Application configuration

This page is about configuring the server process — distinct from authoring a pipeline config. Application config sets the broker, database, storage backend, and LLM providers the server uses; pipeline configs describe the work it runs.

How config loads

Settings come from environment variables (and a .env file in development), namespaced FACTFLOW_<GROUP>_<FIELD>. Each provider group maps to a settings object:

Prefix	Configures
FACTFLOW_APP_*	App name, environment, debug, concurrent-execution ceiling
FACTFLOW_DATABASE_*	PostgreSQL host/port/database
FACTFLOW_QUEUE_*	Queue provider selection, pool size, concurrency
FACTFLOW_ARTEMIS_* / FACTFLOW_PULSAR_*	Provider-specific broker settings
FACTFLOW_STORAGE_*	Storage URI (file://, minio://, s3://), credentials, file-size limits
FACTFLOW_LLM_*	LLM provider profiles and credentials
FACTFLOW_LOGGING_*	Structured logging config (LEVEL, LOG_TO_FILE, LOG_TO_CONSOLE)
FACTFLOW_LOG_LEVEL	Root-logger level set at process startup

Running several pipelines at once

FACTFLOW_APP_MAX_CONCURRENT_EXECUTIONS (default 10) caps how many pipeline executions the orchestrator will admit. Past it, starting a run fails outright rather than queueing:

Failed to start pipeline: Max 10 concurrent executions reached (10 active)

This is the only enforced ceiling on concurrency. Nothing limits the number of broker subscriptions, even though each route holds one — a run of 14 concurrent executions was measured holding 59 subscriptions across 42 routes with the broker healthy. So if concurrent runs are being refused, this is the value to raise; there is no subscription cap to tune.

Raise it with the machine in mind rather than by reflex: every route in every execution holds a broker connection, so the practical limit is what the broker and the host's file descriptors will carry.

Pipeline-config discovery and validation

Two overrides control how the server finds and validates shipped pipeline YAML at startup:

• FACTFLOW_CONFIG_PIPELINES_DIR — override the directory scanned for pipeline configs.
• FACTFLOW_CONFIG_VALIDATOR_STRICT — defaults to strict (abort startup on an invalid shipped YAML); set 0 to downgrade to warn-only.

FACTFLOW_CONFIG_VALIDATOR_STRICT=0 \
FACTFLOW_CONFIG_PIPELINES_DIR=/path/to/pipelines \
  uv run python -m factflow_server serve --embedded

Provider configuration

LLM and embedding provider profiles (model, residency, credentials) are covered in LLM & embedding providers. The broker and storage backends are selected by the FACTFLOW_QUEUE_* and storage settings above.