2024-05-29 Spring IO - Observability Workshop

Transcript

1. Spring I/O 2024-05-29 Jonatan Ivanov & Tommy Ludwig Observability Workshop
2. Hello From Us 👋
3. Spring Team/Micrometer Jonatan Ivanov Tommy Ludwig
4. Cover w/ Image Today's Schedule • Morning Break (20 min) ◦ 10:45 - 11:05 • Lunch Break (60 min) ◦ 13:00 - 14:00 • Afternoon Break (20 min) ◦ 15:30 - 15:50 • Ends 5pm (ish)
5. Cover w/ Image Today's Schedule • Machine Setup (Java + Docker) • Tour of the Spring Boot Applications • HTTP Interface Clients
6. Cover w/ Image Today's Schedule • Observability ◦ Spring and Observability ◦ JDBC observability ◦ Grafana + Prometheus • Manual Instrumentation • Questions and Answers
7. 1. The fundamentals of Application Observability, why do we need it 2. How to apply these fundamentals to realistic scenarios in sample applications where having observability is crucial 3. How Micrometer provides a unified API to instrument your code for various signals and backends 4. What is Micrometer’s new Observation API and how to use it 5. What signals to watch in your own application 6. Spring Boot’s built-in observability features and how to customize them 7. How to avoid common issues 8. How to integrate metrics with distributed tracing and logs 9. How to visualize and analyze observability data to identify issues and optimize performance 10. How to troubleshoot issues faster and more effectively 11. The latest developments around Observability
8. Hello From You 👋
9. Hands Up 󰢨 • You're a Java developer? • You've used Docker? • You've used Spring Boot? • You've used Observability? • You've used OpenTelemetry?
10. WiFi Spring I/O Workshops bootifulBCN24
11. Machine Setup
12. Minimum System Requirements • Git • Docker • Java 17 (or higher) • Java IDE 󰢨 Problems? Please let us know!
13. How to get help? • README.md • HELP.md (copy-paste grand mastery) • A “secret” final branch 😈 [...] ab4ab30 Add org property 2456f04 Initial commit • slack.micrometer.io #springio2024 • Let us know! 󰢨 󰢨 Problems? Please let us know!
14. Machine Setup $ git clone https://github.com/jonatan-ivanov/springio24-observability-workshop $ cd springio24-observability-workshop $ java --version $ ./mvnw --version $ ./mvnw package $ docker compose up $ docker compose ps $ docker compose down #--volumes 󰢨 Problems? Please let us know!
15. Tour of a Spring Boot Application
16. About the Dog Service Sample • It's a very silly application 🤪 • Just enough code to demo what we want! • Spring Boot 3.3 • Java 17 • JPA (Hibernate) • Spring MVC • Spring Security
17. Checkout the Code $ git clone https://github.com/jonatan-ivanov/springio24-observability-workshop $ cd springio24-observability-workshop $ git checkout main $ docker compose up -d $ ./mvnw package Import into your favorite IDE 󰢨 Problems? Please let us know!
18. Check Out the Code - pom.xml • Open pom.xml • We're using spring-boot-starter-parent 3.3 • The java.version property is 17 • We're using starters for: ◦ actuator, web, data-jpa, and security • We're using PostgreSQL for the datastore
19. Check Out the Code - src/main/resources Open src/main/resources Look at schema.sql and data.sql files DOG OWNER • id (pk) • name • owner_id (fk) • id (pk) • name 1 *
20. Check Out the Code - com.example.dogservice.domain • Open com.example.dogservice.domain package • Dog and Owner JPA classes map to the schema • DogRepository and OwnerRepository are Spring Data repositories ◦ findByNameIgnoringCase is converted to JQL automatically • InfoLogger is an ApplicationRunner to log info at startup
21. Check Out the Code - com.example.dogservice.service • Open com.example.dogservice.service package • OwnerService uses constructor injection • Simple facade over repositories • Throws custom NoSuchDogOwnerException
22. Check Out the Code - com.example.dogservice.web • Open com.example.dogservice.web package • DogsController ◦ Simple controller used for testing • OwnerController ◦ Delegates to the OwnerService ◦ Deals with NoSuchDogOwnerException ◦ Note: Meta-annotated @RestController and @GetMapping
23. Check Out the Code - com.example.dogservice.security • Open com.example.dogservice.security package • SecurityConfiguration ◦ Defines our web security • SecurityProperties and UserProperties ◦ @ConfigurationProperties maps from values in src/main/resources/application.yml
24. Check Out the Code - src/main/resources/ • Open src/main/resources/ • Inspect application.yml ◦ Defines the database connection ◦ Configures JPA ◦ Enables JMX ◦ Configures server errors ◦ Exposes all actuator endpoints ◦ Enables actuators over HTTP ◦ Customizes a metric name ◦ Defines the in-memory user details
25. Run the code $ ./mvnw -pl dog-service clean spring-boot:run __ , ," e`--o (( ( | __,' \\~----------------' \_;/ ( / /) ._______________. ) (( ( (( ( ``-' ``-' … 2024-04-09T10:50:22.372-05:00 INFO 151018 --- [dog-service] [ main] [ ] c.example.dogservice.domain.InfoLogger : Found owners [Tommy, Jonatan] 2024-04-09T10:50:22.386-05:00 INFO 151018 --- [dog-service] [ main] [] c.example.dogservice.domain.InfoLogger : Found dogs [Snoopy owned by Tommy, Goofy owned by Tommy, Clifford owned by Jonatan] … 2024-04-09T10:50:28.260-05:00 INFO 151018 --- [dog-service] [nio-8080-exec-1] [ ] o.s.web.servlet.DispatcherServlet : Completed initialization in 2 ms …
26. Checkpoint Everyone has a running application
27. Checkpoint - DogsController Works $ http "http://localhost:8080/dogs" Generates error JSON $ http "http://localhost:8080/dogs?aregood=true" $ http "http://localhost:8080/dogs?aregood=false" $ http "http://localhost:8080/dogs/?aregood=false" Note the trailing slash
28. Checkpoint - OwnerController Works $ http "http://localhost:8080/owner/tommy/dogs" Needs login $ http -a user:password "http://localhost:8080/owner/tommy/dogs" $ http -a user:password "http://localhost:8080/owner/jonatan/dogs" $ http -a user:password "http://localhost:8080/owner/dave/dogs" NoSuchOwnerException mapped to HTTP 404
29. Checkpoint - Actuator Works $ http -a admin:secret "http://localhost:8080/actuator" Open the following in a web browser: http://localhost:8080/actuator http://localhost:8080/actuator/metrics http://localhost:8080/actuator/metrics/http.server.requests
30. Checkpoint - Actuator Over JMX Works $ jconsole Select com.example.dogservice.DogServiceApplication Click Connect Select MBeans on the menu Expand org.springframework.boot, Endpoint
31. Checkpoint Everyone has a working application
32. HTTP Client Interfaces
33. About the Dog Client Sample • It's another very silly application 🤪 • Just enough code to demo what we want! • API using HttpServiceProxy • Another MVC controller • Some configuration
34. HTTP Client Interfaces: interfaces + metadata • Open com.example.dogclient.api.Api @GetExchange("/dogs") DogsResponse dogs(@RequestParam(name = "aregood") boolean areGood); @GetExchange("/owner/{name}/dogs") List<String> ownedDogs(@PathVariable String name);
35. HTTP Client Interfaces + Records • Open com.example.dogclient.api.DogsResponse public record DogsResponse(String message) { }
36. HTTP Client Interfaces - Building the client • Open com.example.dogclient.DogClientApplication WebClient webClient = webClientBuilder.baseUrl("…").build(); WebClientAdapter adapter = WebClientAdapter.create(webClient); HttpServiceProxyFactory factory = HttpServiceProxyFactory.builderFor(adapter).build(); return factory.createClient(Api.class);
37. HTTP Client Interfaces - Using the client // Signature DogsResponse dogs(boolean areGood); List<String> ownedDogs(String name); // Usage api.dogs(true); api.ownedDogs("Tommy");
38. ApplicationRunner • Open com.example.dogclient.DogClientApplication • Prints information when the application starts • Functional interface called when the app starts: void run(ApplicationArguments args) throws Exception
39. Run the application - Check the startup output $ ./mvnw -pl dog-service clean spring-boot:run $ ./mvnw -pl dog-client clean spring-boot:run 2024-04-09T13:26:07.471-05:00 INFO 1618241 --- [dog-client] [ main] [ ] c.e.dogclient.DogClientApplication : Started DogClientApplication in 1.505 seconds (process running for 1.643) DogsResponse[message=We <3 dogs!!!] [Snoopy, Goofy] $ http "http://localhost:8081/owner/tommy/dogs"
40. Checkpoint Client application works
41. Observability
42. What is Observability?
43. What is Observability? How well we can understand the internals of a system based on its outputs (Providing meaningful information about what happens inside) (Data about your app)
44. Why do we need Observability?
45. Why do we need Observability? Today's systems are increasingly complex (cloud) (Death Star Architecture, Big Ball of Mud)
46. Environments can be chaotic You turn a knob here a little and apps are going down there We need to deal with unknown unknowns We can’t know everything Things can be perceived differently by observers Everything is broken for the users but seems ok to you Why do we need Observability?
47. Why do we need Observability? (business perspective) Reduce lost revenue from production incidents Lower mean time to recovery (MTTR) Require less specialized knowledge Shared method of investigating across system Quantify user experience Don't guess, measure!
48. Logging Metrics Distributed Tracing
49. Logging What happened (why)? Emitting events Metrics What is the context? Aggregating data Distributed Tracing Why happened? Recording causal ordering of events Logging - Metrics - Distributed Tracing
50. Examples Latency Logging HTTP request took 140ms Metrics P99.999: 140ms Max: 150 ms Distributed Tracing DB was slow (lot of data was requested) Error Logging Request failed (stacktrace?) Metrics The error rate is 0.001/sec 2 errors in the last 30 minutes Distributed Tracing DB call failed (invalid input)
51. Checkpoint Everyone knows what Observability is
52. 1. The fundamentals of Application Observability, why do we need it 2. How to apply these fundamentals to realistic scenarios in sample applications where having observability is crucial 3. How Micrometer provides a unified API to instrument your code for various signals and backends 4. What is Micrometer’s new Observation API and how to use it 5. What signals to watch in your own application 6. Spring Boot’s built-in observability features and how to customize them 7. How to avoid common issues 8. How to integrate metrics with distributed tracing and logs 9. How to visualize and analyze observability data to identify issues and optimize performance 10. How to troubleshoot issues faster and more effectively 11. The latest developments around Observability
53. Logging with JVM/Spring
54. SLF4J with Logback comes pre-configured SLF4J (Simple Logging Façade for Java) Simple API for logging libraries Logback Natively implements the SLF4J API If you want Log4j2 instead of Logback: - spring-boot-starter-logging + spring-boot-starter-log4j2 Logging with JVM/Spring: SLF4J + Logback
55. Setup Logging - Add org property • We will need something that we can use to query: ◦ All of our apps (spring.application.org) ◦ Only one app (spring.application.name) ◦ Only one instance (we only have one instance/app) spring: application: name: dog-service org: petclinic
56. Setup Centralized Logging - Add Loki4J • Copy From: dog-client/src/main/resources/logback-spring.xml To: dog-service/src/main/resources/logback-spring.xml • Add dependency to pom.xml <dependency> <groupId>com.github.loki4j</groupId> <artifactId>loki-logback-appender</artifactId> <version>1.5.1</version> </dependency>
57. Setup Logging - Do we have logs? • Go to Grafana: http://localhost:3000 • Choose Explore, then Loki from the drop down • Search for application = dog-service • Search for org = petclinic • We will get back to our logs later
58. Checkpoint Everyone has logs in Loki for both services
59. Coffee break (20 minutes) 10:45 - 11:05
60. Metrics with JVM/Spring
61. Metrics with JVM/Spring: Micrometer Dimensional Metrics library on the JVM Like SLF4J, but for metrics API is independent of the configured metrics backend Supports many backends Comes with spring-boot-actuator Spring projects are instrumented using Micrometer Many third-party libraries use Micrometer
62. Supported metrics backends/formats/protocols Ganglia Graphite Humio InfluxDB JMX KairosDB New Relic (/actuator/metrics) OpenTSDB OTLP Prometheus SignalFx Stackdriver (GCP) StatsD Wavefront (VMware) AppOptics Atlas Azure Monitor CloudWatch (AWS) Datadog Dynatrace Elastic
63. Setup Metrics
64. Setup Metrics - Add the org to Observations and Metrics application.yml management: observations: key-values: org: ${spring.application.org} metrics: tags: application: ${spring.application.name} org: ${spring.application.org}
65. Setup Metrics - Let’s check Metrics 🧐 • http://localhost:8080/actuator/prometheus • 401 🧐 • Prometheus? http://localhost:9090/targets • Spring Security! 👀 • Let’s disable it, what could go wrong!? 😈 • Everyone, please don’t do this in prod! • Unless you want everyone to know about it. 😈
66. Setup Metrics - Disable auth for certain endpoints SecurityConfiguration.java requests .requestMatchers("/dogs", "/actuator/**").permitAll();
67. Setup Metrics - Add histogram support for http metrics • We want to see the latency distributions on our dashboards • We want to calculate percentiles (tp99?) management: metrics: distribution: percentiles-histogram: # all: true http.server.requests: true
68. Setup Metrics - Let’s check the HTTP and JVM metrics • Let’s check /actuator/metrics /actuator/metrics/{metricName} /actuator/metrics/{metricName}?tag=key:value • Let’s write a Prometheus query (HELP.md) sum by (application) (rate(http_server_requests_seconds_count[5m])) • Let’s check the dashboards: go to Grafana, then Browse ◦ Spring Boot Statistics ◦ Dogs
69. Checkpoint Everyone has metrics on the dashboards
70. Distributed Tracing with JVM/Spring
71. Distributed Tracing with JVM/Spring 🚀 Boot 2.x: Spring Cloud Sleuth Boot 3.x: Micrometer Tracing (Sleuth w/o Spring dependencies) Provide an abstraction layer on top of tracing libraries - Brave (OpenZipkin), default - OpenTelemetry (CNCF), experimental Instrumentation for Spring Projects, 3rd party libraries, your app Support for various backends
72. Setup Distributed Tracing - Add Micrometer Tracing <dependency> <groupId>io.micrometer</groupId> <artifactId>micrometer-tracing-bridge-brave</artifactId> </dependency> <dependency> <groupId>io.zipkin.reporter2</groupId> <artifactId>zipkin-reporter-brave</artifactId> </dependency>
73. Setup Distributed Tracing - Set sampling probability 🧐 management: tracing: sampling: probability: 1.0
74. Setup Distributed Tracing - Setup log correlation 🚀 • If you are on Spring Boot 3.1 or above, this is not needed • If you are on 3.1 or lower, you need to set logging.pattern.level • We are on 3.3! logging: level: org.springframework.web.servlet.DispatcherServlet: DEBUG
75. Setup Distributed Tracing - Let’s look at correlated logs 2024-03-22T20:13:21.588Z DEBUG 2167 --- [dog-service] [http-nio-8090-exec-5] [65fde66134624d949e80e0d3241ed138-9e80e0d3241ed138] o.s.web.servlet.DispatcherServlet: Completed 200 OK
76. Setup Distributed Tracing - Let’s look at some traces • Go to Grafana, then Explore and choose Tempo • Terminology ◦ Span ◦ Trace ◦ Tags ◦ Annotations
77. Checkpoint Everyone has log correlation and traces in Tempo
78. 1. The fundamentals of Application Observability, why do we need it 2. How to apply these fundamentals to realistic scenarios in sample applications where having observability is crucial 3. How Micrometer provides a unified API to instrument your code for various signals and backends 4. What is Micrometer’s new Observation API and how to use it 5. What signals to watch in your own application 6. Spring Boot’s built-in observability features and how to customize them (to be continued…) 7. How to avoid common issues 8. How to integrate metrics with distributed tracing and logs 9. How to visualize and analyze observability data to identify issues and optimize performance 10. How to troubleshoot issues faster and more effectively 11. The latest developments around Observability (to be continued…)
79. Observation API
80. • Add logs (application logs) • Add metrics ◦ Increment Counters ◦ Start/Stop Timers • Add Distributed Tracing ◦ Start/Stop Spans ◦ Log Correlation ◦ Context Propagation You want to instrument your application…
81. Instrumentation with the Observation API 🚀 Observation observation = Observation.start("talk",registry); try { // TODO: scope doSomething(); } catch (Exception exception) { observation.error(exception); throw exception; } finally { // TODO: attach tags (key-value) observation.stop(); }
82. Configuration with the Observation API 🚀 ObservationRegistry registry = ObservationRegistry.create(); registry.observationConfig() .observationHandler(new MeterHandler(...)) .observationHandler(new TracingHandler(...)) .observationHandler(new LoggingHandler(...)) .observationHandler(new AuditEventHandler(...));
83. Observation API 🚀 Observation.createNotStarted("talk",registry) .lowCardinalityKeyValue("event", "SIO") .highCardinalityKeyValue("uid", userId) .observe(this::talk); @Observed
84. 1. The fundamentals of Application Observability, why do we need it 2. How to apply these fundamentals to realistic scenarios in sample applications where having observability is crucial 3. How Micrometer provides a unified API to instrument your code for various signals and backends (to be continued…) 4. What is Micrometer’s new Observation API and how to use it 5. What signals to watch in your own application 6. Spring Boot’s built-in observability features and how to customize them (to be continued…) 7. How to avoid common issues 8. How to integrate metrics with distributed tracing and logs 9. How to visualize and analyze observability data to identify issues and optimize performance 10. How to troubleshoot issues faster and more effectively 11. The latest developments around Observability (to be continued…)
85. Interoperability
86. Setup Observations - Disable Spring Security Observations SecurityConfiguration.java @Bean ObservationPredicate noSpringSecurityObservations() { return (name,ctx)-> !name.startsWith("spring.security."); }
87. Setup Observations - Disable Actuator Observations ActuatorConfiguration.java if (name.equals("http.server.requests") && ctx instanceof ServerRequestObservationContext sc) { return !sc.getCarrier() .getRequestURI() .startsWith("/actuator"); } else return true;
88. Setup Observations - Enable JDBC Observations 🚀 Tadaya Tsuyukubo 😎 net.ttddyy.observation:datasource-micrometer-spring-boot (1.0.3) jdbc: datasource-proxy: include-parameter-values: true query: enable-logging: true log-level: INFO
89. Setup Observations - Add custom Observation OwnerService.java Observation.createNotStarted("getDogs", registry) .contextualName("gettingOwnedDogs") .highCardinalityKeyValue("owner", owner) .observe(() -> { //… });
90. through traces TraceID ❮ Exemplars Tags ❯ metrics logs traces Interoperability 🚀
91. Interoperability - How to check Exemplars • Exemplars are only available if you request the OpenMetrics format • Your browser does not do this http :8081/actuator/prometheus / 'Accept: application/openmetrics-text;version=1.0.0' | grep trace_id
92. Checkpoint Logs <=> Metrics <=> Traces
93. Setup Observations - Log error and signal it OwnerController.java ProblemDetail onNoSuchDogOwner( HttpServletRequest request, NoSuchDogOwnerException ex) { logger.error("Ooops!", ex); ServerHttpObservationFilter .findObservationContext(request) .ifPresent(context -> context.setError(ex));
94. ObservationFilter tempoErrorFilter() { return context -> { if (context.getError() != null) { context.addHighCardinalityKeyValue( KeyValue.of("error", "true") ); context.addHighCardinalityKeyValue( KeyValue.of( "errorMsg", context.getError().getMessage()) ); } return context; }; } Setup Observations - Hack error reporting for Tempo
95. Setup Observations - Hack DB tags for ServiceGraph if(ctx instanceof DataSourceBaseContext dsCtx){ ctx.addHighCardinalityKeyValue( KeyValue.of( "db.name", dsCtx.getRemoteServiceName() ) ); }
96. Actuator - Add Java, OS, and Process InfoContributors management: info: java: enabled: true os: enabled: true process: enabled: true
97. Checkpoint The applications are observable! 😎
98. 1. The fundamentals of Application Observability, why do we need it 2. How to apply these fundamentals to realistic scenarios in sample applications where having observability is crucial 3. How Micrometer provides a unified API to instrument your code for various signals and backends (to be continued…) 4. What is Micrometer’s new Observation API and how to use it 5. What signals to watch in your own application (to be continued…) 6. Spring Boot’s built-in observability features and how to customize them 7. How to avoid common issues 8. How to integrate metrics with distributed tracing and logs 9. How to visualize and analyze observability data to identify issues and optimize performance 10. How to troubleshoot issues faster and more effectively 11. The latest developments around Observability (to be continued…)
99. Lunch break (1 hour) 13:00 - 14:00
100. Manual Instrumentation
101. Micrometer: MeterRegistry • Meter: interface to collect measurements • MeterRegistry: abstract class to create/store Meters • Backends have implementations of MeterRegistry • SimpleMeterRegistry (debugging, testing, actuator) • SimpleMeterRegistry#getMetersAsString • CompositeMeterRegistry
102. Dimensionality • Dimensional vs. Hierarchical • Dimensional: metrics enriched with key/value pairs • Hierarchical: key/value pairs flattened, added to the name
103. Dimensionality - Dimensional Example (Prometheus) http_server_requests_count{ application="tea-service", exception="None", method="GET", outcome="SUCCESS", profiles="local", status="200", uri="/tea/{name}"} 2.0
104. Dimensionality - Hierarchical Example "http-server-requests-count.tea-service.None .GET.SUCCESS.local.200./tea/{name}": 2.0
105. Cumulative vs. Delta (temporality) Cumulative: the reported value is the total value since the beginning of the measurements Delta: the reported value is the difference in the measurements since the last time it was reported
106. Cumulative vs. Delta (temporality) - Example • We count certain events and report these every minute • The event happened 3 times in the first minute, 2 times in the second, and once in the third • Cumulative says: 3, 5, 6 (running total) • Delta says: 3, 2, 1 (difference)
107. Push vs. Poll Poll: the backend polls the apps for metrics at their leisure (e.g.: Prometheus) Push: the apps send metrics to the backend on a regular interval (e.g.: InfluxDB, ElasticSearch, etc.)
108. Creating a MeterRegistry PrometheusMeterRegistry registry = new PrometheusMeterRegistry(PrometheusConfig.DEFAULT); // [...] System.out.println(registry.scrape());
109. Micrometer basic Meter types - example use case • Counter - cache hits • Gauge - CPU usage % • Timer - HTTP server request timing • DistributionSummary - HTTP request size • LongTaskTimer - timing batch job processing
110. Micrometer: Counter • Records a single metric: a count • Monotonic: only increment(), no decrement() • Example: number of cache hits Counter counter = registry.counter("test"); counter.increment();
111. Micrometer: Counter + Tags registry.counter( "test.counter", "application", "test" ).increment(); test_counter_total{application="test"} 1.0
112. Micrometer: Counter + Builder Counter.builder("test.counter") .description("Test counter") .baseUnit("events") .tag("application", "test") .register(registry) // create or get .increment();
113. High Cardinality 🧐 for (int i = 0; i < 100000; i++) { Counter.builder("test.counter") .tag("userId", String.valueOf(i)) .register(registry) .increment(); }
114. • userId (lots of users) • email (lots of users) • any resourceId (lots of resources) • requestId/txId/traceId/spanId/etc. • Request URL • Any (unsanitized) user input • Please always sanitize/normalize any user input! • Otherwise: DoS 😞 High Cardinality 🧐
115. High Cardinality 🧐 High cardinality should be avoided whenever possible. In cases where it isn’t possible to avoid it, see: • MeterFilter.maximumAllowableMetrics(...); • MeterFilter.maximumAllowableTags(...); • MeterFilter.ignoreTags(...); • HighCardinalityTagsDetector
116. Micrometer: Gauge • A handle to get the current value • Non-monotonic: can increase and decrease • “Asynchronous” • “Heisen-Gauge” • “State” should be mutable and “referenced” • Examples: queue size, number of threads, CPU temperature Never gauge something you can count with a Counter!
117. Micrometer: Gauge private final AtomicLong value = new AtomicLong(); // mutable + referenced // elsewhere “register” registry.gauge("test", value); value.set(2); // elsewhere update the value
118. Micrometer: Gauge 🧐 // immutable :( private Double value = 1024.0; registry.gauge("test", value); value = 1.0; System.out.println(registry.scrape()); // ???
119. Micrometer: Gauge 🧐 // not referenced either :( private Double value = 1024.0; registry.gauge("test", value); value = 1.0; System.gc(); // well… System.out.println(registry.scrape());
120. Micrometer: Gauge private final AtomicLong value = registry.gauge("test", new AtomicLong()); value.set(4); // elsewhere update the value
121. Micrometer: Gauge private final List<String> list = new ArrayList<>(); registry.gauge("test", list, List::size); list.add("test");
122. Micrometer: Gauge private final List<String> list = registry.gauge( "test", Tags.empty(), new ArrayList<>(), // “state object” List::size // “value function” );
123. Micrometer: Gauge private final List<String> list = registry.gaugeCollectionSize( "test", Tags.empty(), new ArrayList<>() );
124. Micrometer: Gauge private final Map<String, Integer> map = registry.gaugeMapSize( "test", Tags.empty(), new HashMap<>() );
125. Micrometer: Gauge private final TemperatureSensor sensor = new TemperatureSensor(); Gauge.builder( "test", () -> sensor.getTemperature() - 273.15 ).register(registry);
126. Micrometer: Gauge private final TemperatureSensor sensor = new TemperatureSensor(); Gauge.builder( "test", sensor::getTemperature ).register(registry);
127. Micrometer: DistributionSummary • Tracks the distribution of recorded values • It has one method: record(amount) • Always reports count, sum, max • Can report: Histograms, SLOs, and Percentiles • Example: payload sizes of requests and responses
128. Micrometer: DistributionSummary DistributionSummary ds = DistributionSummary .builder("response.size") .baseUnit("bytes") .register(registry); ds.record(10); ds.record(20);
129. Micrometer: Timer • Tracks the latency of events • Like DistributionSummary but the unit is time • Multiple ways to record latency • Always reports count, sum, max • Can report: Histograms, SLOs, and Percentiles • Example: processing time of incoming requests Never count something that you can time with a Timer or summarize with a DistributionSummary!
130. count, sum, max count: same as having a Counter (rate) sum: sum of the recorded values (sum/count?) max: max of the recorded values (time-windowed) See the note in this section of the docs.
131. Micrometer: Timer Timer timer = Timer.builder("requests") .register(registry); Sample sample = Timer.start(); doSomething(); sample.stop(timer);
132. Micrometer: Timer timer.record(() -> doSomething()); timer.recordCallable(() -> getSomething()); Runnable r = timer.wrap(() -> doSomething()); Callable c = timer.wrap(() -> getSomething());
133. Micrometer: Timer 🧐 // Don’t do this! If you do, use nanoTime() 🙏 long start = System.nanoTime(); doSomething(); long end = System.nanoTime(); timer.record(end - start, NANOSECONDS);
134. Micrometer: Clock 🧐 • There is a Clock abstraction in Micrometer • wallTime [ms]: for the current time, not for elapsed time • monotonicTime [ns]: for measuring elapsed time • Testing: MockClock, you can set the time with it 😈 no Thread.sleep(...)
135. Coffee break (20 minutes) 15:30 - 15:50
136. Micrometer: LongTaskTimer • “ActiveTaskTimer” 🧐 • Tracks the elapsed time of active events • Timer records latency after the events finished • LongTaskTimer records latency of running events • Timer: past, LongTaskTimer: present • Always reports count, sum, max • Can report: Histograms, SLOs, and Percentiles • Example: processing time of in-progress requests
137. Micrometer: LongTaskTimer LongTaskTimer ltt = LongTaskTimer .builder("test") .register(registry); Sample sample = ltt.start(); doSomething(); sample.stop();
138. Micrometer: LongTaskTimer ltt.record(() -> doSomething()); ltt.recordCallable(() -> getSomething());
139. Micrometer: (Client-Side) Percentiles 🧐 • Timer, DistributionSummary, LongTaskTimer • Approximated on the client side • Not aggregatable and only percentiles configured up-front are available • Use Histogram instead if you can Timer.builder("requests") .publishPercentiles(0.99, 0.999) .register(registry);
140. Micrometer: (Percentile) Histogram • Timer, DistributionSummary, LongTaskTimer • Show the “frequency” of values in a certain range • Arbitrary percentiles are approximated on the backend • Aggregatable! Timer.builder("requests") .publishPercentileHistogram() .register(registry);
141. Micrometer: SLOs • Timer, DistributionSummary, LongTaskTimer • Additional histogram “buckets” • Specific thresholds so you can count recordings above/below the threshold Timer.builder("requests") .serviceLevelObjectives(Duration.ofMillis(10)) .register(registry);
142. Micrometer: MeterProvider 🚀 MeterProvider<Counter> provider = Counter.builder("test") .tag("static", "42") .withRegistry(registry); provider.withTags("dynamic",value).increment();
143. avg, tp95/99, max 🧐 • avg(7,9,12,8,10,11,8,5,168,182) = 42 • (1 - 0.9542) * 100% = 88.40% • avg(tp95) = 😬 • tp95(tp95) = 😬 • max = 😉
144. Latency Expectations vs. Reality 🧐 Expectation • Normal distribution Reality • Long tail distribution • Multi-modal distribution https://commons.wikimedia.org/wiki/File:Joggers.png
145. Alerting • Don’t stare at dashboards; use alerts • Base alerts on metrics that represent business impact • Avoid duplicate alerts • Don’t alert on average latency! • See previous slide about SLOs
146. Micrometer: MeterFilter • Deny or Accept Meters • Transform Meter IDs (name, tags, description, unit) • Configure Distribution Statistics • Separates instrumentation from configuration
147. Micrometer: MeterFilter registry.config() .meterFilter(MeterFilter.commonTags(...)) .meterFilter(MeterFilter.ignoreTags(...)) .meterFilter(MeterFilter.renameTag(...)) .meterFilter(MeterFilter.replaceTagValues(...)) .meterFilter(MeterFilter.denyNameStartsWith(...)) .meterFilter(MeterFilter.acceptNameStartsWith(...));
148. Micrometer: MeterFilter .meterFilter(new MeterFilter() { @Override public Id map(Id id) { if (id.getName().equals("old")) return id.withName("new"); else return id; } });
149. Micrometer Tracing: Span Span span = tracer.nextSpan().name("test"); try (SpanInScope ws = tracer.withSpan(span.start())) { span.tag("userId", userId); span.event("logout"); } finally { span.end(); }
150. Micrometer: Observation 🚀 ObservationRegistry registry = ObservationRegistry.create(); registry.observationConfig().observationHandler(...); Observation observation = Observation.createNotStarted("talk",registry) .contextualName("talk observation") .lowCardinalityKeyValue("event", "SIO") .highCardinalityKeyValue("uid", userId);
151. Micrometer: Observation 🚀 try (Scope scope = observation.start().openScope()) { doSomething(); observation.event(Event.of("question")); } catch (Exception exception) { observation.error(exception); throw exception; } finally { observation.stop(); }
152. Micrometer: ObservationPredicate 🚀 Should the Observation be created or ignored (noop)? registry.observationConfig() .observationPredicate( (name, ctx) -> !name.startsWith("ignored") );
153. Micrometer: ObservationFilter 🚀 Modify the Context registry.observationConfig() .observationFilter( ctx -> ctx.addLowCardinalityKeyValue(...) );
154. Checkpoint Manual instrumentation works! 😎
155. What’s ~new? • Micrometer Tracing (Sleuth w/o Spring deps.) • Micrometer Docs Generator • Micrometer Context Propagation • Observation API • Exemplars • OTLP • Prometheus 1.x
156. 1. The fundamentals of Application Observability, why do we need it 2. How to apply these fundamentals to realistic scenarios in sample applications where having observability is crucial 3. How Micrometer provides a unified API to instrument your code for various signals and backends 4. What is Micrometer’s new Observation API and how to use it 5. What signals to watch in your own application 6. Spring Boot’s built-in observability features and how to customize them 7. How to avoid common issues 8. How to integrate metrics with distributed tracing and logs 9. How to visualize and analyze observability data to identify issues and optimize performance 10. How to troubleshoot issues faster and more effectively 11. The latest developments around Observability
157. Q&A
158. 1. The fundamentals of Application Observability, why do we need it 2. How to apply these fundamentals to realistic scenarios in sample applications where having observability is crucial 3. How Micrometer provides a unified API to instrument your code for various signals and backends 4. What is Micrometer’s new Observation API and how to use it 5. What signals to watch in your own application 6. Spring Boot’s built-in observability features and how to customize them 7. How to avoid common issues 8. How to integrate metrics with distributed tracing and logs 9. How to visualize and analyze observability data to identify issues and optimize performance 10. How to troubleshoot issues faster and more effectively 11. The latest developments around Observability
159. Thank you! slack.micrometer.io #springio2024