requires detailed code review by a maintainer.

Diff size: 607 lines
Author: Slideee (CONTRIBUTOR)

---

Automated review by RockteMQ-AI

[RockteMQ-AI]

Summary

Review of PR #495: [ISSUE #494] support prometheus

Findings: 9 issue(s) identified (2 critical).
CLA: unknown

Please address the inline comments above.

---

Automated review by github-manager-bot

[RockteMQ-AI]

The createSample method completely ignores the additionalLabelNames and additionalLabelValues parameters. These carry the 'quantile' label (e.g., name="quantile", value="0.75") that differentiates histogram percentiles in Prometheus SUMMARY metrics. By always using SOURCE_TASK_LABEL_NAMES and discarding the additional labels, all percentile samples for a histogram end up with an identical metric name and label set, making them indistinguishable and causing Prometheus to reject or arbitrarily deduplicate them. The method should merge additionalLabelNames/additionalLabelValues into the output sample's label names and values.

[RockteMQ-AI]

sanitizeMetricName(dropwizardName + suffix) appends the suffix to the full dropwizardName string before parsing, but sanitizeMetricName extracts only the 2nd comma-separated field (split(":")[1].split(",")[1]). The suffix (e.g., "_count") lands on the last field and is silently lost. This means the histogram count sample (which uses nameSuffix="_count") gets the same metric name as the percentile samples, producing conflicting samples with the same name and label set. The suffix should be appended to the extracted metric name, not to the raw input string.

[RockteMQ-AI]

sanitizeLabelValues performs unchecked array indexing on split(":") and split(",") results, accessing indices 0 through 4 (split[4]) without any bounds validation. If any metric in the registry has a name that does not match the expected 'prefix:group,name,type,connector=X,task=Y' format (e.g., metrics registered by the framework itself or third-party libraries), this will throw ArrayIndexOutOfBoundsException, crashing the entire /metrics endpoint. sanitizeMetricName (line 37) has the same issue. Consider validating the split array length or wrapping in a try-catch that skips malformed metric names.

[RockteMQ-AI]

DropwizardExports is registered with the global CollectorRegistry.defaultRegistry but is never unregistered when the Worker is stopped. On Worker restart (e.g., connector reconfiguration), registering the same collector type again will throw IllegalArgumentException('Collector already registered'), breaking the connector lifecycle. The registration should either use a dedicated CollectorRegistry (not the global singleton), or unregister the collector in the Worker's stop/shutdown method.

[RockteMQ-AI]

The Javalin 'embeddedApp' instance for the metrics server is a local variable in the constructor and is never stored as a field. This means there is no way to stop or shut down the metrics Jetty server when the RestHandler or connect runtime is stopped, causing a port and thread resource leak. The embeddedApp reference should be stored as a field and stopped in the appropriate shutdown method.

[RockteMQ-AI]

Stat.HistogramType.valueOf(metricName.getType()) throws IllegalArgumentException if the metric name's type field does not match a known enum constant. Since this is called inside collect() which iterates over ALL metrics in the registry, a single metric with an unrecognized type will crash the entire metrics collection, making the /metrics endpoint return an error for all metrics. Consider wrapping this in a try-catch that skips the individual metric and logs a warning, similar to how fromGauge handles invalid types.

[RockteMQ-AI]

The default case of the switch in fromSnapshotAndCount creates a duplicate sample: two samples both with quantile="0.5" and snapshot.getMedian(). The original Prometheus DropwizardExports only includes the median sample once. This duplicate should be removed — the second createSample call with Arrays.asList("0.5") and snapshot.getMedian() is redundant.

[RockteMQ-AI]

The parse(HttpServletRequest) method is dead code — it is never called anywhere in the class. It appears to be copied from an upstream Prometheus servlet implementation but was not wired into doGet. Either remove it or use it to support the name[] query parameter for filtering which metrics are returned.

[RockteMQ-AI]

No test coverage is added for any of the new classes (DropwizardExports, PrometheusSampleBuilder, PrometheusMetricsServlet). Given the complex string-parsing logic in PrometheusSampleBuilder and the metric-type dispatch in DropwizardExports, unit tests are especially important to verify correct behavior with well-formed and malformed metric names, and to prevent regressions in the quantile label and suffix handling.

[RockteMQ-AI]

Issue Evaluation

Category: enhancement | Status: Evaluated (PR-like submission)

This issue references #494 and proposes adding Prometheus monitoring support.

Note: This appears to be a PR submission. If you have implementation code ready, please submit it as a pull request directly.

Feasibility: Adding Prometheus metrics exposure is a valuable enhancement for observability.

Age: This issue is from May 2023. If this feature is still desired, please confirm or submit a PR.

---

Automated evaluation by RockteMQ-AI

[RockteMQ-AI]

Summary

This PR adds Prometheus metrics export support to the Connect HTTP connector. The overall architecture is sound — using Dropwizard as the metrics registry and exporting via Prometheus text format is a standard approach.

However, there is a critical parsing bug in PrometheusSampleBuilder that will crash on metric names without the expected name:tag format.

Findings

• [Critical] PrometheusSampleBuilder.java:60 — split(":")[1] and split(",")[1] lack bounds checking → ArrayIndexOutOfBoundsException
• [Warning] WorkerConfig.java:120 — No validation on exporterPort range

Suggestions

1. Add defensive parsing in PrometheusSampleBuilder.buildMetricName() to handle metric names that don't follow the name:tag convention
2. Add port range validation for exporterPort
3. Consider adding a unit test for PrometheusSampleBuilder with edge cases (no colon, no comma, empty name)

---

Automated review by github-manager-bot

Additional notes (not anchored to a changed line)

• [CRITICAL] connectors/rocketmq-connect-http/src/main/java/org/apache/rocketmq/connect/http/metrics/PrometheusSampleBuilder.java:60 — [Critical] name.split(":")[1] will throw ArrayIndexOutOfBoundsException if the metric name does not contain a colon (e.g., a plain metric name like my_metric). Add a bounds check:

String[] parts = name.split(":");
String metricName = parts.length > 1 ? parts[1] : parts[0];

Similarly, parts[1].split(",")[1] on line 61 can throw if there is no comma. Consider defensive parsing. (line outside diff)

• [WARNING] connect/connector-runtime/src/main/java/org/apache/rocketmq/connect/runtime/WorkerConfig.java:120 — [Warning] exporterPort defaults to 5557 with no validation. Consider adding a port range check (1-65535) in the setter or during config initialization to prevent silent failures when an invalid port is configured. (line outside diff)

[RockteMQ-AI]

Summary

The Prometheus integration has multiple unguarded string-parsing paths that will crash the /metrics endpoint on unexpected metric names, a duplicate-sample bug in the histogram default branch, a leaked Jetty server with no shutdown hook, and no test coverage.

Findings

• [CRITICAL] metric-exporter/src/main/java/org/apache/rocketmq/connect/metrics/DropwizardExports.java:155 — The default case in fromSnapshotAndCount adds the median (quantile "0.5") sample twice — the first two entries in the Arrays.asList(...) are identical. This produces duplicate samples in every Prometheus scrape for any histogram type that falls through to default, which will confuse or error in Prometheus ingestion.
• [CRITICAL] metric-exporter/src/main/java/org/apache/rocketmq/connect/metrics/DropwizardExports.java:127 — Stat.HistogramType.valueOf(metricName.getType()) will throw IllegalArgumentException if the metric's type string doesn't exactly match an enum constant. There is no try/catch, so a single unrecognized histogram metric name will abort the entire collect() call, breaking the whole /metrics endpoint. Wrap this in a try/catch and log+skip the offending metric.
• [CRITICAL] metric-exporter/src/main/java/org/apache/rocketmq/connect/metrics/PrometheusSampleBuilder.java:35 — sanitizeMetricName splits on ":" then "," and accesses index [1] without bounds checking. If any Dropwizard metric name doesn't follow the exact expected format (e.g. internal JVM metrics, or third-party metrics), this throws ArrayIndexOutOfBoundsException and crashes the scrape. Add defensive parsing with a fallback.
• [CRITICAL] metric-exporter/src/main/java/org/apache/rocketmq/connect/metrics/PrometheusSampleBuilder.java:39 — sanitizeLabelValues has the same unguarded split pattern and accesses split[0] through split[4]. Any metric name with fewer than 5 comma-separated segments will throw ArrayIndexOutOfBoundsException. Additionally, split[3].substring(var3.indexOf("=") + 1) will misbehave if var3 doesn't contain "=" (indexOf returns -1, yielding substring(0) — silently wrong rather than failing loudly).
• [CRITICAL] rocketmq-connect-runtime/src/main/java/org/apache/rocketmq/connect/runtime/rest/RestHandler.java:78 — The Prometheus Jetty Server is started in the constructor but the embeddedApp reference is local and never stored. This means: (1) there is no way to shut it down on application stop — the server and its thread pool will leak; (2) if exporterPort is already bound, the exception propagates uncaught from the constructor, preventing the entire Worker from starting. Store the reference and wire it into the shutdown lifecycle.
• [WARNING] rocketmq-connect-runtime/src/main/java/org/apache/rocketmq/connect/runtime/connectorwrapper/Worker.java:154 — CollectorRegistry.defaultRegistry.register(...) is called with no corresponding unregister(). If the Worker is re-created (e.g., in tests or on restart within the same JVM), the second registration will throw IllegalArgumentException: Collector already registered. Use an instance-scoped CollectorRegistry instead of the global default, or add unregister logic on shutdown.
• [WARNING] rocketmq-connect-runtime/src/main/java/org/apache/rocketmq/connect/runtime/rest/PrometheusMetricsServlet.java:60 — This class hand-rolls the Prometheus text exposition format instead of using io.prometheus.client.exporter.common.TextFormat (which is part of simpleclient_common). The manual writer uses raw integer literals (123, 125, 10, 32) instead of char literals ('{', '}', '\n', ' '), hurting readability. More importantly, it is missing # HELP and # TYPE lines that Prometheus expects, so the output is not fully spec-compliant.
• [WARNING] rocketmq-connect-runtime/src/main/java/org/apache/rocketmq/connect/runtime/rest/PrometheusMetricsServlet.java:120 — The parse() method reads name[] query parameters for metric filtering but is never called anywhere. Dead code — either wire it into doGet to support selective metric export or remove it.
• [INFO] rocketmq-connect-runtime/src/main/java/org/apache/rocketmq/connect/runtime/rest/PrometheusMetricsServlet.java:124 — doPost delegates to doGet. Exposing metrics via POST is unconventional and unnecessary — Prometheus only scrapes via GET. Remove doPost to reduce the attack surface.
• [WARNING] rocketmq-connect-runtime/src/main/java/org/apache/rocketmq/connect/runtime/config/WorkerConfig.java:61 — The exporterPort field (default 5557) is not populated from the properties map like other config fields (e.g., httpPort). There is no buildWorkerConfig() / init logic that reads it from a config key, so the setter is the only way to change it. Add it to the config loading path to be consistent with the rest of the class.
• [INFO] metric-exporter/src/main/java/org/apache/rocketmq/connect/metrics/DropwizardExports.java:234 — Missing newline at end of file. Minor, but some tools and checkers flag this.
• [INFO] metric-exporter/src/main/java/org/apache/rocketmq/connect/metrics/DropwizardExports.java:107 — new Long(counter.getCount()).doubleValue() uses the deprecated Long constructor. Replace with (double) counter.getCount() or Long.valueOf(...).doubleValue().
• [WARNING] rocketmq-connect-runtime/src/main/java/org/apache/rocketmq/connect/runtime/rest/PrometheusMetricsServlet.java:42 — No tests are included for any of the new classes (DropwizardExports, PrometheusSampleBuilder, PrometheusMetricsServlet). The PR description claims unit tests are written (>80% coverage), but no test files are present in the diff. This is a significant gap for a metrics-exporting feature where parsing bugs can silently produce wrong dashboards.

---

Automated review by github-manager-bot

[RockteMQ-AI]

The default case in fromSnapshotAndCount adds the median (quantile "0.5") sample twice — the first two entries in the Arrays.asList(...) are identical. This produces duplicate samples in every Prometheus scrape for any histogram type that falls through to default, which will confuse or error in Prometheus ingestion.

[RockteMQ-AI]

Stat.HistogramType.valueOf(metricName.getType()) will throw IllegalArgumentException if the metric's type string doesn't exactly match an enum constant. There is no try/catch, so a single unrecognized histogram metric name will abort the entire collect() call, breaking the whole /metrics endpoint. Wrap this in a try/catch and log+skip the offending metric.

[RockteMQ-AI]

sanitizeMetricName splits on ":" then "," and accesses index [1] without bounds checking. If any Dropwizard metric name doesn't follow the exact expected format (e.g. internal JVM metrics, or third-party metrics), this throws ArrayIndexOutOfBoundsException and crashes the scrape. Add defensive parsing with a fallback.

[RockteMQ-AI]

sanitizeLabelValues has the same unguarded split pattern and accesses split[0] through split[4]. Any metric name with fewer than 5 comma-separated segments will throw ArrayIndexOutOfBoundsException. Additionally, split[3].substring(var3.indexOf("=") + 1) will misbehave if var3 doesn't contain "=" (indexOf returns -1, yielding substring(0) — silently wrong rather than failing loudly).

[RockteMQ-AI]

The Prometheus Jetty Server is started in the constructor but the embeddedApp reference is local and never stored. This means: (1) there is no way to shut it down on application stop — the server and its thread pool will leak; (2) if exporterPort is already bound, the exception propagates uncaught from the constructor, preventing the entire Worker from starting. Store the reference and wire it into the shutdown lifecycle.

[RockteMQ-AI]

doPost delegates to doGet. Exposing metrics via POST is unconventional and unnecessary — Prometheus only scrapes via GET. Remove doPost to reduce the attack surface.

[RockteMQ-AI]

The exporterPort field (default 5557) is not populated from the properties map like other config fields (e.g., httpPort). There is no buildWorkerConfig() / init logic that reads it from a config key, so the setter is the only way to change it. Add it to the config loading path to be consistent with the rest of the class.

[RockteMQ-AI]

Missing newline at end of file. Minor, but some tools and checkers flag this.

[RockteMQ-AI]

new Long(counter.getCount()).doubleValue() uses the deprecated Long constructor. Replace with (double) counter.getCount() or Long.valueOf(...).doubleValue().

[RockteMQ-AI]

No tests are included for any of the new classes (DropwizardExports, PrometheusSampleBuilder, PrometheusMetricsServlet). The PR description claims unit tests are written (>80% coverage), but no test files are present in the diff. This is a significant gap for a metrics-exporting feature where parsing bugs can silently produce wrong dashboards.

[RockteMQ-AI]

LGTM. Trivial change, looks good.

---

Automated review by github-manager-bot