Open s-volkov-1 opened 3 weeks ago
What I did as a workaround:
@Configuration(proxyBeanMethods = false)
class DbSchedulerConfig(
private val applicationTaskExecutor: ThreadPoolTaskExecutor
) {
@Bean
fun dbSchedulerCustomizer(): DbSchedulerCustomizer = object : DbSchedulerCustomizer {
// this is needed to register executor service metrics for db-scheduler
override fun executorService(): Optional<ExecutorService> {
return Optional.of(applicationTaskExecutor.threadPoolExecutor)
}
}
}
application.yml
spring:
task:
execution: # ! used as db-scheduler pool !
pool:
core-size: 9
max-size: 9
allow-core-thread-timeout: false
thread-name-prefix: task-executor-
shutdown:
await-termination: true
await-termination-period: PT4M # just in case, same as db-scheduler.shutdown-max-wait
db-scheduler:
# removed thread number setting
immediate-execution-enabled: true
shutdown-max-wait: 240s # 4 minutes
And can see executor's load now:
Prerequisites
Proposal
I am using db-scheduler-spring-boot-starter. It would be great if default ExecutorService (i.e. workers thread pool) could automatically be attached to micrometer metrics for executors.
Example metrics, which are automatically registered for two default spring's thread pools (for
@Async
and@Scheduled
):These executors are exposed as beans and connected to MeterRegistry in TaskExecutorMetricsAutoConfiguration.
So scheduler's ExecutorService could be added as well:
Workaround
One could take spring's executor (e.g. applicationTaskExecutor) and use it as db-scheduler worker ExecutorService via DbSchedulerCustomizer.executorService()
Thoughts
It seems like ExecutorService was intentionally not exposed as a bean, which possibly prevents it from being registered by TaskExecutorMetricsAutoConfiguration.
Bonus idea: metric returning overdue task count by names
Motivation
With ExecutorService's metrics users could monitor scheduler's load
Context
Involved classes: