Prometheus Integration

At the front of the TSE, there is a Prometheus-compatible adapter. Without a separate Prometheus server, you can use collection (remote_write) and visualization (PromQL) exactly as in the Prometheus ecosystem.

[Prometheus · Grafana Agent] --remote_write--> TSE --> Cassandra
[Grafana (Prometheus 데이터소스)] --PromQL--> TSE

Not a separate process

The adapter runs as the same process inside the TSE's Jetty. There is nothing to launch and no extra port to open — paths are simply split at /prometheus (the TSE native API is at /api/v1/*). Authentication also uses the TSE's Basic Auth as-is.

1. Pushing data in — `remote_write`

Write this in prometheus.yml or in your Grafana Agent configuration.

remote_write:
  - url: http://<TSE 호스트>:7800/prometheus/api/v1/write
    basic_auth:
      username: tse
      password: "<PP_TSE_PASSWORD>"

The payload is snappy-compressed + protobuf (remote_write 1.0), and Prometheus and Agent handle it automatically. On success you get 204, on a malformed payload 400 (no retry), and on a temporary failure 500 (with retry). NaN and ±Inf are discarded at the storage stage.

2. Viewing in Grafana

Choose Prometheus as the datasource type and fill it in like this.

Item	Value
URL	`http://<TSE 호스트>:7800/prometheus`
Access	Server (proxy)
Auth	Basic auth — `tse` / `<PP_TSE_PASSWORD>`
HTTP Method	POST recommended

When you click "Save & Test", Grafana recognizes it as Prometheus via /prometheus/api/v1/status/buildinfo. From then on, you can use PromQL as-is in panels.

Different path from the bundled PlantPulse Dashboard

TSE also has a bundled PlantPulse Dashboard that uses a dedicated datasource plugin. You can choose either one — if you're comfortable with PromQL, use this one; if you want ready-made dashboards, use the bundled one.

3. How it maps

Prometheus	TSE
`__name__` label	Metric name
Other labels	Tags
Sample `(ts_ms, float64)`	DataPoint (double, ms)

Prometheus's naming rules ([a-zA-Z_:][a-zA-Z0-9_:]*) are compatible with TSE's, so everything moves over without loss.

4. Endpoints

All are under /prometheus and follow the Prometheus HTTP API conventions ({"status":"success","data":…}).

Method·Path	Purpose
`POST /prometheus/api/v1/write`	remote_write reception
`GET`·`POST` `/prometheus/api/v1/query`	Instant query
`GET`·`POST` `/prometheus/api/v1/query_range`	Range query (`start`·`end`·`step`)
`GET /prometheus/api/v1/labels`	Label names list
`GET /prometheus/api/v1/label/{name}/values`	Label values list
`GET`·`POST` `/prometheus/api/v1/series`	Series matching (`match[]`)
`GET /prometheus/api/v1/status/buildinfo`	Grafana recognition

Time is in Unix seconds (fractional allowed) or RFC3339, and step is 15s·1m·1h·500ms or unitless seconds. Timestamps in responses are in seconds, and values are strings (per the Prometheus convention).

5. Supported PromQL

Category	Supported
Selectors	bare · matchers (`=` `!=` `=~` `!~`) · `__name__` matcher · range selectors (`[5m]`, unit `ms s m h d w y`)
Arithmetic	numeric literals · unary `-` · parentheses · `+ - * /` (scalar∘scalar, vector∘scalar, vector∘vector 1-to-1)
Comparison	`== != > < >= <=` (+ `bool`)
Functions	`rate` `irate` `increase` `delta` · `abs` `clamp_min` `clamp_max` · `histogram_quantile`
Aggregation	`sum` `avg` `min` `max` `count` + `by`/`without` · `topk` `bottomk`

rate(errors[5m]) / rate(total[5m])
topk(5, sum(rate(req_total[5m])) by (path))
histogram_quantile(0.95, rate(latency_seconds_bucket[5m]))

6. Not supported — rejected with 400, not silently wrong

Category	Not supported
Vector matching	`on(…)` · `ignoring(…)` · `group_left` · `group_right`
Set operations	`and` · `or` · `unless`
Time functions	`*_over_time` · `predict_linear` · `holt_winters`
Transformations	subquery(`expr[5m:1m]`) · `@` modifier · `offset`
Others	`scalar()` `vector()` `label_replace()` `label_join()` `sort()` `sort_desc()` `absent()`
Protocol	`remote_read` · remote_write 2.0 · exemplars · native histogram

Unsupported things come as "errors", not "wrong values"

Unsupported syntax is rejected with 400 {"status":"error","errorType":"bad_data",…}. This means it never silently provides plausible-looking numbers, so your dashboards will not lie without a word.

7. Subtle differences

This is not a full Prometheus, but a compatible adapter. In practice this is usually not an issue, but you should be aware of it when you need to compare values precisely.

What	How it differs
`rate`·`increase` extrapolation	It linearly extrapolates over the entire observed range (Prometheus's boundary extrapolation cap is not implemented) — slight differences at range boundaries
range window	`[t-range, t]` includes both endpoints (Prometheus uses left-open intervals). The lookback for instant selectors is the same as `(t-5m, t]`, with a default of 5 minutes
`/label/<name>/values`	`__name__` is accurate, but other labels are over-approximated with global tag values — sufficient for autocompletion
scalar comparison	Always evaluates to 0 or 1 (Prometheus requires `bool`). `3 > 2` maps directly to `1`

1. Pushing data in — remote_write​

2. Viewing in Grafana​

3. How it maps​

4. Endpoints​

5. Supported PromQL​

6. Not supported — rejected with 400, not silently wrong​

7. Subtle differences​

See also​