drip

drip is a small, opt-in companion binary that collects host metrics on a Linux box and POSTs them into NanoTDB using line protocol. It is intended for edge boxes — Raspberry Pi, NAS appliances, small Linux servers — where you want one place to look at CPU, memory, disk, network, load, and temperature without setting up a separate metrics pipeline.

NanoTDB's default dashboard.json is pre-wired to the metric names drip emits, so a Pi + nanotdb + drip deployment is a usable dashboard from the first start.

You do not need drip to use NanoTDB. Any source that can POST line protocol to /api/v1/import works.

---

What it collects

Collector	Sample metrics
cpu	cpu.user, cpu.system, cpu.idle, cpu.iowait, cpu.busy_pct, cpu.temp_mdeg, sys.uptime_sec
memory	mem.total, mem.free, mem.available, mem.cached, mem.swaptotal, mem.swapfree
process	proc.{exe}.count, proc.{exe}.rss_bytes, proc.{exe}.cpu_pct per configured executable
disk	per block device: disk.{dev}.reads, disk.{dev}.busy_pct, disk.{dev}.iops, disk.{dev}.read_kbps, disk.{dev}.write_kbps; per mount: diskfs.{mount}.bytes_used, diskfs.{mount}.used_pct
io	io.pgpgin, io.pgpgout, io.pswpin, io.pswpout
sd_write_probe	disk.sd_write_probe_ms — measured write+fsync latency on the configured directory
network	per interface: net.{iface}.rx_bytes, net.{iface}.tx_bytes, plus packet and error counters
loadavg	sys.load1, sys.load5, sys.load15, sys.procs_running, sys.procs_total
onewire	temp.{name}.mdeg for DS18B20 sensors, with friendly-name mapping

Each collector can be enabled or disabled independently in drip.toml.

---

Configuration

drip.toml controls the server endpoint, the destination database, the collection cadence, and per-collector options.

Minimum:

[drip]
server_url = "http://localhost:8428"
database = "metrics"
collection_interval_ms = 10000
timeout_ms = 5500

Per-collector enable flags follow:

[collectors.cpu]
enabled = true

[collectors.memory]
enabled = true

[collectors.process]
enabled = true
exe_names = ["drip", "nanotdb"]

[collectors.disk]
enabled = true

[collectors.sd_write_probe]
enabled = true
directory = "/tmp"
bytes = 32000
every_n_cycles = 6
metric = "disk.sd_write_probe_ms"

[collectors.network]
enabled = true
skip = ["lo", "br-*", "docker*"]

[collectors.loadavg]
enabled = true

[collectors.onewire]
enabled = true
auto_discover = true
base_path = "/sys/bus/w1/devices"
max_valid_mdeg = 85000

[collectors.onewire.devices]
"28-0120481ddd35" = "office_wet"
"28-01214506d428" = "out_dry"
"28-0121450f31ef" = "office_dry"

See the in-repo cmd/drip/drip.toml for the full annotated example, which is what the default dashboard.json is tuned to.

---

Running

Build it from source:

go build -o drip ./cmd/drip

Or download a prebuilt binary from Releases. drip is published for Linux ARM/x86_64 targets only.

Run it pointing at a running NanoTDB:

./drip --config drip.toml

It will start posting samples on every collection_interval_ms tick.

To verify, check the destination DB once data starts arriving:

curl "http://localhost:8428/api/v1/metrics?db=metrics"

---

Running as a service

drip has a systemd template alongside the NanoTDB one. See RUN_AS_A_SERVICE.md for the recommended pi-user layout and unit files.

---

Notes on edge use

• The sd_write_probe collector measures write+fsync latency on a directory you choose. Point it at an SD-card-backed path on a Pi and you'll see the card's actual responsiveness over time — extremely useful when SD cards start degrading.
• Process CPU is emitted as a per-cycle delta. The first cycle starts at 0.
• One-wire devices that aren't explicitly named fall back to using their device ID as the metric suffix.