Lotus emits :telemetry events for query execution, cache operations, and schema introspection. These events integrate with monitoring tools like Phoenix LiveDashboard, AppSignal, Datadog, and others.

Events

Query Execution

EventMeasurementsMetadata
[:lotus, :query, :start]system_timesource, statement, context
[:lotus, :query, :stop]duration, row_countsource, statement, context, row_count, result
[:lotus, :query, :exception]durationsource, statement, context, kind, reason, stacktrace

Duration is measured in native time units. Use System.convert_time_unit/3 to convert to milliseconds or microseconds. On :stop, row_count appears in both the measurements and the metadata — it is the same value.

The source field is the data source name ("main", "warehouse"), not a repo module. The statement field is a %Lotus.Query.Statement{}: read statement.body for the adapter-native payload (SQL text for Ecto-backed sources, a JSON object or AST for others) and statement.params for the bound values. Pre-v1 :repo, :sql and :params metadata keys are gone.

The context field carries whatever value the caller passed as the :context option to Lotus.run_statement/3 or Lotus.run_query/2. It defaults to nil when not provided. Typical uses include request IDs, controller names, or OpenTelemetry span contexts for trace correlation. The :scope option is not in the metadata — it is caller identity used for cache keys and visibility, not instrumentation.

The events bracket statement execution — sanitization, preflight, :before_execute and the query itself — so :start fires before any of those run. Any failure among them — a denied table, a halted :before_execute plug, a driver error — ends the run at :exception, not :stop.

Because Lotus turns those failures into {:error, reason} rather than letting them raise, the :exception metadata is uniform: kind is always :error, reason is the {:error, reason} tuple the caller receives, and stacktrace is []. Match on reason rather than expecting an exception struct.

Query telemetry brackets what a result cache hit skips — authorizing the statement and executing it — so a query served from cache emits no [:lotus, :query, *] events at all; [:lotus, :cache, :hit] is the event to count for those. On a miss the span includes the cache write, which stores only the result and the relations. The :before_query and :after_query middleware run outside that phase, on every call: a plug that halts there yields {:error, reason} to the caller without emitting query events, because no statement ran. For the same reason :stop carries the result as the source returned it — an :after_query plug that redacts rows has not run yet, so a handler that logs metadata.result logs the unredacted rows.

Cache Operations

EventMeasurementsMetadata
[:lotus, :cache, :hit]countkey
[:lotus, :cache, :miss]countkey
[:lotus, :cache, :put]countkey, ttl_ms

Schema Introspection

EventMeasurementsMetadata
[:lotus, :schema, :introspection, :start]system_timeoperation, repo
[:lotus, :schema, :introspection, :stop]durationoperation, repo, result

The operation field is one of: :list_schemas, :list_tables, :describe_table, :get_table_stats, or :list_relations.

The result field is :ok or :error.

Note: these two events kept the metadata key :repo, unlike the query events which renamed it to :source. The value is the same thing in both — the data source name ("main"), not an Ecto repo module.

Content Changes

EventMeasurementsMetadata
[:lotus, :content, :change, :start]system_timeop, resource, record, context
[:lotus, :content, :change, :stop]durationop, resource, record, changes, context
[:lotus, :content, :change, :exception]durationop, resource, record, context, kind, reason, stacktrace

A content change is one create, update or delete of a query, visualization, dashboard, dashboard card, dashboard filter or filter mapping, including enabling or disabling public sharing and moving a card in a reorder. The events bracket every change, refused or not, and always carry the :context option the caller passed, or nil.

op is :create, :update, :delete, :enable_sharing or :disable_sharing. On :start, record is the struct as stored (nil on a create); on :stop it is the struct as written, and the metadata is the :after_content_change middleware payload. changes holds each changed field with its written value; it is %{} on a delete and on an update that wrote nothing, which still emits :stop.

:exception is emitted when a :before_content_change plug refused the change, the changeset was invalid, or the write raised, threw or exited. reason is the error the caller receives — {:halted, reason} for a refusal, the changeset for a failed validation — and stacktrace is present only when the call raised, threw or exited. A reorder emits one span per card it moves, and every card shares the outcome of the call.

Setup

Attach handlers in your application's start/2 callback:

# lib/my_app/application.ex
def start(_type, _args) do
  :telemetry.attach_many(
    "lotus-telemetry",
    [
      [:lotus, :query, :stop],
      [:lotus, :query, :exception],
      [:lotus, :cache, :hit],
      [:lotus, :cache, :miss]
    ],
    &MyApp.LotusInstrumentation.handle_event/4,
    nil
  )

  children = [
    # ...
  ]

  Supervisor.start_link(children, strategy: :one_for_one)
end

Example Handler

defmodule MyApp.LotusInstrumentation do
  require Logger

  def handle_event([:lotus, :query, :stop], measurements, metadata, _config) do
    duration_ms = System.convert_time_unit(measurements.duration, :native, :millisecond)

    Logger.info(
      "Lotus query completed",
      duration_ms: duration_ms,
      row_count: measurements.row_count,
      source: metadata.source
    )
  end

  def handle_event([:lotus, :query, :exception], measurements, metadata, _config) do
    duration_ms = System.convert_time_unit(measurements.duration, :native, :millisecond)

    Logger.error(
      "Lotus query failed",
      duration_ms: duration_ms,
      reason: inspect(metadata.reason),
      source: metadata.source,
      statement: inspect(metadata.statement.body)
    )
  end

  def handle_event([:lotus, :cache, :hit], _measurements, metadata, _config) do
    Logger.debug("Lotus cache hit", key: metadata.key)
  end

  def handle_event([:lotus, :cache, :miss], _measurements, metadata, _config) do
    Logger.debug("Lotus cache miss", key: metadata.key)
  end
end

Phoenix LiveDashboard Integration

If you use Phoenix LiveDashboard, you can add Lotus metrics to your telemetry supervisor:

# lib/my_app_web/telemetry.ex
defp metrics do
  [
    # Lotus query metrics
    summary("lotus.query.stop.duration",
      unit: {:native, :millisecond},
      description: "Lotus query execution time"
    ),
    counter("lotus.query.stop.duration",
      description: "Total Lotus queries executed"
    ),
    counter("lotus.query.exception.duration",
      description: "Total Lotus query failures"
    ),

    # Cache metrics
    counter("lotus.cache.hit.count",
      description: "Lotus cache hits"
    ),
    counter("lotus.cache.miss.count",
      description: "Lotus cache misses"
    )
  ]
end

Event Reference

For the complete list of events, measurements, and metadata fields, see Lotus.Telemetry.