|
| 1 | +/* |
| 2 | + * Copyright The OpenTelemetry Authors |
| 3 | + * SPDX-License-Identifier: Apache-2.0 |
| 4 | + */ |
| 5 | + |
| 6 | +package io.opentelemetry; |
| 7 | + |
| 8 | +import static org.assertj.core.api.Assertions.assertThat; |
| 9 | + |
| 10 | +import io.opentelemetry.api.trace.Span; |
| 11 | +import io.opentelemetry.api.trace.Tracer; |
| 12 | +import io.opentelemetry.sdk.OpenTelemetrySdk; |
| 13 | +import io.opentelemetry.sdk.common.CompletableResultCode; |
| 14 | +import io.opentelemetry.sdk.trace.SdkTracerProvider; |
| 15 | +import io.opentelemetry.sdk.trace.data.SpanData; |
| 16 | +import io.opentelemetry.sdk.trace.export.SimpleSpanProcessor; |
| 17 | +import io.opentelemetry.sdk.trace.export.SpanExporter; |
| 18 | +import java.util.ArrayList; |
| 19 | +import java.util.Collection; |
| 20 | +import java.util.List; |
| 21 | +import java.util.concurrent.TimeUnit; |
| 22 | +import org.crac.Context; |
| 23 | +import org.crac.Resource; |
| 24 | +import org.junit.jupiter.api.Disabled; |
| 25 | +import org.junit.jupiter.api.Test; |
| 26 | + |
| 27 | +/** |
| 28 | + * Integration-style lifecycle tests for CRaC (Coordinated Restore at Checkpoint) support. |
| 29 | + * |
| 30 | + * <p>These tests use {@link MockCracContext} to simulate the CRaC checkpoint/restore lifecycle |
| 31 | + * without a CRaC-enabled JDK. Resources register with the mock context; the test then drives {@code |
| 32 | + * beforeCheckpoint} and {@code afterRestore} callbacks directly. |
| 33 | + * |
| 34 | + * <p>See: <a href="https://github.com/open-telemetry/opentelemetry-java/issues/6756">#6756</a> |
| 35 | + */ |
| 36 | +class CracLifecycleIntegrationTest { |
| 37 | + |
| 38 | + /** |
| 39 | + * Demonstrates the failure mode when the SDK is naively shut down at checkpoint with no |
| 40 | + * corresponding restore logic. This is what happens today without proper CRaC support: the SDK is |
| 41 | + * a one-shot object, so spans emitted after a restore are silently dropped. |
| 42 | + */ |
| 43 | + @Test |
| 44 | + void spansDroppedAfterRestore_naiveCracIntegration() throws Exception { |
| 45 | + MockCracContext cracContext = new MockCracContext(); |
| 46 | + InMemorySpanExporter exporter = new InMemorySpanExporter(); |
| 47 | + OpenTelemetrySdk sdk = buildSdk(exporter); |
| 48 | + Tracer tracer = sdk.getTracer("crac-lifecycle-test"); |
| 49 | + |
| 50 | + // Naive CRaC resource: shuts the SDK down at checkpoint, does nothing on restore. |
| 51 | + cracContext.register( |
| 52 | + new Resource() { |
| 53 | + @Override |
| 54 | + public void beforeCheckpoint(Context<? extends Resource> context) { |
| 55 | + sdk.getSdkTracerProvider().shutdown().join(10, TimeUnit.SECONDS); |
| 56 | + } |
| 57 | + |
| 58 | + @Override |
| 59 | + public void afterRestore(Context<? extends Resource> context) { |
| 60 | + // No restore logic — this is the gap that #6756 addresses. |
| 61 | + } |
| 62 | + }); |
| 63 | + |
| 64 | + emitSpan(tracer, "before-checkpoint"); |
| 65 | + sdk.getSdkTracerProvider().forceFlush().join(10, TimeUnit.SECONDS); |
| 66 | + assertThat(exporter.exportedCount()).isEqualTo(1); |
| 67 | + |
| 68 | + cracContext.simulateCheckpoint(); |
| 69 | + cracContext.simulateRestore(); |
| 70 | + |
| 71 | + // Post-restore span is silently dropped: the SDK is shut down and has no way to reinitialize. |
| 72 | + emitSpan(tracer, "after-restore"); |
| 73 | + sdk.getSdkTracerProvider().forceFlush().join(10, TimeUnit.SECONDS); |
| 74 | + assertThat(exporter.exportedCount()).isEqualTo(1); |
| 75 | + } |
| 76 | + |
| 77 | + /** |
| 78 | + * Describes the desired behavior once the SDK properly implements {@link Resource}: spans emitted |
| 79 | + * after a CRaC restore should be exported normally. |
| 80 | + * |
| 81 | + * <p>This test is disabled until <a |
| 82 | + * href="https://github.com/open-telemetry/opentelemetry-java/issues/6756">#6756</a> is addressed. |
| 83 | + * When that work lands, the SDK (or an adapter it exposes) should register with the CRaC context |
| 84 | + * so that {@code beforeCheckpoint} flushes and quiesces, and {@code afterRestore} reinitializes |
| 85 | + * exporters and processors. Replace the TODO below with the real SDK API. |
| 86 | + */ |
| 87 | + @Test |
| 88 | + @Disabled("Expected to fail until #6756 adds checkpoint/restore-safe SDK lifecycle") |
| 89 | + void spansExportedAfterRestore_properCracIntegration() throws Exception { |
| 90 | + MockCracContext cracContext = new MockCracContext(); |
| 91 | + InMemorySpanExporter exporter = new InMemorySpanExporter(); |
| 92 | + OpenTelemetrySdk sdk = buildSdk(exporter); |
| 93 | + Tracer tracer = sdk.getTracer("crac-lifecycle-test"); |
| 94 | + |
| 95 | + // TODO(#6756): replace this placeholder with the real SDK CRaC API, e.g.: |
| 96 | + // cracContext.register(sdk.asCracResource()); |
| 97 | + cracContext.register( |
| 98 | + new Resource() { |
| 99 | + @Override |
| 100 | + public void beforeCheckpoint(Context<? extends Resource> context) throws Exception { |
| 101 | + sdk.getSdkTracerProvider().shutdown().join(10, TimeUnit.SECONDS); |
| 102 | + } |
| 103 | + |
| 104 | + @Override |
| 105 | + public void afterRestore(Context<? extends Resource> context) throws Exception { |
| 106 | + // Reinitialize: reopen connections, restart background threads. |
| 107 | + // No SDK API exists for this yet — this is the body of #6756. |
| 108 | + } |
| 109 | + }); |
| 110 | + |
| 111 | + emitSpan(tracer, "before-checkpoint"); |
| 112 | + sdk.getSdkTracerProvider().forceFlush().join(10, TimeUnit.SECONDS); |
| 113 | + assertThat(exporter.exportedCount()).isEqualTo(1); |
| 114 | + |
| 115 | + cracContext.simulateCheckpoint(); |
| 116 | + cracContext.simulateRestore(); |
| 117 | + |
| 118 | + emitSpan(tracer, "after-restore"); |
| 119 | + sdk.getSdkTracerProvider().forceFlush().join(10, TimeUnit.SECONDS); |
| 120 | + assertThat(exporter.exportedCount()).isEqualTo(2); |
| 121 | + } |
| 122 | + |
| 123 | + private static OpenTelemetrySdk buildSdk(SpanExporter exporter) { |
| 124 | + return OpenTelemetrySdk.builder() |
| 125 | + .setTracerProvider( |
| 126 | + SdkTracerProvider.builder() |
| 127 | + .addSpanProcessor(SimpleSpanProcessor.create(exporter)) |
| 128 | + .build()) |
| 129 | + .build(); |
| 130 | + } |
| 131 | + |
| 132 | + private static void emitSpan(Tracer tracer, String name) { |
| 133 | + Span span = tracer.spanBuilder(name).startSpan(); |
| 134 | + span.end(); |
| 135 | + } |
| 136 | + |
| 137 | + private static final class InMemorySpanExporter implements SpanExporter { |
| 138 | + private final List<SpanData> spans = new ArrayList<>(); |
| 139 | + private boolean shutdown; |
| 140 | + |
| 141 | + @Override |
| 142 | + public CompletableResultCode export(Collection<SpanData> spans) { |
| 143 | + if (shutdown) { |
| 144 | + return CompletableResultCode.ofFailure(); |
| 145 | + } |
| 146 | + this.spans.addAll(spans); |
| 147 | + return CompletableResultCode.ofSuccess(); |
| 148 | + } |
| 149 | + |
| 150 | + @Override |
| 151 | + public CompletableResultCode flush() { |
| 152 | + return CompletableResultCode.ofSuccess(); |
| 153 | + } |
| 154 | + |
| 155 | + @Override |
| 156 | + public CompletableResultCode shutdown() { |
| 157 | + shutdown = true; |
| 158 | + return CompletableResultCode.ofSuccess(); |
| 159 | + } |
| 160 | + |
| 161 | + int exportedCount() { |
| 162 | + return spans.size(); |
| 163 | + } |
| 164 | + } |
| 165 | +} |
0 commit comments