MongoDBintermediate

MongoDB Monitoring

The commands and metrics that show whether a MongoDB deployment is healthy, and the profiler settings that find slow operations.

2 min readIntermediateUpdated 31 Jul 2026Edit this page

MongoDB

Metrics that matter

Replication lag

Source: rs.status() member optimeDate

Seconds each secondary is behind the primary. Alert against the staleness budget your reads depend on.

Oplog window

Source: db.getReplicationInfo()

How far back the oplog reaches. If it falls below your maintenance window, a restarted secondary needs a full resync.

wiredTiger.cache bytes / configured max

Source: serverStatus()

Cache pressure. Sustained eviction with a full cache means the working set exceeds memory.

globalLock.currentQueue

Source: serverStatus()

Operations queued waiting for locks. Non-zero for sustained periods means contention.

connections.current / available

Source: serverStatus()

Connection saturation. Each connection costs about a megabyte of stack plus driver state.

opcounters and opLatencies

Source: serverStatus()

Operation mix and latency by type. A rising read latency with a flat operation count usually means cache or index trouble.

db.serverStatus().wiredTiger.cache["bytes currently in the cache"];
db.serverStatus().connections;
db.serverStatus().opLatencies;
rs.printSecondaryReplicationInfo();

Live operations

// What is running now, longer than a second.
db.currentOp({ "secs_running": { $gt: 1 }, "active": true });
 
// Kill a runaway operation by its opid.
db.killOp(12345);

Profiler

// Level 1 records operations slower than slowms. Level 2 records everything — very expensive.
db.setProfilingLevel(1, { slowms: 100, sampleRate: 1.0 });
 
db.system.profile.find({ millis: { $gt: 100 } })
  .sort({ ts: -1 }).limit(10)
  .forEach(op => printjson({
    ts: op.ts, ms: op.millis, ns: op.ns,
    plan: op.planSummary, docsExamined: op.docsExamined, nreturned: op.nreturned
  }));

planSummary: "COLLSCAN" is the clearest signal of a missing index. A large gap between docsExamined and nreturned means the index that exists is not selective enough.

The profiler writes to a capped collection in each database, so it self-limits — but it does add write overhead. Sample rather than record everything on a busy system.

Free tooling

mongostat --uri="mongodb://…" 5      # operations, queues, cache, network per interval
mongotop --uri="mongodb://…" 5       # read and write time per collection

mongotop is the fastest way to find which collection is consuming the server's time.

Alerts worth having

No primary in the replica set.
Replication lag beyond the read-staleness budget.
Oplog window below the longest expected maintenance outage.
Connection count approaching the limit.
WiredTiger cache eviction pressure sustained.
Any member not in PRIMARY or SECONDARY state.
Disk usage projection for the data volume.
For sharded clusters: balancer errors, jumbo chunks, and chunk imbalance between shards.