[
https://issues.apache.org/jira/browse/IGNITE-29229?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Egor Baranov updated IGNITE-29229:
----------------------------------
Description:
There is currently no way to detect AssertionError occurrences in the node JVM.
These errors are frequently swallowed by catch (Throwable) blocks (500+ in the
codebase), making them
invisible during test runs.
This change adds a LongAdderMetric named sys.AssertionErrorCount to
GridMetricManager. The metric is backed by a JFR jdk.JavaErrorThrow event
consumer (RecordingStream) that listens
for thrown java.lang.AssertionError events and increments the counter. The
stream is started/stopped with the manager lifecycle.
The metric is exposed through all standard export channels (SQL metrics
system view, JMX, REST) and requires no JVM agent.
was:
h2. Description
Currently, when AssertionError events occur in the Ignite node JVM (e.g.
due to internal invariant violations triggered by assert statements), there is
no built-in mechanism to
count or detect them. These errors can be silently swallowed by
catch(Throwable) blocks (500+ occurrences in the codebase), rethrown as wrapped
exceptions, or terminate threads,
making them difficult to track during test runs and production diagnostics.
h3. Solution
A new long-adder metric h1.AssertionErrorCount is added to the h1.sys
metrics registry (fully qualified name: sys.AssertionErrorCount). The metric is
backed by a JFR
h1.jdk.JavaErrorThrow event consumer via
h1.jdk.jfr.consumer.RecordingStream which is started and stopped as part of the
h1.GridMetricManager lifecycle.
The metric increments on every thrown h1.AssertionError regardless of
whether the error is caught, rethrown, wrapped, or uncaught. This provides 100%
detection coverage without
requiring bytecode instrumentation or JVM agents.
h3. Implementation details
* Metric registered in h1.GridMetricManager constructor under h1.sys
registry
* JFR h1.RecordingStream started in h1.start(), closed in h1.stop()
* Event listener filters on h1.thrownClass h1.equals
java.lang.AssertionError
* Metric exposed via all standard Ignite metric export channels (SQL
h1.metrics system view, JMX, REST)
h3. Files changed
*
modules/core/src/main/java/org/apache/ignite/internal/processors/metric/GridMetricManager.java
*
modules/core/src/test/java/org/apache/ignite/internal/processors/metric/GridMetricManagerAssertionMetricTest.java
*
modules/core/src/test/java/org/apache/ignite/testsuites/IgniteKernalSelfTestSuite.java
h3. Verification
Metric is visible via:
{code:sql}
SELECT * FROM metrics WHERE metric_name = 'sys.AssertionErrorCount';
{/code}
Unit test h1.GridMetricManagerAssertionMetricTest verifies both the metric
metadata (name, description) and the increment behavior upon a thrown
AssertionError. Test passes with and
without global JFR enabled.
> Add AssertionErrorCount monitoring metric to detect AssertionError
> occurrences in the JVM
> -----------------------------------------------------------------------------------------
>
> Key: IGNITE-29229
> URL: https://issues.apache.org/jira/browse/IGNITE-29229
> Project: Ignite
> Issue Type: New Feature
> Reporter: Egor Baranov
> Assignee: Egor Baranov
> Priority: Major
>
> There is currently no way to detect AssertionError occurrences in the node
> JVM. These errors are frequently swallowed by catch (Throwable) blocks (500+
> in the codebase), making them
> invisible during test runs.
> This change adds a LongAdderMetric named sys.AssertionErrorCount to
> GridMetricManager. The metric is backed by a JFR jdk.JavaErrorThrow event
> consumer (RecordingStream) that listens
> for thrown java.lang.AssertionError events and increments the counter.
> The stream is started/stopped with the manager lifecycle.
> The metric is exposed through all standard export channels (SQL metrics
> system view, JMX, REST) and requires no JVM agent.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
