prometheus 每次显示指标时都无法获取指标答案

【问题标题】：promethus can not get metrics everytime resultting gaps when displaying metricsprometheus 每次显示指标时都无法获取指标
【发布时间】：2021-05-24 08:08:51
【问题描述】：

我在 k8s 上以独立模式部署了几个 flink，并通过一个 promethus-pushgateway 导出它们的指标。

问题在于：
指标数据间歇性地到达promethus，导致在grafana中显示时点之间出现间隙

click me, show the gapped graph

普罗米修斯目标：

monitoring/pushgateway/0 (1/1 up)
Endpoint: http://172.19.88.111:9091/metrics
State   : UP
Labels: endpoint="tcp" instance="172.19.88.111:9091" job="pushgateway" namespace="flink-sql" pod="pushgateway-76d64545dd-6prdn" service="pushgateway"

我直接查询pushgateway，但每次都获取不到所有的metris

bash-5.0# date  &&  curl  -s http://pushgateway.flink-sql:9091/metrics      | grep flink_jobmanager_numRegisteredTaskManagers
Mon May 24 07:15:17 UTC 2021
# HELP flink_jobmanager_numRegisteredTaskManagers numRegisteredTaskManagers (scope: jobmanager)
# TYPE flink_jobmanager_numRegisteredTaskManagers gauge
flink_jobmanager_numRegisteredTaskManagers{host="flink_jobmanager",instance="",job="flink-sql"} 0

bash-5.0# date  &&  curl  -s http://pushgateway.flink-sql:9091/metrics      | grep flink_jobmanager_numRegisteredTaskManagers
Mon May 24 07:15:18 UTC 2021
# HELP flink_jobmanager_numRegisteredTaskManagers numRegisteredTaskManagers (scope: jobmanager)
# TYPE flink_jobmanager_numRegisteredTaskManagers gauge
flink_jobmanager_numRegisteredTaskManagers{host="flink_jobmanager",instance="",job="flink-sql"} 0

bash-5.0# date  &&  curl  -s http://pushgateway.flink-sql:9091/metrics      | grep flink_jobmanager_numRegisteredTaskManagers
Mon May 24 07:15:18 UTC 2021
# HELP flink_jobmanager_numRegisteredTaskManagers numRegisteredTaskManagers (scope: jobmanager)
# TYPE flink_jobmanager_numRegisteredTaskManagers gauge
flink_jobmanager_numRegisteredTaskManagers{host="172_19_90_175",instance="",job="model1122"} 8
flink_jobmanager_numRegisteredTaskManagers{host="flink_jobmanager",instance="",job="flink-sql"} 0
bash-5.0# date  &&  curl  -s http://pushgateway.flink-sql:9091/metrics      | grep flink_jobmanager_numRegisteredTaskManagers
Mon May 24 07:15:19 UTC 2021
# HELP flink_jobmanager_numRegisteredTaskManagers numRegisteredTaskManagers (scope: jobmanager)
# TYPE flink_jobmanager_numRegisteredTaskManagers gauge
flink_jobmanager_numRegisteredTaskManagers{host="172_19_90_175",instance="",job="model1122"} 8
bash-5.0# date  &&  curl  -s http://pushgateway.flink-sql:9091/metrics      | grep flink_jobmanager_numRegisteredTaskManagers
Mon May 24 07:15:20 UTC 2021
# HELP flink_jobmanager_numRegisteredTaskManagers numRegisteredTaskManagers (scope: jobmanager)
# TYPE flink_jobmanager_numRegisteredTaskManagers gauge
flink_jobmanager_numRegisteredTaskManagers{host="flink_jobmanager",instance="",job="flink-sql"} 0
bash-5.0# date  &&  curl  -s http://pushgateway.flink-sql:9091/metrics      | grep flink_jobmanager_numRegisteredTaskManagers
Mon May 24 07:15:20 UTC 2021
# HELP flink_jobmanager_numRegisteredTaskManagers numRegisteredTaskManagers (scope: jobmanager)
# TYPE flink_jobmanager_numRegisteredTaskManagers gauge
flink_jobmanager_numRegisteredTaskManagers{host="172_19_90_175",instance="",job="model1122"} 8
flink_jobmanager_numRegisteredTaskManagers{host="flink_jobmanager",instance="",job="flink-sql"} 0
flink_jobmanager_numRegisteredTaskManagers{host="jobmanager",instance="",job="model"} 20
bash-5.0# date  &&  curl  -s http://pushgateway.flink-sql:9091/metrics      | grep flink_jobmanager_numRegisteredTaskManagers
Mon May 24 07:15:20 UTC 2021
# HELP flink_jobmanager_numRegisteredTaskManagers numRegisteredTaskManagers (scope: jobmanager)
# TYPE flink_jobmanager_numRegisteredTaskManagers gauge
flink_jobmanager_numRegisteredTaskManagers{host="172_19_90_175",instance="",job="model1122"} 8
flink_jobmanager_numRegisteredTaskManagers{host="flink_jobmanager",instance="",job="flink-sql"} 0
bash-5.0# date  &&  curl  -s http://pushgateway.flink-sql:9091/metrics      | grep flink_jobmanager_numRegisteredTaskManagers
Mon May 24 07:15:21 UTC 2021
# HELP flink_jobmanager_numRegisteredTaskManagers numRegisteredTaskManagers (scope: jobmanager)
# TYPE flink_jobmanager_numRegisteredTaskManagers gauge
flink_jobmanager_numRegisteredTaskManagers{host="flink_jobmanager",instance="",job="flink-sql"} 0
bash-5.0# date  &&  curl  -s http://pushgateway.flink-sql:9091/metrics      | grep flink_jobmanager_numRegisteredTaskManagers
Mon May 24 07:15:22 UTC 2021
# HELP flink_jobmanager_numRegisteredTaskManagers numRegisteredTaskManagers (scope: jobmanager)
# TYPE flink_jobmanager_numRegisteredTaskManagers gauge
flink_jobmanager_numRegisteredTaskManagers{host="172_19_90_175",instance="",job="model1122"} 8
flink_jobmanager_numRegisteredTaskManagers{host="flink_jobmanager",instance="",job="flink-sql"} 0
bash-5.0# date  &&  curl  -s http://pushgateway.flink-sql:9091/metrics      | grep flink_jobmanager_numRegisteredTaskManagers
Mon May 24 07:15:22 UTC 2021
# HELP flink_jobmanager_numRegisteredTaskManagers numRegisteredTaskManagers (scope: jobmanager)
# TYPE flink_jobmanager_numRegisteredTaskManagers gauge
flink_jobmanager_numRegisteredTaskManagers{host="172_19_90_175",instance="",job="model1122"} 8
bash-5.0# date  &&  curl  -s http://pushgateway.flink-sql:9091/metrics      | grep flink_jobmanager_numRegisteredTaskManagers
Mon May 24 07:15:23 UTC 2021
# HELP flink_jobmanager_numRegisteredTaskManagers numRegisteredTaskManagers (scope: jobmanager)
# TYPE flink_jobmanager_numRegisteredTaskManagers gauge
flink_jobmanager_numRegisteredTaskManagers{host="flink_jobmanager",instance="",job="flink-sql"} 0

bash-5.0# date  &&  curl  -s http://pushgateway.flink-sql:9091/metrics      | grep flink_jobmanager_numRegisteredTaskManagers
Mon May 24 07:15:23 UTC 2021
# HELP flink_jobmanager_numRegisteredTaskManagers numRegisteredTaskManagers (scope: jobmanager)
# TYPE flink_jobmanager_numRegisteredTaskManagers gauge
flink_jobmanager_numRegisteredTaskManagers{host="172_19_90_175",instance="",job="model1122"} 8
flink_jobmanager_numRegisteredTaskManagers{host="flink_jobmanager",instance="",job="flink-sql"} 0
flink_jobmanager_numRegisteredTaskManagers{host="jobmanager",instance="",job="model"} 20

bash-5.0# date  &&  curl  -s http://pushgateway.flink-sql:9091/metrics      | grep flink_jobmanager_numRegisteredTaskManagers
Mon May 24 07:15:24 UTC 2021
# HELP flink_jobmanager_numRegisteredTaskManagers numRegisteredTaskManagers (scope: jobmanager)
# TYPE flink_jobmanager_numRegisteredTaskManagers gauge
flink_jobmanager_numRegisteredTaskManagers{host="flink_jobmanager",instance="",job="flink-sql"} 0

bash-5.0# date  &&  curl  -s http://pushgateway.flink-sql:9091/metrics      | grep flink_jobmanager_numRegisteredTaskManagers
Mon May 24 07:15:24 UTC 2021
# HELP flink_jobmanager_numRegisteredTaskManagers numRegisteredTaskManagers (scope: jobmanager)
# TYPE flink_jobmanager_numRegisteredTaskManagers gauge
flink_jobmanager_numRegisteredTaskManagers{host="172_19_90_175",instance="",job="model1122"} 8
flink_jobmanager_numRegisteredTaskManagers{host="flink_jobmanager",instance="",job="flink-sql"} 0
bash-5.0# date  &&  curl  -s http://pushgateway.flink-sql:9091/metrics      | grep flink_jobmanager_numRegisteredTaskManagers
Mon May 24 07:15:25 UTC 2021

bash-5.0# date  &&  curl  -s http://pushgateway.flink-sql:9091/metrics      | grep flink_jobmanager_numRegisteredTaskManagers
Mon May 24 07:15:26 UTC 2021
# HELP flink_jobmanager_numRegisteredTaskManagers numRegisteredTaskManagers (scope: jobmanager)
# TYPE flink_jobmanager_numRegisteredTaskManagers gauge
flink_jobmanager_numRegisteredTaskManagers{host="flink_jobmanager",instance="",job="flink-sql"} 0

bash-5.0# date  &&  curl  -s http://pushgateway.flink-sql:9091/metrics      | grep flink_jobmanager_numRegisteredTaskManagers
Mon May 24 07:15:27 UTC 2021
# HELP flink_jobmanager_numRegisteredTaskManagers numRegisteredTaskManagers (scope: jobmanager)
# TYPE flink_jobmanager_numRegisteredTaskManagers gauge
flink_jobmanager_numRegisteredTaskManagers{host="flink_jobmanager",instance="",job="flink-sql"} 0

我的 flink-conf.yaml 中的配置

metrics.reporter.promgateway.class: org.apache.flink.metrics.prometheus.PrometheusPushGatewayReporter
metrics.reporter.promgateway.host: pushgateway.flink-sql
metrics.reporter.promgateway.port: 9091
metrics.reporter.promgateway.jobName: flink-sql
metrics.reporter.promgateway.randomJobNameSuffix: false
metrics.reporter.promgateway.deleteOnShutdown: false
metrics.reporter.promgateway.interval: 3 SECONDS

即使将promethus Scrape intervalmetrics.reporter.promgateway.interval设置为1秒，也没有效果；

【问题讨论】：

我已经经历过类似的事情，这是由于内存使用率高。基本上，检查您的 Flink 作业是否处于背压状态并使用 htop 检查内存使用情况
@Felipe 即使没有运行 flink 作业也没有运气，
我不会尝试每 1 秒刮 Flink，5 秒或 15 秒就足够了。我认为这可能是差距的根源，因为一次刮擦没有完成，而普罗米修斯已经开始了另一次刮擦。差距是否仅出现在flink_jobmanager_numRegisteredTaskManagers 或其他人身上？
也许你可以尝试注册2个TM并测试一下。然后注册 4 个 TM 并进行测试。然后 6 TM 和测试。如果您对 2 个 TM 没有问题，那么在您注册更多 TM 时，它会在某个时候出现。
是的，所有指标都会出现这种情况

标签： apache-flink prometheus prometheus-pushgateway

【解决方案1】：

我猜：

promethus 的间隙图结果没有存储连续数据。
Prometheus 的指标数据来自 PushaGateWay。
PushGateWay 的指标数据来自 JobManager/TaskManager。
JobManager/TaskManager 上报给 PushaGateWay 的数据没有被 PushaGateWay 缓存。
所以promethus定期查询Pushgateway时，只得到pushgateway当前的数据，而不是JobManager/TaskManager上报的所有数据。

我所经历的似乎是这样的，但还没有定论。PushGateWay毕竟必须发挥作用。当然，不考虑 Flink 的 metric reportor 是否按预期定期上报数据

现在我通过 promethus 直接从 Jobmanage/Taskmanager 抓取数据的新解决方案解决了差距问题。

【讨论】：