Forecast error evaluator plugin

The Forecast Error Evaluator Plugin validates forecast model accuracy for time series data in InfluxDB 3 Enterprise by comparing predicted values with actual observations. On every scheduled run the plugin matches the two measurements over a time window, computes an error metric (MSE, MAE, RMSE, MAPE, or SMAPE) for each matched timestamp, and notifies for the points that reach a configured threshold. It includes debounce logic to suppress transient anomalies and supports multi-channel notifications via the Notification Sender Plugin.

The metric is computed per timestamp rather than aggregated over the window, which is what lets min_condition_duration measure how long an elevated error persists. As a consequence rmse yields the same value as mae: the root of a single squared difference is its absolute value.

Configuration

Plugin parameters may be specified as key-value pairs in the --trigger-arguments flag (CLI) or in the trigger_arguments field (API) when creating a trigger. Some plugins support TOML configuration files, which can be specified using the plugin’s config_file_path parameter.

If a plugin supports multiple trigger specifications, some parameters may depend on the trigger specification that you use.

Plugin metadata

This plugin includes a JSON metadata schema in its docstring that defines supported trigger types and configuration parameters. This metadata enables the InfluxDB 3 Explorer UI to display and configure the plugin.

Required parameters

Thresholds

Levels are INFO, WARN, ERROR and CRITICAL, and each threshold must be above 0. Every supported metric is non-negative, so a threshold of 0 or below would flag every point of the window; such a level is skipped with a warning.

Levels are evaluated independently: a point whose error reaches several thresholds produces one notification per level, each with its own debounce state. Configure only the levels you want to be paged about. Levels are processed from the highest threshold down, so max_notifications_per_run is spent on the most severe alerts first.

Malformed segments, and a level given twice, are skipped with a warning and the remaining levels still apply. If no level survives parsing, the run logs an error and stops without notifying.

Notification parameters

Timing parameters

Authentication parameters

Sender-specific parameters

Slack notifications

Discord notifications

HTTP notifications

SMS notifications (via Twilio)

Matching forecast to actual values

Forecast and actual rows are matched with an inner join on the timestamp plus the tags that both measurements share. Tags present in only one of them are ignored with a warning, so a forecast table without the tag columns of the actual table still works.

Set rounding_freq when the two series are written with slightly different timestamps. Rounding coarser than the sampling interval puts several points into the same slot; only the earliest point of each slot is kept, and the number of collapsed rows is logged. Without that, matching would join every forecast of the slot against every actual value in it.

Rows where either value is missing are dropped before the metric is computed. For mape, rows with actual = 0 are skipped; for smape, rows where both values are 0 are skipped.

Alert state

Debounce and alert state live in the trigger-local cache, keyed by measurement, field, level and tag values:

• While an error stays above a threshold for less than min_condition_duration, the plugin logs the pending state and waits. The duration is measured in data time, and a pending start that has scrolled out of the window is discarded, so a gap in the data cannot stand in for a persistent error.
• After an alert is delivered, its timestamp is recorded and earlier or equal timestamps are never alerted again. Overlapping windows on successive runs therefore do not resend the same point.
• If delivery fails after all retries, the state is left untouched so the next run alerts on that point again.
• The cache is in-memory and trigger-local: restarting the server clears the debounce and last-alert state.

TOML configuration

To use a TOML configuration file, set the PLUGIN_DIR environment variable and specify the config_file_path in the trigger arguments. This is in addition to the --plugin-dir flag when starting InfluxDB 3 Enterprise. Relative paths are resolved against the first directory that is set: PLUGIN_DIR, then INFLUXDB3_PLUGIN_DIR, then the parent of VIRTUAL_ENV. Only that directory is used — the file is not looked up in the remaining ones.

When config_file_path is set, the TOML file provides the whole configuration and inline trigger arguments are ignored. INFLUXDB3_AUTH_TOKEN from the environment still applies when influxdb3_auth_token is not set in the file. In TOML, senders and error_thresholds can use native structures (a list and a table) instead of the inline string formats, though the inline strings are also accepted.

Example TOML configuration

forecast_error_config_scheduler.toml

For more information on using TOML configuration files, see the Using TOML Configuration Files section in the influxdb3_plugins/README.md.

Software Requirements

• InfluxDB 3 Enterprise: with the Processing Engine enabled.
• Notification Sender Plugin for InfluxDB 3 Enterprise: Required for sending notifications. See the influxdata/notifier plugin.
• Python packages:
  • influxdata-plugin-utils>=0.3.0 (configuration loading, parsing, and schema introspection)
  • pandas (for data processing)
  • requests (for HTTP notifications)

Installation steps

1. Start InfluxDB 3 Enterprise with the Processing Engine enabled (--plugin-dir /path/to/plugins):

   influxdb3 serve \
     --node-id node0 \
     --object-store file \
     --data-dir ~/.influxdb3 \
     --plugin-dir ~/.plugins

2. Install required Python packages:

   influxdb3 install package influxdata-plugin-utils
   influxdb3 install package pandas
   influxdb3 install package requests

3. Install the influxdata/notifier plugin (required)

Trigger setup

Scheduled forecast validation

Run forecast error evaluation periodically:

influxdb3 create trigger \
  --database weather_forecasts \
  --path "gh:influxdata/forecast_error_evaluator/forecast_error_evaluator.py" \
  --trigger-spec "every:30m" \
  --trigger-arguments 'forecast_measurement=temperature_forecast,actual_measurement=temperature_actual,forecast_field=predicted_temp,actual_field=temp,error_metric=rmse,error_thresholds=INFO-"0.5":WARN-"1.0":ERROR-"2.0",window=1h,senders=slack,slack_webhook_url="$SLACK_WEBHOOK_URL"' \
  forecast_validation

Set SLACK_WEBHOOK_URL to your Slack incoming webhook URL.

Example usage

Example 1: Temperature forecast validation with Slack alerts

Validate temperature forecast accuracy and send Slack notifications:

# Create the trigger
influxdb3 create trigger \
  --database weather_db \
  --path "gh:influxdata/forecast_error_evaluator/forecast_error_evaluator.py" \
  --trigger-spec "every:15m" \
  --trigger-arguments 'forecast_measurement=temp_forecast,actual_measurement=temp_actual,forecast_field=predicted,actual_field=temperature,error_metric=rmse,error_thresholds=INFO-"0.5":WARN-"1.0":ERROR-"2.0":CRITICAL-"3.0",window=30min,senders=slack,slack_webhook_url="$SLACK_WEBHOOK_URL",min_condition_duration=10min' \
  temp_forecast_check

# Write forecast data
influxdb3 write \
  --database weather_db \
  "temp_forecast,location=station1 predicted=22.5"

# Write actual data  
influxdb3 write \
  --database weather_db \
  "temp_actual,location=station1 temperature=21.8"

# Check logs after trigger runs
influxdb3 query \
  --database YOUR_DATABASE \
  "SELECT * FROM system.processing_engine_logs WHERE trigger_name = 'temp_forecast_check'"

Expected output

• Plugin computes the error between forecast and actual values for every matched timestamp
• Points with an error of 0.5 or more send an INFO notification, 1.0 or more a WARN notification, and so on for each configured level
• A point is only alerted after its error has stayed above the level for 10 minutes (debounce)

Set SLACK_WEBHOOK_URL to your Slack incoming webhook URL.

Notification example:

[WARN] Forecast error alert in temp_actual.temperature: rmse=1.2. Tags: location=station1

The message names the actual measurement and field, because that is where the observed value comes from.

Example 2: Multi-metric validation with multiple channels

Monitor multiple forecast metrics with different notification channels:

# Create trigger with Discord and HTTP notifications
influxdb3 create trigger \
  --database analytics \
  --path "gh:influxdata/forecast_error_evaluator/forecast_error_evaluator.py" \
  --trigger-spec "every:1h" \
  --trigger-arguments 'forecast_measurement=sales_forecast,actual_measurement=sales_actual,forecast_field=predicted_sales,actual_field=sales_amount,error_metric=mae,error_thresholds=WARN-"1000":ERROR-"5000":CRITICAL-"10000",window=6h,senders=discord.http,discord_webhook_url="$DISCORD_WEBHOOK_URL",http_webhook_url="$HTTP_WEBHOOK_URL",notification_text="[$$level] Sales forecast error: $$metric=$$error (threshold exceeded)",rounding_freq=5min' \
  sales_forecast_monitor

Set DISCORD_WEBHOOK_URL and HTTP_WEBHOOK_URL to your webhook URLs.

Example 3: SMS alerts for critical forecast failures

Set up SMS notifications for critical forecast accuracy issues:

# Set environment variables (recommended for sensitive data)
export TWILIO_SID="your_twilio_sid"
export TWILIO_TOKEN="your_twilio_token"

# Create trigger with SMS notifications
influxdb3 create trigger \
  --database production_forecasts \
  --path "gh:influxdata/forecast_error_evaluator/forecast_error_evaluator.py" \
  --trigger-spec "every:5m" \
  --trigger-arguments 'forecast_measurement=demand_forecast,actual_measurement=demand_actual,forecast_field=predicted_demand,actual_field=actual_demand,error_metric=mse,error_thresholds=CRITICAL-"100000",window=15min,senders=sms,twilio_from_number="+1234567890",twilio_to_number="+0987654321",notification_text="CRITICAL: Production demand forecast error exceeded threshold. MSE: $$error",min_condition_duration=2min' \
  critical_forecast_alert

Using TOML Configuration Files

This plugin supports using TOML configuration files for complex configurations.

Important Requirements

To use TOML configuration files, you must set the PLUGIN_DIR environment variable in the InfluxDB 3 Enterprise host environment:

PLUGIN_DIR=~/.plugins influxdb3 serve \
  --node-id node0 \
  --object-store file \
  --data-dir ~/.influxdb3 \
  --plugin-dir ~/.plugins

Example TOML Configuration

# forecast_error_config_scheduler.toml
forecast_measurement = "temperature_forecast"
actual_measurement = "temperature_actual"
forecast_field = "predicted_temp"
actual_field = "temperature"
error_metric = "rmse"
error_thresholds = 'INFO-"0.5":WARN-"1.0":ERROR-"2.0":CRITICAL-"3.0"'
window = "1h"
senders = "slack"
slack_webhook_url = "$SLACK_WEBHOOK_URL"
min_condition_duration = "10min"
rounding_freq = "1min"
notification_text = "[$$level] Forecast validation alert: $$metric=$$error in $$measurement.$$field"

# Authentication (use environment variables instead when possible)
influxdb3_auth_token = "your_token_here"

Set SLACK_WEBHOOK_URL to your Slack incoming webhook URL.

Create trigger using TOML config

influxdb3 create trigger \
  --database weather_db \
  --path "gh:influxdata/forecast_error_evaluator/forecast_error_evaluator.py" \
  --trigger-spec "every:30m" \
  --trigger-arguments config_file_path=forecast_error_config_scheduler.toml \
  forecast_validation_trigger

Code overview

Files

• forecast_error_evaluator.py: The main plugin code containing scheduler handler for forecast validation
• forecast_error_config_scheduler.toml: Example TOML configuration file
• test_forecast_error_evaluator.py: Pytest suite, runs without a live InfluxDB 3 Enterprise server
• requirements.txt: Runtime dependencies (influxdata-plugin-utils>=0.3.0, pandas, requests)
• requirements-dev.txt: Development dependencies (pytest)

Logging

Logs are stored in the trigger’s database in the system.processing_engine_logs table. To view logs:

influxdb3 query --database YOUR_DATABASE "SELECT * FROM system.processing_engine_logs WHERE trigger_name = 'your_trigger_name'"

Log columns:

• event_time: Timestamp of the log event
• trigger_name: Name of the trigger that generated the log
• log_level: Severity level (INFO, WARN, ERROR)
• log_text: Message describing validation results or errors

Main functions

process_scheduled_call(influxdb3_local, call_time, args)

Handles scheduled forecast validation tasks. Queries forecast and actual measurements, computes error metrics, and triggers notifications.

Key operations:

1. Parses configuration from arguments or TOML file
2. Verifies that both measurements exist and resolves the tags they share
3. Queries forecast and actual measurements within the time window
4. Rounds timestamps, collapses duplicate keys and matches the two series
5. Computes the error metric for every matched timestamp
6. Evaluates each threshold level, applies debounce logic and skips already-alerted points
7. Sends notifications via configured channels, up to max_notifications_per_run

compute_error(influxdb3_local, merged, error_metric, task_id)

Adds a per-timestamp error column to the matched frame.

align_frames(influxdb3_local, df_forecast, df_actual, tags, rounding_freq, task_id)

Rounds timestamps, keeps the earliest row per key and inner-joins the two frames on the timestamp and the shared tags.

parse_error_thresholds(influxdb3_local, config, task_id)

Parses the inline <level>-<value> string or the TOML table into a {level: threshold} mapping, skipping unknown levels, non-numeric values and thresholds at or below zero.

Troubleshooting

Common issues

Issue: No overlapping timestamps between forecast and actual data

Solution: Check that both measurements have data in the specified time window and use rounding_freq for alignment:

influxdb3 query --database mydb "SELECT time, field_value FROM forecast_measurement WHERE time >= now() - 1h"
influxdb3 query --database mydb "SELECT time, field_value FROM actual_measurement WHERE time >= now() - 1h"

Issue: Notifications not being sent

Solution: Verify the Notification Sender Plugin is installed and webhook URLs are correct:

# Check if notifier plugin exists
ls ~/.plugins/notifier_plugin.py

# Test webhook URL manually
curl -X POST "your_webhook_url" -d '{"text": "test message"}'

Issue: Error threshold format not recognized

Solution: Use proper threshold format with level prefixes. Note that MAPE and SMAPE thresholds are in percentages:

# For absolute metrics (MSE, MAE, RMSE)
--trigger-arguments 'error_thresholds=INFO-"0.5":WARN-"1.0":ERROR-"2.0":CRITICAL-"3.0"'

# For percentage metrics (MAPE, SMAPE)
--trigger-arguments 'error_thresholds=INFO-"5.0":WARN-"10.0":ERROR-"20.0":CRITICAL-"30.0"'

A Skipping threshold warning names the level that was dropped and why. No valid error thresholds configured means every level was rejected, so nothing was evaluated.

Issue: Trigger fails with “Failed to load configuration”

Solution: The message names the offending parameter. Common causes are a duration without a supported unit (use min, not m), a window of 0s, a port_override outside 1-65535 and an error_metric outside mse, mae, rmse, mape, smape.

Issue: Many rows collapse into one timestamp

Solution: A Collapsed N forecast and M actual rows sharing a rounded timestamp line means rounding_freq is coarser than the sampling interval, so only the earliest point of each slot is compared. Lower rounding_freq to match how far apart the two series are actually written.

Issue: Notifications stop mid-run

Solution: Suppressed N notifications after reaching max_notifications_per_run means the per-run cap was hit. Raise max_notifications_per_run, raise the thresholds, or set min_condition_duration so short spikes are not alerted.

Issue: MAPE/SMAPE calculation errors with zero values

Solution: MAPE cannot be calculated when actual values are zero, and SMAPE cannot be calculated when both forecast and actual are zero. The plugin automatically skips such rows and logs warnings. For datasets with frequent zero values, consider using MAE or RMSE instead.

Issue: Environment variables not loaded

Solution: Set environment variables before starting InfluxDB:

export INFLUXDB3_AUTH_TOKEN="your_token"
export TWILIO_SID="your_sid"
influxdb3 serve --plugin-dir ~/.plugins

Debugging tips

1. Check data availability in both measurements:

influxdb3 query --database mydb \
 "SELECT COUNT(*) FROM forecast_measurement WHERE time >= now() - window"

2. Verify timestamp alignment with rounding frequency:

--trigger-arguments 'rounding_freq=5min'

3. Test with shorter windows for faster debugging:

--trigger-arguments 'window=10min,min_condition_duration=1min'

4. Monitor notification delivery in logs:

influxdb3 query --database YOUR_DATABASE \
 "SELECT * FROM system.processing_engine_logs WHERE log_text LIKE '%notification%'"

Performance considerations

• Data alignment: Use appropriate rounding_freq to balance accuracy and performance
• Window size: Larger windows evaluate more points per run, and every point is checked against every configured level
• Debounce duration: Balance between noise suppression and alert responsiveness
• Notification throttling: Deliveries are sequential with up to three retries each, so max_notifications_per_run bounds how long a run can take
• Trigger interval: An interval shorter than window re-reads the overlap on every run; the last-alert state keeps it from resending, but the data is queried again
• Memory usage: Plugin processes data in pandas DataFrames - consider memory for large datasets

Report an issue

For plugin issues, see the Plugins repository issues page.

Find support for InfluxDB 3 Enterprise

The InfluxDB Discord server is the best place to find support for InfluxDB 3 Core and InfluxDB 3 Enterprise. For other InfluxDB versions, see the Support and feedback options.