InfluxDB中合并Heapster测量指标：如何创建自定义数据库筛选所需指标

Great question! Let’s tackle your two asks one by one—first creating a custom InfluxDB instance with only those three Heapster measurements, then merging those metrics for easier querying.

Absolutely, you can set up an InfluxDB instance that only stores these three specific measurements. There are two common approaches depending on whether you're starting fresh or migrating from an existing Heapster-InfluxDB setup:

• Option 1: Deploy a new InfluxDB and configure Heapster to write only to it

  1. First, deploy a standalone InfluxDB instance in your Kubernetes cluster (using a Deployment + Service + PersistentVolumeClaim for persistence). Once it's up, create a dedicated database for your metrics:

     influx -host <influxdb-service-ip> -execute "CREATE DATABASE heapster_custom"
     

  2. Modify your Heapster Deployment to target this new InfluxDB instance. Update the command arguments to include the InfluxDB sink pointing to your custom database:

     args:
       - --source=kubernetes:https://kubernetes.default
       - --sink=influxdb:http://<influxdb-service-name>:8086?db=heapster_custom
     

     Heapster’s default schema only writes the cpu, memory, and network measurements (along with their respective fields for usage/bandwidth), so this setup will automatically populate your custom InfluxDB with exactly those three measurements—no extra data will be stored here.

• Option 2: Migrate existing metrics to a new custom InfluxDB
  If you already have an existing Heapster-InfluxDB setup with extra measurements, you can export just the three you care about and import them into a new database:

  1. Export the targeted measurements from your existing InfluxDB:

     influxd export -database heapster -measurement cpu -measurement memory -measurement network > heapster_custom_backup.bkp
     

  2. Create the new database in your target InfluxDB (same as step 1 in Option 1), then import the backup:

     influx -host <target-influxdb-ip> -database heapster_custom -import -path=heapster_custom_backup.bkp
     

Merging these metrics typically means querying them together so you can view correlated data (e.g., CPU and memory usage for the same pod at the same time). The approach depends on which version of InfluxDB you’re using:

For InfluxDB 1.x (common with Heapster) #

Use InfluxQL’s JOIN clause to correlate metrics by shared tags (like pod_name, node_name, or namespace) and timestamps. Since Heapster uses consistent tags across all three measurements, this works smoothly.

Example query to merge pod-level metrics over the last hour, aggregated into 1-minute intervals:

SELECT 
  cpu_agg.value AS cpu_usage,
  memory_agg.value AS memory_usage,
  network_agg.value AS network_bandwidth
FROM (
  SELECT MEAN(value) AS value 
  FROM cpu 
  WHERE time > now() - 1h 
  GROUP BY time(1m), pod_name
) cpu_agg
JOIN (
  SELECT MEAN(value) AS value 
  FROM memory 
  WHERE time > now() - 1h 
  GROUP BY time(1m), pod_name
) memory_agg 
ON cpu_agg.time = memory_agg.time AND cpu_agg.pod_name = memory_agg.pod_name
JOIN (
  SELECT MEAN(value) AS value 
  FROM network 
  WHERE time > now() - 1h 
  GROUP BY time(1m), pod_name
) network_agg 
ON cpu_agg.time = network_agg.time AND cpu_agg.pod_name = network_agg.pod_name

This query first aggregates each metric to smooth out noise, then joins them by matching timestamps and pod names to get a unified view of resource usage per pod.

For InfluxDB 2.x (if you’ve upgraded) #

Use Flux (InfluxDB 2.x’s query language) which has built-in pivoting functionality to merge measurements into a single table:

from(bucket: "heapster_custom")
  |> range(start: -1h)
  |> filter(fn: (r) => r._measurement == "cpu" or r._measurement == "memory" or r._measurement == "network")
  |> pivot(
    rowKey: ["_time", "pod_name"],
    columnKey: ["_measurement"],
    valueColumn: "_value"
  )

This will pivot the data so each row contains the CPU, memory, and network values for a specific pod at a specific time—much cleaner than nested JOINs.

内容的提问来源于stack exchange，提问作者samar