OAL
基础知识
根本介绍
OAL(Observability Analysis Language)
是一门用来剖析流式数据的语言。因为
OAL
聚焦于度量Service
、Service Instance
和Endpoint
的指标,所以它学习和应用起来非常简单。
OAL
基于altlr
与javassist
将oal
脚本转化为动静生成的类文件。自从
6.3
版本后,OAL
引擎内置在OAP
服务器中,能够看做oal-rt(OAL Runtime)
。OAL
脚本地位OAL
配置目录下 (/config/oal
),使用者可能更改脚本并重启失效。留神:OAL
脚本依然是一门编译语言,oal-rt
动静的生成Java
代码。如果你配置了环境变量
SW_OAL_ENGINE_DEBUG=Y
,能在工作目录下的oal-rt
目录下找到生成的Class
文件。
语法
// 申明一个指标
METRICS_NAME = from(SCOPE.(* | [FIELD][,FIELD ...])) // 从某一个 SCOPE 中获取数据
[.filter(FIELD OP [INT | STRING])] // 能够过滤掉局部数据
.FUNCTION([PARAM][, PARAM ...]) // 应用某个聚合函数将数据聚合
// 禁用一个指标
disable(METRICS_NAME);
语法案例
oap-server/server-bootstrap/src/main/resources/oal/java-agent.oal
// 从 ServiceInstanceJVMMemory 的 used 获取数据,只须要 heapStatus 为 true 的数据,并取 long 型的平均值
instance_jvm_memory_heap = from(ServiceInstanceJVMMemory.used).filter(heapStatus == true).longAvg();
org.apache.skywalking.oap.server.core.source.ServiceInstanceJVMMemory
@ScopeDeclaration(id = SERVICE_INSTANCE_JVM_MEMORY, name = "ServiceInstanceJVMMemory", catalog = SERVICE_INSTANCE_CATALOG_NAME)
@ScopeDefaultColumn.VirtualColumnDefinition(fieldName = "entityId", columnName = "entity_id", isID = true, type = String.class)
public class ServiceInstanceJVMMemory extends Source {
@Override
public int scope() {return DefaultScopeDefine.SERVICE_INSTANCE_JVM_MEMORY;}
@Override
public String getEntityId() {return String.valueOf(id);
}
@Getter @Setter
private String id;
@Getter @Setter @ScopeDefaultColumn.DefinedByField(columnName = "name", requireDynamicActive = true)
private String name;
@Getter @Setter @ScopeDefaultColumn.DefinedByField(columnName = "service_name", requireDynamicActive = true)
private String serviceName;
@Getter @Setter @ScopeDefaultColumn.DefinedByField(columnName = "service_id")
private String serviceId;
@Getter @Setter
private boolean heapStatus;
@Getter @Setter
private long init;
@Getter @Setter
private long max;
@Getter @Setter
private long used;
@Getter @Setter
private long committed;
}
可供参考的官网文档:Observability Analysis Language
从一个案例开始剖析 OAL
原理
短少的类加载信息监控
默认的 APM/Instance
页面,短少对于 JVM Class
的信息(如下图所示),故这次将相干信息补齐。由这次案例来剖析 OAL
的原理。
在 Skywalking-04:扩大 Metric 监控信息 中,讲到了如何在已有 Source
类的状况下,减少一些指标。
这次间接连 Source
类以及 OAL
词法语法关键字都本人定义。
可供参考的官网文档:Source and Scope extension for new metrics
确定减少的指标
通过 Java ManagementFactory 解析这篇文章,能够确定监控指标为“以后加载类的数量”、“已卸载类的数量”、“一共加载类的数量”三个指标
ClassLoadingMXBean classLoadingMXBean = ManagementFactory.getClassLoadingMXBean();
// 以后加载类的数量
int loadedClassCount = classLoadingMXBean.getLoadedClassCount();
// 已卸载类的数量
long unloadedClassCount = classLoadingMXBean.getUnloadedClassCount();
// 一共加载类的数量
long totalLoadedClassCount = classLoadingMXBean.getTotalLoadedClassCount();
定义 agent
与 oap server
通信类
在 apm-protocol/apm-network/src/main/proto/language-agent/JVMMetric.proto
协定文件中减少如下定义。
在 apm-protocol/apm-network
目录下执行 mvn clean package -DskipTests=true
会生成新的相干 Java
类,org.apache.skywalking.apm.network.language.agent.v3.Class
该类就是咱们在代码中实际操作的类。
message Class {
int64 loadedClassCount = 1;
int64 unloadedClassCount = 3;
int64 totalLoadedClassCount = 2;
}
message JVMMetric {
int64 time = 1;
CPU cpu = 2;
repeated Memory memory = 3;
repeated MemoryPool memoryPool = 4;
repeated GC gc = 5;
Thread thread = 6;
// 在 JVM 指标中增加 Class 的定义
Class clazz = 7;
}
收集 agent
的信息后,将信息发送至 oap server
收集 Class
相干的指标信息
package org.apache.skywalking.apm.agent.core.jvm.clazz;
import org.apache.skywalking.apm.network.language.agent.v3.Class;
import java.lang.management.ClassLoadingMXBean;
import java.lang.management.ManagementFactory;
public enum ClassProvider {
/**
* instance
*/
INSTANCE;
private final ClassLoadingMXBean classLoadingMXBean;
ClassProvider() {this.classLoadingMXBean = ManagementFactory.getClassLoadingMXBean();
}
// 构建 class 的指标信息
public Class getClassMetrics() {int loadedClassCount = classLoadingMXBean.getLoadedClassCount();
long unloadedClassCount = classLoadingMXBean.getUnloadedClassCount();
long totalLoadedClassCount = classLoadingMXBean.getTotalLoadedClassCount();
return Class.newBuilder().setLoadedClassCount(loadedClassCount)
.setUnloadedClassCount(unloadedClassCount)
.setTotalLoadedClassCount(totalLoadedClassCount)
.build();}
}
在 org.apache.skywalking.apm.agent.core.jvm.JVMService#run
办法中,将 class
相干指标设置到 JVM
指标类中
@Override
public void run() {long currentTimeMillis = System.currentTimeMillis();
try {JVMMetric.Builder jvmBuilder = JVMMetric.newBuilder();
jvmBuilder.setTime(currentTimeMillis);
jvmBuilder.setCpu(CPUProvider.INSTANCE.getCpuMetric());
jvmBuilder.addAllMemory(MemoryProvider.INSTANCE.getMemoryMetricList());
jvmBuilder.addAllMemoryPool(MemoryPoolProvider.INSTANCE.getMemoryPoolMetricsList());
jvmBuilder.addAllGc(GCProvider.INSTANCE.getGCList());
jvmBuilder.setThread(ThreadProvider.INSTANCE.getThreadMetrics());
// 设置 class 的指标
jvmBuilder.setClazz(ClassProvider.INSTANCE.getClassMetrics());
// 将 JVM 的指标放在阻塞队列中
// org.apache.skywalking.apm.agent.core.jvm.JVMMetricsSender#run 办法,会将相干信息发送至 oap server
sender.offer(jvmBuilder.build());
} catch (Exception e) {LOGGER.error(e, "Collect JVM info fail.");
}
}
创立 Source
类
public class DefaultScopeDefine {
public static final int SERVICE_INSTANCE_JVM_CLASS = 11000;
/** Catalog of scope, the metrics processor could use this to group all generated metrics by oal rt. */
public static final String SERVICE_INSTANCE_CATALOG_NAME = "SERVICE_INSTANCE";
}
package org.apache.skywalking.oap.server.core.source;
import lombok.Getter;
import lombok.Setter;
import static org.apache.skywalking.oap.server.core.source.DefaultScopeDefine.SERVICE_INSTANCE_CATALOG_NAME;
import static org.apache.skywalking.oap.server.core.source.DefaultScopeDefine.SERVICE_INSTANCE_JVM_CLASS;
@ScopeDeclaration(id = SERVICE_INSTANCE_JVM_CLASS, name = "ServiceInstanceJVMClass", catalog = SERVICE_INSTANCE_CATALOG_NAME)
@ScopeDefaultColumn.VirtualColumnDefinition(fieldName = "entityId", columnName = "entity_id", isID = true, type = String.class)
public class ServiceInstanceJVMClass extends Source {
@Override
public int scope() {return SERVICE_INSTANCE_JVM_CLASS;}
@Override
public String getEntityId() {return String.valueOf(id);
}
@Getter @Setter
private String id;
@Getter @Setter @ScopeDefaultColumn.DefinedByField(columnName = "name", requireDynamicActive = true)
private String name;
@Getter @Setter @ScopeDefaultColumn.DefinedByField(columnName = "service_name", requireDynamicActive = true)
private String serviceName;
@Getter @Setter @ScopeDefaultColumn.DefinedByField(columnName = "service_id")
private String serviceId;
@Getter @Setter
private long loadedClassCount;
@Getter @Setter
private long unloadedClassCount;
@Getter @Setter
private long totalLoadedClassCount;
}
将从 agent
获取到的信息,发送至 SourceReceive
在 org.apache.skywalking.oap.server.analyzer.provider.jvm.JVMSourceDispatcher
进行如下批改
public void sendMetric(String service, String serviceInstance, JVMMetric metrics) {long minuteTimeBucket = TimeBucket.getMinuteTimeBucket(metrics.getTime());
final String serviceId = IDManager.ServiceID.buildId(service, NodeType.Normal);
final String serviceInstanceId = IDManager.ServiceInstanceID.buildId(serviceId, serviceInstance);
this.sendToCpuMetricProcess(service, serviceId, serviceInstance, serviceInstanceId, minuteTimeBucket, metrics.getCpu());
this.sendToMemoryMetricProcess(service, serviceId, serviceInstance, serviceInstanceId, minuteTimeBucket, metrics.getMemoryList());
this.sendToMemoryPoolMetricProcess(service, serviceId, serviceInstance, serviceInstanceId, minuteTimeBucket, metrics.getMemoryPoolList());
this.sendToGCMetricProcess(service, serviceId, serviceInstance, serviceInstanceId, minuteTimeBucket, metrics.getGcList());
this.sendToThreadMetricProcess(service, serviceId, serviceInstance, serviceInstanceId, minuteTimeBucket, metrics.getThread());
// class 指标解决
this.sendToClassMetricProcess(service, serviceId, serviceInstance, serviceInstanceId, minuteTimeBucket, metrics.getClazz());
}
private void sendToClassMetricProcess(String service,
String serviceId,
String serviceInstance,
String serviceInstanceId,
long timeBucket,
Class clazz) {
// 拼装 Source 对象
ServiceInstanceJVMClass serviceInstanceJVMClass = new ServiceInstanceJVMClass();
serviceInstanceJVMClass.setId(serviceInstanceId);
serviceInstanceJVMClass.setName(serviceInstance);
serviceInstanceJVMClass.setServiceId(serviceId);
serviceInstanceJVMClass.setServiceName(service);
serviceInstanceJVMClass.setLoadedClassCount(clazz.getLoadedClassCount());
serviceInstanceJVMClass.setUnloadedClassCount(clazz.getUnloadedClassCount());
serviceInstanceJVMClass.setTotalLoadedClassCount(clazz.getTotalLoadedClassCount());
serviceInstanceJVMClass.setTimeBucket(timeBucket);
// 将 Source 对象发送至 SourceReceive 进行解决
sourceReceiver.receive(serviceInstanceJVMClass);
}
在 OAL
词法定义和语法定义中退出 Source
相干信息
在 oap-server/oal-grammar/src/main/antlr4/org/apache/skywalking/oal/rt/grammar/OALLexer.g4
定义 Class
关键字
// Keywords
FROM: 'from';
FILTER: 'filter';
DISABLE: 'disable';
SRC_ALL: 'All';
SRC_SERVICE: 'Service';
SRC_SERVICE_INSTANCE: 'ServiceInstance';
SRC_ENDPOINT: 'Endpoint';
SRC_SERVICE_RELATION: 'ServiceRelation';
SRC_SERVICE_INSTANCE_RELATION: 'ServiceInstanceRelation';
SRC_ENDPOINT_RELATION: 'EndpointRelation';
SRC_SERVICE_INSTANCE_JVM_CPU: 'ServiceInstanceJVMCPU';
SRC_SERVICE_INSTANCE_JVM_MEMORY: 'ServiceInstanceJVMMemory';
SRC_SERVICE_INSTANCE_JVM_MEMORY_POOL: 'ServiceInstanceJVMMemoryPool';
SRC_SERVICE_INSTANCE_JVM_GC: 'ServiceInstanceJVMGC';
SRC_SERVICE_INSTANCE_JVM_THREAD: 'ServiceInstanceJVMThread';
SRC_SERVICE_INSTANCE_JVM_CLASS:'ServiceInstanceJVMClass'; // 在 OAL 词法定义中增加 Class 的关键字
SRC_DATABASE_ACCESS: 'DatabaseAccess';
SRC_SERVICE_INSTANCE_CLR_CPU: 'ServiceInstanceCLRCPU';
SRC_SERVICE_INSTANCE_CLR_GC: 'ServiceInstanceCLRGC';
SRC_SERVICE_INSTANCE_CLR_THREAD: 'ServiceInstanceCLRThread';
SRC_ENVOY_INSTANCE_METRIC: 'EnvoyInstanceMetric';
在 oap-server/oal-grammar/src/main/antlr4/org/apache/skywalking/oal/rt/grammar/OALParser.g4
增加 Class
关键字
source
: SRC_ALL | SRC_SERVICE | SRC_DATABASE_ACCESS | SRC_SERVICE_INSTANCE | SRC_ENDPOINT |
SRC_SERVICE_RELATION | SRC_SERVICE_INSTANCE_RELATION | SRC_ENDPOINT_RELATION |
SRC_SERVICE_INSTANCE_JVM_CPU | SRC_SERVICE_INSTANCE_JVM_MEMORY | SRC_SERVICE_INSTANCE_JVM_MEMORY_POOL |
SRC_SERVICE_INSTANCE_JVM_GC | SRC_SERVICE_INSTANCE_JVM_THREAD | SRC_SERVICE_INSTANCE_JVM_CLASS |// 在 OAL 语法定义中增加词法定义中定义的关键字
SRC_SERVICE_INSTANCE_CLR_CPU | SRC_SERVICE_INSTANCE_CLR_GC | SRC_SERVICE_INSTANCE_CLR_THREAD |
SRC_ENVOY_INSTANCE_METRIC |
SRC_BROWSER_APP_PERF | SRC_BROWSER_APP_PAGE_PERF | SRC_BROWSER_APP_SINGLE_VERSION_PERF |
SRC_BROWSER_APP_TRAFFIC | SRC_BROWSER_APP_PAGE_TRAFFIC | SRC_BROWSER_APP_SINGLE_VERSION_TRAFFIC
;
在 oap-server/oal-grammar
目录下执行 mvn clean package -DskipTests=true
会生成新的相干 Java
类
定义 OAL
指标
在 oap-server/server-bootstrap/src/main/resources/oal/java-agent.oal
中增加基于 OAL
语法的 Class
相干指标定义
// 以后加载类的数量
instance_jvm_class_loaded_class_count = from(ServiceInstanceJVMClass.loadedClassCount).longAvg();
// 已卸载类的数量
instance_jvm_class_unloaded_class_count = from(ServiceInstanceJVMClass.unloadedClassCount).longAvg();
// 一共加载类的数量
instance_jvm_class_total_loaded_class_count = from(ServiceInstanceJVMClass.totalLoadedClassCount).longAvg();
配置 UI
面板
将如下界面配置导入 APM
面板中
{
"name": "Instance",
"children": [{
"width": "3",
"title": "Service Instance Load",
"height": "250",
"entityType": "ServiceInstance",
"independentSelector": false,
"metricType": "REGULAR_VALUE",
"metricName": "service_instance_cpm",
"queryMetricType": "readMetricsValues",
"chartType": "ChartLine",
"unit": "CPM - calls per minute"
},
{
"width": 3,
"title": "Service Instance Throughput",
"height": "250",
"entityType": "ServiceInstance",
"independentSelector": false,
"metricType": "REGULAR_VALUE",
"metricName": "service_instance_throughput_received,service_instance_throughput_sent",
"queryMetricType": "readMetricsValues",
"chartType": "ChartLine",
"unit": "Bytes"
},
{
"width": "3",
"title": "Service Instance Successful Rate",
"height": "250",
"entityType": "ServiceInstance",
"independentSelector": false,
"metricType": "REGULAR_VALUE",
"metricName": "service_instance_sla",
"queryMetricType": "readMetricsValues",
"chartType": "ChartLine",
"unit": "%",
"aggregation": "/",
"aggregationNum": "100"
},
{
"width": "3",
"title": "Service Instance Latency",
"height": "250",
"entityType": "ServiceInstance",
"independentSelector": false,
"metricType": "REGULAR_VALUE",
"metricName": "service_instance_resp_time",
"queryMetricType": "readMetricsValues",
"chartType": "ChartLine",
"unit": "ms"
},
{
"width": 3,
"title": "JVM CPU (Java Service)",
"height": "250",
"entityType": "ServiceInstance",
"independentSelector": false,
"metricType": "REGULAR_VALUE",
"metricName": "instance_jvm_cpu",
"queryMetricType": "readMetricsValues",
"chartType": "ChartLine",
"unit": "%",
"aggregation": "+",
"aggregationNum": ""
},
{
"width": 3,
"title": "JVM Memory (Java Service)",
"height": "250",
"entityType": "ServiceInstance",
"independentSelector": false,
"metricType": "REGULAR_VALUE",
"metricName": "instance_jvm_memory_heap, instance_jvm_memory_heap_max,instance_jvm_memory_noheap, instance_jvm_memory_noheap_max",
"queryMetricType": "readMetricsValues",
"chartType": "ChartLine",
"unit": "MB",
"aggregation": "/",
"aggregationNum": "1048576"
},
{
"width": 3,
"title": "JVM GC Time",
"height": "250",
"entityType": "ServiceInstance",
"independentSelector": false,
"metricType": "REGULAR_VALUE",
"metricName": "instance_jvm_young_gc_time, instance_jvm_old_gc_time",
"queryMetricType": "readMetricsValues",
"chartType": "ChartLine",
"unit": "ms"
},
{
"width": 3,
"title": "JVM GC Count",
"height": "250",
"entityType": "ServiceInstance",
"independentSelector": false,
"metricType": "REGULAR_VALUE",
"queryMetricType": "readMetricsValues",
"chartType": "ChartBar",
"metricName": "instance_jvm_young_gc_count, instance_jvm_old_gc_count"
},
{
"width": 3,
"title": "JVM Thread Count (Java Service)",
"height": "250",
"entityType": "ServiceInstance",
"independentSelector": false,
"metricType": "REGULAR_VALUE",
"queryMetricType": "readMetricsValues",
"chartType": "ChartLine",
"metricName": "instance_jvm_thread_live_count, instance_jvm_thread_daemon_count, instance_jvm_thread_peak_count,instance_jvm_thread_deadlocked,instance_jvm_thread_monitor_deadlocked"
},
{
"width": 3,
"title": "JVM Thread State Count (Java Service)",
"height": "250",
"entityType": "ServiceInstance",
"independentSelector": false,
"metricType": "REGULAR_VALUE",
"metricName": "instance_jvm_thread_new_thread_count,instance_jvm_thread_runnable_thread_count,instance_jvm_thread_blocked_thread_count,instance_jvm_thread_wait_thread_count,instance_jvm_thread_time_wait_thread_count,instance_jvm_thread_terminated_thread_count",
"queryMetricType": "readMetricsValues",
"chartType": "ChartBar"
},
{
"width": 3,
"title": "JVM Class Count (Java Service)",
"height": "250",
"entityType": "ServiceInstance",
"independentSelector": false,
"metricType": "REGULAR_VALUE",
"metricName": "instance_jvm_class_loaded_class_count,instance_jvm_class_unloaded_class_count,instance_jvm_class_total_loaded_class_count",
"queryMetricType": "readMetricsValues",
"chartType": "ChartArea"
},
{
"width": 3,
"title": "CLR CPU (.NET Service)",
"height": "250",
"entityType": "ServiceInstance",
"independentSelector": false,
"metricType": "REGULAR_VALUE",
"metricName": "instance_clr_cpu",
"queryMetricType": "readMetricsValues",
"chartType": "ChartLine",
"unit": "%"
},
{
"width": 3,
"title": "CLR GC (.NET Service)",
"height": "250",
"entityType": "ServiceInstance",
"independentSelector": false,
"metricType": "REGULAR_VALUE",
"metricName": "instance_clr_gen0_collect_count, instance_clr_gen1_collect_count, instance_clr_gen2_collect_count",
"queryMetricType": "readMetricsValues",
"chartType": "ChartBar"
},
{
"width": 3,
"title": "CLR Heap Memory (.NET Service)",
"height": "250",
"entityType": "ServiceInstance",
"independentSelector": false,
"metricType": "REGULAR_VALUE",
"metricName": "instance_clr_heap_memory",
"queryMetricType": "readMetricsValues",
"chartType": "ChartLine",
"unit": "MB",
"aggregation": "/",
"aggregationNum": "1048576"
},
{
"width": 3,
"title": "CLR Thread (.NET Service)",
"height": "250",
"entityType": "ServiceInstance",
"independentSelector": false,
"metricType": "REGULAR_VALUE",
"queryMetricType": "readMetricsValues",
"chartType": "ChartLine",
"metricName": "instance_clr_available_completion_port_threads,instance_clr_available_worker_threads,instance_clr_max_completion_port_threads,instance_clr_max_worker_threads"
}
]
}
后果校验
能够看到导入的界面中,曾经有 Class
相干指标了
代码奉献
- Add some new thread metric and class metric to JVMMetric #7230
- add some new thread metric and class metric to JVMMetric #52
- Remove Terminated State and New State in JVMMetric (#7230) #53
- Add some new thread metric and class metric to JVMMetric (#7230) #7243
参考文档
- Observability Analysis Language
- Source and Scope extension for new metrics
- Java ManagementFactory 解析
分享并记录所学所见