Current metrics catalog
Exact metric names, units, kinds, and labels the Linux agent collects today, plus how that catalog relates to the CloudStats website.
Updated Nov 7, 2026
This is the canonical V1 metric list the Linux agent collects and CloudStats accepts. Names are lowercase dotted identifiers. Units and kinds below match the agent catalog.
Collectors always attempt this full list on Linux (from /proc and filesystem stats). The CloudStats website is a separate application. It can display latest and history values for names in this catalog, including labels. This page does not claim that every name has its own chart. Server detail is expected to show CPU, memory, load, and filesystem values.
Catalog#
| Metric | Unit | Kind | Labels |
|---|---|---|---|
system.uptime.seconds | seconds | gauge | |
load.1 | load average | gauge | |
load.5 | load average | gauge | |
load.15 | load average | gauge | |
cpu.utilization.percent | percent | gauge | |
cpu.user.percent | percent | gauge | |
cpu.system.percent | percent | gauge | |
memory.total.bytes | bytes | gauge | |
memory.used.bytes | bytes | gauge | |
memory.available.bytes | bytes | gauge | |
memory.utilization.percent | percent | gauge | |
swap.total.bytes | bytes | gauge | |
swap.used.bytes | bytes | gauge | |
filesystem.total.bytes | bytes | gauge | device, mountpoint, fstype |
filesystem.used.bytes | bytes | gauge | device, mountpoint, fstype |
filesystem.available.bytes | bytes | gauge | device, mountpoint, fstype |
filesystem.utilization.percent | percent | gauge | device, mountpoint, fstype |
disk.read.bytes_total | bytes | counter | device |
disk.write.bytes_total | bytes | counter | device |
disk.read.bytes_per_second | bytes/second | derived rate | device |
disk.write.bytes_per_second | bytes/second | derived rate | device |
network.rx.bytes_total | bytes | counter | interface |
network.tx.bytes_total | bytes | counter | interface |
network.rx.bytes_per_second | bytes/second | derived rate | interface |
network.tx.bytes_per_second | bytes/second | derived rate | interface |
Kinds:
- Gauge: value at collection time
- Counter: running total from the host counters
- Derived rate: bytes/second from counter deltas. A gap or counter reset invalidates that interval (no point is emitted)
Percent values are in 0..100. Byte gauges, counters, rates, and load values are non-negative.
Collector notes (Linux)#
These notes are from the current agent collectors, not from website layout.
| Area | Source | Behavior |
|---|---|---|
| Uptime | /proc/uptime | First field, seconds |
| Load | /proc/loadavg | 1 / 5 / 15 minute averages |
| CPU | /proc/stat (cpu line) | Percents from the delta versus the previous sample. Omitted on the first collect. Utilization is busy time (user + nice + system + irq + softirq + steal) over total. cpu.user.percent is user + nice. cpu.system.percent is system + irq + softirq. |
| Memory | /proc/meminfo | memory.used.bytes is total minus available. Available falls back to free + buffers + cached when MemAvailable is missing. Swap used is total minus free. |
| Filesystem | /proc/mounts + statfs | One series set per mountpoint. Skips virtual types such as proc, sysfs, cgroup, overlay, and paths under /proc and /sys. tmpfs is collected (it is not in the skip list). |
| Disk | /proc/diskstats | Whole disks. Skips names starting with loop, ram, sr, fd, zram, nbd, and partition names (sda1, nvme0n1p1). Counters are 512-byte sectors converted to bytes. Rates omitted until a second sample for that device; omitted on reset. |
| Network | /proc/net/dev | Skips lo. Rates omitted until a second sample for that interface; omitted on reset. |
One collector failure is logged (collector failed) and does not stop the others.
Default collect interval is 10 seconds. See Service, status, and logs.
Website versus collectors#
| Layer | What it has |
|---|---|
| Linux agent | Collects the catalog above, with first-sample and skip rules |
| Latest snapshot | Last accepted batch, labels kept (not flattened to one value per name) |
| History | Per metric name plus labels |
| CloudStats website | Separate app. Expected detail charts: CPU, memory, load, filesystem. Other catalog names may or may not have a dedicated widget. |
Do not assume a missing website panel means the agent is not collecting that metric. Confirm with Troubleshooting.