MongoDB Instance Summary dashboard shows "No data" for mongos services on PMM 3.6.0

Percona Monitoring and Management (PMM)
1

Description:

The MongoDB Instance Summary dashboard in PMM 3.6.0 shows “No data” on several panels when a mongos service is selected. The Service Name dropdown includes mongos services, so users expect data to appear, but the panel queries only use mongod-specific metrics (mongodb_mongod_metrics_document_total, mongodb_mongod_metrics_query_executor_total) that are not available on mongos instances.

Additionally, when using --enable-all-collectors on mongos via the Percona Operator’s mongosParams, the indexstats collector queries $indexStats across all shards, causing the same collection+index combination to be returned multiple times (once per shard) with identical labels. This results in duplicate metric errors for mongodb_indexstats_accesses_ops.

Steps to Reproduce:

Issue 1: Instance Summary “No data” on mongos

  1. Deploy a sharded MongoDB cluster using Percona Operator with PMM enabled

  2. Navigate to MongoDB > Instance Summary dashboard

  3. Select a mongos service in the Service Name dropdown

  4. Observe “No data” on Document Operations and Query Efficiency panels

Issue 2: --enable-all-collectors duplicate metric error on mongos

  1. Set mongosParams: "--enable-all-collectors" in the PerconaServerMongoDB CR

  2. Wait for mongos pods to restart with the new PMM config

  3. Check pmm-client container logs: kubectl logs <mongos-pod> -c pmm-client

  4. Observe repeated duplicate metric errors

Version:

  1. PMM Server: 3.6.0
  2. PMM Client: 3.6.0 (percona/pmm-client:3.6.0)
  3. Percona Server for MongoDB: 8.0.8-3
  4. Percona Operator for MongoDB: 1.21.1
  5. Platform: AWS EKS

Logs:

pmm-client container log from mongos pod (Issue 2):

level=ERROR source=http_error_logger.go:47 msg="error gathering metrics:" 
error="[from Gatherer #2] collected metric \"mongodb_indexstats_accesses_ops\" 
{ label:{name:\"cl_role\" value:\"mongos\"} 
  label:{name:\"collection\" value:\"lookup_franchise_no_stage_2026_01_09\"} 
  label:{name:\"database\" value:\"cdp\"} 
  label:{name:\"key_name\" value:\"hashed_id_idx\"} 
  untyped:{value:0}} 
was collected before with the same name and label values" 
total_errors=36

This error repeats for every collection/index combination across all shards (36 errors per scrape cycle).

Expected Result:

Issue 1:

  • Instance Summary should either filter mongos services out of the Service Name dropdown, or conditionally switch queries to mongos-compatible metrics (e.g., mongodb_mongod_op_latencies_* which works on both mongod and mongos), or hide panels that are not applicable for mongos.

Issue 2:

  • --enable-all-collectors on mongos should work without duplicate metric errors. The exporter should either deduplicate index stats from multiple shards, or add a shard label to differentiate them.

Actual Result:

Issue 1:

  • Document Operations panel shows “No data” because it queries mongodb_mongod_metrics_document_total which is mongod-only.

  • Query Efficiency panel shows “No data” for the same reason.

  • Other panels like Operation Latencies work correctly because mongodb_mongod_op_latencies_* is collected from mongos as well.

Issue 2:

  • 36 duplicate metric errors per scrape cycle, one for each collection+index combination across shards.

  • The indexstats collector on mongos queries all shards via $indexStats, returning the same collection+index from multiple shards with identical labels, causing the Prometheus collector to reject duplicates.

Additional Information:

  • Cluster topology: 3 shards x 3 replset nodes + 3 mongos + 3 config servers

  • The Router Summary dashboard correctly displays mongos metrics. This issue is specific to Instance Summary.

  • Current workaround for Issue 2: only use --enable-all-collectors on mongodParams, not mongosParams:

yaml

pmm:
  enabled: true
  mongodParams: "--enable-all-collectors --environment=prod"
  mongosParams: "--environment=prod"
  • Some mongodb_mongod_* metrics are collected from mongos (e.g., op_latencies), while others are not (e.g., metrics_document_total). This naming inconsistency makes it difficult to determine which metrics are available for mongos without trial and error.
2 
mongosParams: "--enable-all-collectors --disable-collectors=indexstats --environment=stage"

It worked for me.

3

glad to hear it’s sorted out