nori_regression: correct the plugin's Synthefy-facing details (key env var, model list, required model)

influxdata/nori_regression/README.md

@@ -60,8 +60,8 @@ The Nori gateway API key is a secret and is **never** read from trigger argument
 body (both are logged). It is resolved, in order:
 
 1. a non-empty `X-Nori-Api-Key: <key>` request header (HTTP trigger only), then
-2. the `NORI_API_KEY` environment variable set on the InfluxDB host (required for the scheduled
-   trigger).
+2. the `SYNTHEFY_NORI_API_KEY` environment variable set on the InfluxDB host (required for the
+   scheduled trigger).
 
 The key is intentionally **not** accepted in the `Authorization` header: InfluxDB parses
 `Authorization` for its own request authorization, so a key placed there never reaches the plugin.
@@ -84,6 +84,7 @@ InfluxDB host. It must be an `https://` URL; plain `http://` is accepted only fo
 | `measurement` | string | required | Source measurement (table) to read from. |
 | `field` | string | required | The numeric field to predict. The plugin trains on the rows where it is present and predicts the rows where it is null. |
 | `feature_fields` | string | required | Numeric feature columns (X) used to predict `field`, **space-separated** (for example `temp humidity`). Use spaces, not commas (`--trigger-arguments` splits argument pairs on commas) and not dots (a field name may contain a `.`). |
+| `model` | string | required | The Nori gateway slug to call. There is no default: the slug selects a priced model, so the plugin will not choose one for you. See [Supported models](#supported-models). Trigger argument only. |
 
 A column name that contains a space cannot be expressed in `feature_fields` as a trigger argument,
 because every string form splits on whitespace. Name such a column from a TOML array
@@ -97,7 +98,6 @@ because every string form splits on whitespace. Name such a column from a TOML a
 | `start_time` | string | *(none)* | ISO 8601 start of a fixed window. Given alone, the window ends now. |
 | `end_time` | string | *(none)* | ISO 8601 end of a fixed window. Given alone, the window starts one `window` earlier. |
 | `tags` | string | *(none)* | Filter to a single series. Format: `key:val key2:val2` (space-separated pairs, one value per key). A token without a `:` is rejected. Required when the window holds more than one series. |
-| `model` | string | `synthefy/nori-30m` | The Nori gateway slug to call. See [Supported models](#supported-models). |
 | `output_measurement` | string | `<measurement>_regressed` | Measurement to write predictions to. Must differ from `measurement`. |
 | `target_database` | string | *(trigger db)* | Write predictions to a different database. |
 | `dry_run` | boolean | `false` | Log the first few predictions and return them all, without writing anything. |
@@ -107,7 +107,7 @@ because every string form splits on whitespace. Name such a column from a TOML a
 | `max_predict_rows` | integer | `5000` | Cap on rows predicted per run; the most recent rows are kept and the rest wait for a later run. |
 | `max_read_rows` | integer | `50000` | Ceiling on rows read from InfluxDB in one run, applied as a `LIMIT` on the query. The most recent rows are read, and a truncated read is logged with a warning. This bounds the plugin's memory: a row costs roughly 0.7 KB while it is held, so the default is about 35 MB. |
 | `predict_batch_size` | integer | `1000` | Rows per gateway call. Each batch re-sends the training context and is billed separately, so a larger value costs less. |
-| `request_timeout` | string | `300s` | Timeout for one gateway call. A cold start has been measured at 60-130 seconds, so keep this well above that. |
+| `request_timeout` | string | `300s` | Timeout for one gateway call. A model that has scaled to zero cold-starts on the first request, measured between roughly one and four minutes depending on the variant, so keep this well above the warm response time. |
 | `max_retries` | integer | `3` | Maximum attempts per gateway call and per write. `1` disables retry. |
 | `config_file_path` | string | *(none)* | Path to a TOML file supplying every parameter, relative to `PLUGIN_DIR`. Cannot be combined with other inline arguments or a request body. |
 
@@ -185,7 +185,7 @@ influxdb3 create trigger \
 3. Set the Nori gateway key on the InfluxDB host, so the scheduled trigger can read it:
 
    ```bash
-   export NORI_API_KEY="<your Nori API key>"
+   export SYNTHEFY_NORI_API_KEY="<your Nori API key>"
    ```
 
 ### Data requirements
@@ -273,7 +273,7 @@ influxdb3 create trigger \
   --database mydb \
   --path "gh:influxdata/nori_regression/nori_regression.py" \
   --trigger-spec "every:15m" \
-  --trigger-arguments measurement=sensors,field=pressure,feature_fields="temp humidity",tags=site:A,min_history=3 \
+  --trigger-arguments measurement=sensors,field=pressure,feature_fields="temp humidity",tags=site:A,model=synthefy/nori-30m,min_history=3 \
   nori_example
 ```
 
@@ -305,7 +305,7 @@ Call the HTTP endpoint (exposed at `/api/v3/engine/<path>`), passing the Nori ke
 
 ```bash
 curl -X POST http://localhost:8181/api/v3/engine/nori_regress \
-  -H "X-Nori-Api-Key: $NORI_API_KEY" \
+  -H "X-Nori-Api-Key: $SYNTHEFY_NORI_API_KEY" \
   -H "Content-Type: application/json" \
   -d '{"measurement":"sensors","field":"pressure","feature_fields":["temp","humidity"],"tags":{"site":"A"}}'
 ```
@@ -335,7 +335,7 @@ The top-level `status` is one of `success`, `partial`, `skipped`, `dry_run` or `
 
 ```bash
 curl -X POST http://localhost:8181/api/v3/engine/nori_regress \
-  -H "X-Nori-Api-Key: $NORI_API_KEY" \
+  -H "X-Nori-Api-Key: $SYNTHEFY_NORI_API_KEY" \
   -H "Content-Type: application/json" \
   -d '{"measurement":"sensors","field":"pressure","feature_fields":["temp","humidity"],"tags":{"site":"A"},"start_time":"2026-01-01T00:00:00Z","end_time":"2026-02-01T00:00:00Z"}'
 ```
@@ -347,7 +347,7 @@ reads the `window` before it.
 
 ```bash
 curl -X POST http://localhost:8181/api/v3/engine/nori_regress \
-  -H "X-Nori-Api-Key: $NORI_API_KEY" \
+  -H "X-Nori-Api-Key: $SYNTHEFY_NORI_API_KEY" \
   -H "Content-Type: application/json" \
   -d '{"measurement":"sensors","field":"pressure","feature_fields":["temp","humidity"],"tags":{"site":"A"},"dry_run":true}'
 ```
@@ -414,15 +414,16 @@ ORDER BY 1 DESC
 
 ## Supported models
 
-The `model` argument is the Nori gateway slug your API key is granted:
+The `model` argument is the Nori gateway slug your API key is granted. It is **required**: the
+slug selects a priced model, so the plugin will not choose one on your behalf. Synthefy's own
+client and local package take the same position.
 
-| Slug | Parameters | Notes |
-|---|---|---|
-| `synthefy/nori-30m` | ~29M | The default, and the variant Synthefy's own documentation recommends. Priced higher and slower to cold-start (measured at ~125s). |
-| `synthefy/nori-6m` | ~6M | Cheaper per request and faster to cold-start (measured at ~69s). |
+Synthefy publishes the current models, their sizes and their slugs at
+[docs.synthefy.com/nori/quickstart#models](https://docs.synthefy.com/nori/quickstart#models). That list is the authoritative one:
+it changes when Synthefy releases a variant, and a slug not on it will not route.
 
-Which one predicts better depends on your data; try both with `dry_run=true` before committing a
-schedule to one.
+Which model predicts better depends on your data, and the larger ones cost more per request and
+take longer to cold-start. Try a couple with `dry_run=true` before committing a schedule to one.
 
 The bare `synthefy/nori` slug has been retired and no longer routes; the plugin rejects it with a
 pointed message rather than letting the gateway answer `404`. One API key from the
@@ -469,8 +470,8 @@ caller-facing message with the full detail in `processing_engine_logs`.
 
 The plugin cannot find a Nori gateway key.
 
-**Solution:** set `NORI_API_KEY` on the InfluxDB host, or pass an `X-Nori-Api-Key: <key>` header
-when calling the HTTP trigger (see
+**Solution:** set `SYNTHEFY_NORI_API_KEY` on the InfluxDB host, or pass an
+`X-Nori-Api-Key: <key>` header when calling the HTTP trigger (see
 [Authentication](#authentication-for-the-nori-gateway)). An empty header value is ignored and the
 environment variable is used instead.
 
@@ -526,9 +527,9 @@ names `time`/`y`.
 
 #### Cold-start latency and timeouts
 
-The models scale to zero, so the first request after an idle period is slow: about 69 seconds for
-`synthefy/nori-6m` and 125 seconds for `synthefy/nori-30m` in measurement, and it can return a
-`503` or a non-JSON body from the fronting proxy once.
+The models scale to zero, so the first request after an idle period is slow: measurements have
+ranged from roughly one minute to nearly four, with the larger variants slower, and it can return
+a `503` or a non-JSON body from the fronting proxy once.
 
 **Solution:** the default `request_timeout` of `300s` and `max_retries` of `3` are set to absorb
 this; a `503`, a `429` and a connection error are retried with backoff. A read timeout is **not**

influxdata/nori_regression/nori_regression.py

@@ -9,7 +9,7 @@
         {"name": "start_time", "example": "2026-01-01T00:00:00Z", "description": "ISO start of a fixed window instead of a trailing one. With skip_existing left on, a schedule over a fixed window backfills and then stops calling the gateway.", "required": false},
         {"name": "end_time", "example": "2026-02-01T00:00:00Z", "description": "ISO end of a fixed window. Given alone, the window starts one `window` earlier.", "required": false},
         {"name": "tags", "example": "site:A", "description": "Filter to a single series. Format: key:val key2:val2 (space-separated pairs, single value per key). A token without ':' is rejected. Required if the window holds more than one series.", "required": false},
-        {"name": "model", "example": "synthefy/nori-30m", "description": "The Nori gateway slug to call. Current slugs: synthefy/nori-30m (default, ~29M params) and synthefy/nori-6m (cheaper and faster to cold-start). The bare 'synthefy/nori' slug is retired. Your API key must be granted the slug.", "required": false},
+        {"name": "model", "example": "synthefy/nori-30m", "description": "The Nori gateway slug to call. Required: there is no default, because the slug selects a priced model. The current list of models and their slugs is at https://docs.synthefy.com/nori/quickstart#models. The bare 'synthefy/nori' slug is retired. Your API key must be granted the slug.", "required": true},
         {"name": "output_measurement", "example": "sensors_regressed", "description": "Where to write predictions. Default: <measurement>_regressed.", "required": false},
         {"name": "target_database", "example": "predictions", "description": "Write predictions to this database instead of the trigger's own.", "required": false},
         {"name": "dry_run", "example": "false", "description": "If true, log predictions but do not write them.", "required": false},
@@ -32,7 +32,7 @@
         {"name": "start_time", "example": "2026-01-01T00:00:00Z", "description": "ISO start of the window. May also be in the request body. Given alone, the window ends now.", "required": false},
         {"name": "end_time", "example": "2026-02-01T00:00:00Z", "description": "ISO end of the window. May also be in the request body. Given alone, the window starts one `window` earlier.", "required": false},
         {"name": "dry_run", "example": "false", "description": "If true, log predictions but do not write them. May also be in the request body.", "required": false},
-        {"name": "model", "example": "synthefy/nori-30m", "description": "The Nori gateway slug to call. Trigger argument only: it selects a billed model, so the request body cannot override it.", "required": false},
+        {"name": "model", "example": "synthefy/nori-30m", "description": "The Nori gateway slug to call. Required, and a trigger argument only: it selects a billed model, so there is no default and the request body cannot override it.", "required": true},
         {"name": "output_measurement", "example": "sensors_regressed", "description": "Where to write predictions. Trigger argument only: the request body cannot override a write target.", "required": false},
         {"name": "target_database", "example": "predictions", "description": "Write predictions to this database instead of the trigger's own. Trigger argument only: the request body cannot override a write target.", "required": false},
         {"name": "min_history", "example": "50", "description": "Minimum labeled rows required to train. Trigger argument only.", "required": false},
@@ -80,18 +80,31 @@
 DEFAULT_GATEWAY_URL = "https://inference.baseten.co/predict"
 GATEWAY_URL_ENV_VAR = "NORI_GATEWAY_URL"
 
-# The key is a SECRET: it is read from the NORI_API_KEY environment variable on the InfluxDB host,
-# or (HTTP trigger only) from an incoming X-Nori-Api-Key request header. It is NEVER read from
-# trigger args or the request body (both are logged), and never from the incoming `Authorization`
-# header (InfluxDB consumes that header for its own request authorization).
-API_KEY_ENV_VAR = "NORI_API_KEY"
+# The key is a SECRET: it is read from the SYNTHEFY_NORI_API_KEY environment variable on the
+# InfluxDB host, or (HTTP trigger only) from an incoming X-Nori-Api-Key request header. It is NEVER
+# read from trigger args or the request body (both are logged), and never from the incoming
+# `Authorization` header (InfluxDB consumes that header for its own request authorization).
+#
+# The name carries the vendor prefix because an InfluxDB host runs plugins from several authors:
+# an unprefixed NORI_API_KEY says nothing about whose service it authenticates against.
+API_KEY_ENV_VAR = "SYNTHEFY_NORI_API_KEY"
 API_KEY_HEADER = "X-Nori-Api-Key"
 
 # Slugs name their parameter count so the model behind a slug never silently changes. The bare
 # `synthefy/nori` slug was retired and now returns 404, so it is rejected with a pointed message.
-DEFAULT_MODEL_SLUG = "synthefy/nori-30m"
+# There is no default model. Synthefy's own client and local package both require an explicit
+# size and raise rather than pick one, because a variant is a priced choice: selecting one on the
+# operator's behalf spends their money and pins them to a slug they never named. This plugin
+# follows that. The slug below is only the example shown in help text and suggested when a retired
+# slug is used.
+EXAMPLE_MODEL_SLUG = "synthefy/nori-30m"
 RETIRED_MODEL_SLUGS = {"synthefy/nori"}
 
+# Which models exist, and their sizes, is Synthefy's to publish and changes when they release one.
+# This plugin names only its default and points at the vendor's list, so a new variant does not
+# make the plugin's documentation wrong.
+MODEL_LIST_URL = "https://docs.synthefy.com/nori/quickstart#models"
+
 # Gateway faults worth another attempt: 408/409/425 (transient conflicts), 429 (per-key rate limit,
 # 50/min) and every 5xx (a cold start can 503, or 500 once). A 4xx outside that set is a permanent
 # input, key or slug problem, so retrying it only wastes the caller's time and the gateway's quota.
@@ -120,8 +133,9 @@
 # that STARTS WITH "@": @format and @jinja interpolate `env`, @read_file reads the host filesystem,
 # @get reads other settings. Every one of dynaconf's ~30 tokens begins with this single character,
 # and nothing else triggers them (a leading space or a doubled @ is inert), so refusing a leading
-# "@" is a complete guard. Without it a request body of {"measurement": "@format {env[NORI_API_KEY]}"}
-# comes back resolved in the error message - the plugin's own secret, handed to the caller.
+# "@" is a complete guard. Without it a request body of
+# {"measurement": "@format {env[SYNTHEFY_NORI_API_KEY]}"} comes back resolved in the error
+# message - the plugin's own secret, handed to the caller.
 DYNACONF_TOKEN_PREFIX = "@"
 
 # Parameters an earlier revision accepted. They are rejected by name rather than ignored, so an
@@ -186,7 +200,7 @@ class GatewayError(PublicError):
     Validator("field", default="", cast=str),
     Validator("feature_fields", default="", cast=parse_delimited_list),
     Validator("window", default="30d", cast=parse_timedelta),
-    Validator("model", default=DEFAULT_MODEL_SLUG, cast=str),
+    Validator("model", default="", cast=str),
     Validator("dry_run", default=False, cast=parse_bool),
     Validator("skip_existing", default=True, cast=parse_bool),
     Validator("min_history", default=50, gte=1, cast=int),
@@ -320,11 +334,17 @@ def _normalize_config(cfg) -> dict:
             f"feature_fields cannot include the target field or 'time': {reserved}"
         )
 
-    model = str(cfg.get("model") or DEFAULT_MODEL_SLUG).strip()
+    model = str(cfg.get("model") or "").strip()
+    if not model:
+        raise ConfigError(
+            "`model` is required: the Nori gateway slug to call, for example "
+            f"{EXAMPLE_MODEL_SLUG!r}. There is no default, because the slug selects a priced "
+            f"model. The current slugs are listed at {MODEL_LIST_URL}."
+        )
     if model in RETIRED_MODEL_SLUGS:
         raise ConfigError(
-            f"model slug {model!r} is retired and no longer routes. Use 'synthefy/nori-30m' "
-            f"(~29M parameters) or 'synthefy/nori-6m' (cheaper, faster cold start)."
+            f"model slug {model!r} is retired and no longer routes. Use {EXAMPLE_MODEL_SLUG!r} "
+            f"or another current slug, listed at {MODEL_LIST_URL}."
         )
 
     try:
@@ -648,7 +668,7 @@ def _gateway_url() -> str:
 
 def _get_api_key(request_headers=None) -> str:
     """Resolve the gateway key: a non-empty X-Nori-Api-Key header wins (HTTP trigger), else the
-    NORI_API_KEY environment variable.
+    SYNTHEFY_NORI_API_KEY environment variable.
 
     An empty or non-string header value falls through to the environment rather than winning it: a
     header sent with no value used to suppress the fallback and send `Api-Key ` with no key at all.

influxdata/nori_regression/nori_regression_config_scheduler.toml

@@ -24,9 +24,11 @@ feature_fields = ["temp", "humidity"]
 # Filter to a single series. A run must resolve to one series; the plugin fails if the window
 # holds more. Default: no filter.
 #tags = { site = "A" }
-# Nori gateway slug. "synthefy/nori-30m" (default, ~29M parameters) or "synthefy/nori-6m"
-# (cheaper, faster cold start). The bare "synthefy/nori" slug is retired.
-#model = "synthefy/nori-30m"
+# Nori gateway slug. REQUIRED: there is no default, because the slug selects a priced model.
+# The current models and their slugs are
+# listed at https://docs.synthefy.com/nori/quickstart#models
+# The bare "synthefy/nori" slug is retired.
+model = "synthefy/nori-30m"
 # Measurement to write predictions to. Default: "<measurement>_regressed".
 #output_measurement = "sensors_regressed"
 # Write predictions to another database. Default: the trigger's database.