Storage problems rarely announce themselves as storage problems. A full disk stops a database from writing and an application from logging. A slow volume looks like a slow API. A failing drive shows up as occasional errors long before it dies. Storage monitoring catches these early, while there is still time to add space, move data or replace hardware.
Key metrics
- Capacity: space used and free, and how fast it is growing. The growth rate tells you when you will run out, which matters more than today's percentage.
- Inodes: on Linux file systems, millions of small files can exhaust inodes while space remains.
- IOPS: read and write operations per second, compared with what the device or volume can deliver.
- Throughput: megabytes per second read and written.
- Latency: how long each read or write takes. Rising latency is usually the first sign of an overloaded device.
- Queue depth and utilization: how many requests are waiting, and how busy the device is.
- Health: SMART warnings on drives, degraded RAID arrays, replication lag and I/O errors.
Types of storage to cover
- Direct-attached storage: local disks and SSDs in servers.
- SAN: block storage shared over Fibre Channel or iSCSI.
- NAS: file storage shared over NFS or SMB.
- Cloud block storage: volumes such as Amazon EBS, which have their own IOPS and throughput limits, and burst balances on some volume types.
- Object storage: services such as Amazon S3, where you track size, request errors and cost rather than IOPS.
- Persistent volumes in Kubernetes, which can fill up just like any disk.
Warning signs worth an alert
- A forecast that a volume will be full within a day or a week
- Sustained high latency or a long I/O queue on a database volume
- Exhausted burst credits on a cloud volume
- A degraded RAID array or a drive reporting SMART errors
- Replication falling behind
Storage monitoring with Atatus
Atatus infrastructure monitoring tracks disk metrics on every host alongside CPU, memory and processes, and Kubernetes monitoring alerts when persistent volumes run low on capacity. For databases, database monitoring ranks slow queries and shows lock waits and replication metrics, which often explain storage-related slowdowns. See also server monitoring.