A tiny push-based metrics store and live dashboard. In-memory, bit-packed, gone in a few hours.
dash is a minimal Grafana/Graphite-style tool for watching what is happening right now. Clients push metrics over HTTP+JSON; dash holds them in memory using Gorilla-style compression (delta-of-delta timestamps, XOR-encoded values — a couple of bytes per point instead of sixteen) and serves a built-in live dashboard. Retention is a few hours, enforced by dropping whole compressed chunks, and there is deliberately no persistence: restart the process and you start fresh.
Metric names form a Graphite-style tree split at the dots
(web.frontend.latency_ms), browsable in the UI and queryable with globs.
The storage itself stays flat — the hierarchy is derived on demand, so empty
branches disappear on their own as data rotates out.
Everything ships in one static binary: the web UI and its chart library are embedded at compile time.
- Push model over HTTP+JSON — batch-friendly, curl-friendly
- Gorilla-compressed in-memory storage, sharded for concurrent ingest
- Dot-hierarchy metric names with Graphite-style
/findand*globs - Built-in live dashboard: collapsible series tree, filtering, time-range presets, auto-refresh, light/dark theme, shareable URLs, named bookmarks
- Retention by chunk rotation — no compaction, no cleanup jobs, no disk
- Single binary, four dependencies, no configuration required
cargo run &
# push a metric (timestamp optional, epoch milliseconds)
curl -s localhost:9090/ingest \
-H 'content-type: application/json' \
-d '{"m":"test.answer","v":42}'
# or feed it a demo hierarchy
./scripts/demo.sh &Then open http://localhost:9090/.
Environment variables only:
| Variable | Default | Description |
|---|---|---|
DASH_ADDR |
127.0.0.1:9090 |
Listen address |
DASH_RETENTION_SECS |
10800 |
How long data is kept (3 h) |
DASH_CHUNK_SECS |
1800 |
Chunk window; expiry drops whole chunks |
Body is a JSON array of points, or a single point object. ts is epoch
milliseconds and defaults to server time. Points that are out of order for
their series, older than the retention window, or non-finite are rejected
(counted, not an error).
curl -s localhost:9090/ingest -H 'content-type: application/json' -d '[
{"m":"web.frontend.requests","v":1042},
{"m":"web.frontend.latency_ms","v":73.2,"ts":1788103068401}
]'
# {"accepted":2,"rejected":0}m may repeat and may contain *, which matches within one dot-separated
segment (web.*.latency_ms matches web.backend.latency_ms, not
web.a.b.latency_ms). from/to accept epoch ms, now, or relative values
like -15m; step accepts 10s-style durations or plain ms. Defaults:
from=-1h, to=now, step auto-picked for ~300 buckets. Expansion is
capped at 20 series ("truncated":true past that). All series share one
time grid. Each bucket keeps its last sample, and empty buckets carry the
newest retained value forward; buckets before the first known value are
null.
curl -s 'localhost:9090/query?m=web.*.latency_ms&from=-15m&step=10s'
# {"step":10000,"from":...,"to":...,"truncated":false,
# "series":[{"m":"web.backend.latency_ms","points":[[1788103060000,71.8],...]}, ...]}One level of the metric tree, Graphite-style. q defaults to *; its
segment count selects the depth.
curl -s 'localhost:9090/find?q=web.*'
# [{"path":"web.backend","name":"backend","leaf":false,"leaves":2},
# {"path":"web.frontend","name":"frontend","leaf":false,"leaves":2}]Flat sorted list of every stored metric name.
Totals for probes and curiosity: {"series":14,"points":56,"bytes":572}.
The published image listens on 0.0.0.0:9090:
docker run --rm -p 9090:9090 ghcr.io/nsg/dash:latest