config.example.yaml

log:
  level: info
  format: json

telemetry:
  service_name: rss2msg
  # instance_id sets service.instance.id on every signal and, crucially, is added
  # as a dimension/tag on CloudWatch and Graphite metrics so multiple replicas
  # don't collide into one series. Defaults to OTEL_SERVICE_INSTANCE_ID, then the
  # hostname. Leave unset (commented) to use the hostname.
  # instance_id: ""
  traces:   { enabled: true }
  metrics:  { enabled: true }
  logs:     { enabled: false }
  prometheus:
    enabled: false
    listen: ":9090"
  graphite:                       # push metrics to a Carbon (Graphite) endpoint
    enabled: false
    address: "localhost:2003"     # Carbon plaintext TCP endpoint (host:port)
    prefix: "rss2msg"             # metric-path prefix prepended to every metric
    interval: 10s                 # push cadence (0 uses the SDK default)
  sentry:                          # optional error/crash reporting (disabled by default)
    enabled: false
    # dsn: ${SENTRY_DSN}           # required when enabled; falls back to SENTRY_DSN env
    # environment: production      # falls back to SENTRY_ENVIRONMENT
    # release: ${SENTRY_RELEASE}   # falls back to SENTRY_RELEASE
    # server_name: ""              # optional host/instance label
    level: error                   # min log level forwarded (trace|debug|info|warn|error|fatal|panic)
    sample_rate: 1.0               # error-event sampling, [0.0, 1.0]
    traces_sample_rate: 0.0        # performance/transaction sampling, [0.0, 1.0]
    debug: false                   # Sentry SDK debug logging
  posthog:                         # optional PostHog telemetry (disabled by default)
    enabled: false
    # api_key: ${POSTHOG_API_KEY}  # project API key; required when enabled, falls back to POSTHOG_API_KEY
    endpoint: "https://us.i.posthog.com"  # ingestion host; falls back to POSTHOG_ENDPOINT (EU: https://eu.i.posthog.com)
    # distinct_id: ""              # distinct id attached to events; defaults to service name/host
    level: error                   # min log level forwarded; error+ becomes a $exception (Error Tracking) event
    flush_interval: 0s             # batch flush cadence (0 uses the SDK default)
  cloudwatch:                      # optional AWS CloudWatch telemetry (disabled by default)
    enabled: false
    # region: us-east-1            # AWS region; empty uses the default SDK chain
    # endpoint_url: ""             # optional endpoint override (e.g. LocalStack)
    logs:                          # ship log events to CloudWatch Logs
      enabled: false
      # log_group: /rss2msg/app    # required when logs enabled
      # log_stream: ""             # defaults to the hostname
      level: info                  # min log level forwarded (trace|debug|info|warn|error|fatal|panic)
      batch_interval: 5s           # flush cadence (0 uses the default)
      create_group: false          # auto-create the log group/stream if missing
    metrics:                       # push metrics to CloudWatch Metrics (PutMetricData)
      enabled: false
      namespace: rss2msg           # CloudWatch namespace
      interval: 60s                # push cadence (0 uses the OTEL SDK default)

http:
  user_agent: "rss2msg/0.1 (+https://example.com)"
  timeout: 30s
  retry:                 # in-tick retry of a feed fetch on transient errors
    max_attempts: 3      # total tries including the first; 1 disables retry
    base_delay: 500ms    # initial backoff (exponential, full jitter)
    max_delay: 10s       # cap on the backoff delay

retry:
  max_attempts: 3
  base_delay: 500ms
  max_delay: 10s

runtime:
  shutdown_drain_timeout: 30s
  run_once_concurrency: 8
  # Bound a single sink delivery (all retry attempts plus the DLQ handoff) so one
  # wedged sink can't stall a feed's poll loop. 0 (or omitted) disables it.
  deliver_timeout: 0s

heartbeat:
  # Opt-in liveness signal. When enabled, emit one info log line every interval
  # ("heartbeat: service alive") for as long as the service runs, so a
  # quiet-but-healthy daemon still produces an alertable signal in the logs.
  enabled: false
  interval: 1m

health:
  enabled: true
  listen: ":8080"               # probe listener address
  liveness_path: /healthz       # 200 while the process is alive
  readiness_path: /readyz       # 200 when started, not draining, deps reachable
  startup_path: /startupz       # 503 until boot completes, then 200

admin:
  # Opt-in admin HTTP API: JSON introspection (status, feeds, members, health)
  # and safe actions (reconcile, per-feed poll, state prune). Disabled by default.
  enabled: false
  listen: ":8090"               # dedicated listener; required when enabled
  tls:
    enabled: false
    cert_file: ""               # server cert (PEM); required when tls.enabled
    key_file: ""                # server key  (PEM); required when tls.enabled
    client_ca_file: ""          # set => require & verify client certs (mTLS)
  auth:
    enabled: true               # ON by default; set false to deliberately disable token auth
    bearer_tokens:
      - name: ops
        token: "${ADMIN_TOKEN}"
    basic_users: []
    api_keys: []
    api_key_header: X-API-Key
  cors:
    allowed_origins: []         # empty => CORS off. e.g. ["https://ops.example.com"]

coordination:
  driver: memory   # memory | postgres | redis ; default memory
  # postgres:
  #   dsn: ${POSTGRES_DSN}  # optional; falls back to state.postgres.dsn
  #   tls:                  # only takes effect when DSN does not set sslmode=disable
  #     ca_file: /etc/ssl/pg-ca.pem
  #     cert_file: /etc/ssl/pg-client.pem   # mTLS: both cert_file and key_file or neither
  #     key_file:  /etc/ssl/pg-client.key
  #     server_name: pg.internal             # SNI / cert-verify hostname override
  #     insecure_skip_verify: false          # true = disable cert verification (test only)
  # redis:
  #   mode: single           # single (default) | sentinel | cluster
  #   lock_ttl: 30s          # optional, default 30s
  #   renewal_interval: 10s  # optional, default = lock_ttl / 3
  #   # --- single mode (default) ---
  #   url: ${REDIS_URL}      # e.g. redis://localhost:6379/0 or rediss://...
  #   # --- sentinel mode (tested / supported HA path) ---
  #   sentinel:
  #     master_name: mymaster
  #     addrs: [sentinel-a:26379, sentinel-b:26379]
  #     username: redisuser           # data-node credentials (master/replicas)
  #     password: ${REDIS_PASSWORD}   # data-node credentials (master/replicas)
  #     sentinel_username: sentuser           # sentinel-node credentials
  #     sentinel_password: ${SENTINEL_PASSWORD} # sentinel-node credentials
  #     db: 0
  #   # --- cluster mode (best-effort; not covered by CI integration tests) ---
  #   cluster:
  #     addrs: [node-a:6379, node-b:6379]
  #     username: redisuser
  #     password: ${REDIS_PASSWORD}
  #   # --- shared TLS block (applies to all modes) ---
  #   # For single, requires rediss:// URL. For sentinel/cluster, set server_name —
  #   # there is no URL to derive the SNI hostname from.
  #   tls:
  #     ca_file: /etc/ssl/redis-ca.pem
  #     cert_file: /etc/ssl/redis-client.pem   # mTLS: both cert_file and key_file or neither
  #     key_file:  /etc/ssl/redis-client.key
  #     server_name: redis.internal             # SNI / cert-verify hostname override
  #     insecure_skip_verify: false             # true = disable cert verification (test only)
  # assignment: membership + rendezvous-hash partitioning so each feed is polled
  # by exactly one instance. Default off (every instance polls every feed, the
  # per-tick lease decides). Requires a distributed driver to have any effect.
  assignment:
    enabled: false
    strategy: rendezvous      # only "rendezvous" is supported today
    heartbeat_interval: 10s
    member_ttl: 30s           # must exceed heartbeat_interval; dead members expire after this
    rebalance_grace: 5s

state:
  driver: sqlite   # postgres | sqlite | dynamodb | cosmosdb
  # item_ttl: 720h            # retention since an item was last seen; 0/unset = keep forever (default).
                              # Honored by every backend: SQL sweeps; DynamoDB/Cosmos prune natively.
  sqlite:
    path: ./rss2msg.db    # filesystem path; ":memory:" and "?_pragma=..." are accepted
    # cleanup_interval: 1h    # SQL-only: how often this instance sweeps expired items (default 1h when item_ttl > 0)
  # postgres:
  #   dsn: ${POSTGRES_DSN}
  #   cleanup_interval: 1h    # SQL-only: sweep cadence (default 1h when item_ttl > 0)
  #   tls:                  # only takes effect when DSN does not set sslmode=disable
  #     ca_file: /etc/ssl/pg-ca.pem
  #     cert_file: /etc/ssl/pg-client.pem   # mTLS: both cert_file and key_file or neither
  #     key_file:  /etc/ssl/pg-client.key
  #     server_name: pg.internal             # SNI / cert-verify hostname override
  #     insecure_skip_verify: false          # true = disable cert verification (test only)
  # dynamodb:                # shared, distributed-safe store (good with a redis/postgres coordinator)
  #   table: rss2msg-state   # required; table with PK feed_url (S) + SK item_id (S), provisioned out of band
  #   region: us-east-1      # optional; SDK default chain (env / shared config) when empty
  #   endpoint_url:          # optional; LocalStack / DynamoDB Local override
  #   ttl_attribute: expires_at   # required when state.item_ttl is set; names the TTL attribute the table prunes on

sinks:
  # A composite fans every change out to its children. Used as `default`, so
  # feeds with no explicit `sinks:` list publish to both children.
  - name: default
    driver: composite
    composite:
      children: [console, pretty]
  - name: console
    driver: stdout
    stdout:
      target: stdout                              # stdout (default) | stderr
      format: json                                # json (default, NDJSON) | pretty (indented)
  - name: pretty
    driver: stdout
    stdout:
      format: pretty
  # The blocks below are reference examples for the other sink drivers.
  # Uncomment + supply credentials, and add the sink name to a feed's
  # sinks: [...] list to route to it.
  # - name: pg-main
  #   driver: postgres
  #   postgres:
  #     dsn: ${POSTGRES_DSN}
  #     table: feed_changes
  #     tls:                                        # custom CA / mTLS — see docs/how-to/secure-connections-tls.md#sinks
  #       ca_file: /etc/ssl/certs/db-ca.pem
  # - name: kafka-main
  #   driver: kafka
  #   kafka:
  #     brokers: ["kafka-1:9092", "kafka-2:9092"]
  #     topic: feed.changes
  #     acks: all
  #     compression: snappy
  #     tls:                                        # kafka has no URL scheme — set enabled to turn TLS on
  #       enabled: true
  #       ca_file: /etc/ssl/certs/kafka-ca.pem
  #   dead_letter: dlq-main
  # - name: dlq-main
  #   driver: postgres
  #   postgres:
  #     dsn: ${POSTGRES_DSN}
  #     table: feed_changes_dlq
  # - name: sqs-main
  #   driver: sqs
  #   sqs:
  #     queue_url: https://sqs.us-east-1.amazonaws.com/123456789012/feed-changes
  #     region: us-east-1
  #     # endpoint_url: http://localhost:4566           # for LocalStack
  #     # message_group: feed_url                       # FIFO only: feed_url (default) | item_id | sink
  #   dead_letter: dlq-main
  # - name: dynamodb-main
  #   driver: dynamodb                                  # datastore target (change-log table); consume via DynamoDB Streams or polling
  #   dynamodb:
  #     table: feed-changes                             # table must already exist
  #     region: us-east-1
  #     # endpoint_url: http://localhost:4566           # for DynamoDB Local / LocalStack
  #     # partition_key: feed_url                       # default feed_url
  #     # sort_key: item_id                             # default item_id (re-publishing the same item overwrites the row)
  #     # ttl_attribute: expires_at                     # must match the table's configured TTL attribute
  #     # item_ttl: 720h                                # detected_at + item_ttl, written as Unix epoch seconds
  #   dead_letter: dlq-main
  # - name: sns-main
  #   driver: sns
  #   sns:
  #     topic_arn: arn:aws:sns:us-east-1:123456789012:feed-changes
  #     region: us-east-1
  #     # message_group: feed_url                       # FIFO only: feed_url (default) | item_id | sink
  #   dead_letter: dlq-main
  # - name: pubsub-main
  #   driver: gcp_pubsub
  #   gcp_pubsub:
  #     project_id: my-gcp-project
  #     topic_id: feed-changes                          # topic must already exist
  #     # endpoint: localhost:8085                      # for the Pub/Sub emulator
  #     # ordering_key: feed_url                        # ordered delivery: feed_url | item_id | sink
  #   dead_letter: dlq-main
  # - name: dapr-main
  #   driver: dapr_pubsub                               # publish via the local Dapr sidecar; broker is chosen by Dapr component YAML
  #   dapr_pubsub:
  #     pubsub_name: rss-pubsub                         # Dapr pub/sub component name
  #     topic: rss.changes
  #     # address: localhost:50001                      # sidecar gRPC endpoint; default = SDK env / localhost:50001
  #     # content_type: application/json                # payload content type (default application/json)
  #     # metadata:                                     # static per-publish metadata (e.g. broker partition key)
  #     #   partitionKey: feeds
  #   dead_letter: dlq-main
  # - name: rmq-main
  #   driver: amqp091
  #   amqp091:
  #     url: amqp://guest:guest@rabbit-1:5672/      # or amqps:// for TLS
  #     exchange: feed.changes
  #     exchange_type: topic                        # direct (default) | topic | fanout | headers
  #     routing_key: feed.changes
  #     declare: true                               # declare the exchange at startup
  #     durable: true                               # only meaningful when declare=true
  #     tls:                                        # use an amqps:// url above; adds custom CA / mTLS
  #       ca_file: /etc/ssl/certs/rabbit-ca.pem
  #   dead_letter: dlq-main
  # - name: bus-main
  #   driver: amqp10
  #   amqp10:
  #     url: amqps://broker:5671                      # amqp:// or amqps://
  #     target: /queues/changes                       # node/queue/exchange address (required)
  #     username: ${AMQP_USER}                        # optional; else URL userinfo
  #     password: ${AMQP_PASS}
  #     tls:                                          # use an amqps:// url; adds custom CA / mTLS
  #       ca_file: /etc/ssl/certs/broker-ca.pem
  #   dead_letter: dlq-main
  # - name: stream-main
  #   driver: rabbitmq_stream
  #   rabbitmq_stream:
  #     uris: ["rabbitmq-stream://guest:guest@rabbit-1:5552/%2f"]  # or a single url:
  #     stream: feed.changes                          # target stream (required)
  #     declare: true                                 # create the stream if absent
  #     max_age: 168h                                 # optional retention (declare only)
  #     max_length_bytes: 5368709120                  # optional retention (declare only)
  #     # username: ${RMQ_USER}                       # optional; else URI userinfo
  #     # password: ${RMQ_PASS}
  #     tls:                                          # use rabbitmq-stream+tls / IsTLS; custom CA / mTLS
  #       ca_file: /etc/ssl/certs/rabbit-ca.pem
  #   dead_letter: dlq-main
  # - name: nats-main
  #   driver: nats
  #   nats:
  #     url: nats://nats-1:4222                       # comma-separate for a cluster; tls:// for TLS
  #     subject: feed.changes
  #     # token: ${NATS_TOKEN}                         # at most one auth group:
  #     # username: rss                                #   token | username+password | creds_file
  #     # password: ${NATS_PASSWORD}
  #     # creds_file: /etc/nats/rss.creds              # JWT + NKey credentials file
  #     # jetstream: true                              # persisted, server-acked publish (stream must exist)
  #     # tls:                                         # custom CA / mTLS; use a tls:// url above
  #     #   ca_file: /etc/ssl/certs/nats-ca.pem
  #   dead_letter: dlq-main
  # - name: asb-main
  #   driver: azureservicebus
  #   azureservicebus:
  #     connection_string: ${AZURE_SERVICEBUS_CONNECTION_STRING}  # SAS auth; or use namespace:
  #     # namespace: my-bus.servicebus.windows.net                # Azure AD (DefaultAzureCredential)
  #     queue: feed-changes                          # set exactly one of queue / topic
  #     # topic: feed-changes
  #   dead_letter: dlq-main
  # - name: webhook
  #   driver: http
  #   http:
  #     url: https://example.com/hook               # http:// or https://
  #     method: POST                                # POST (default) | PUT
  #     headers:
  #       Authorization: "Bearer ${HOOK_TOKEN}"
  #     timeout: 10s
  #     success_codes: [200, 201, 202, 204]
  #     http3: false                                # dial over HTTP/3 (QUIC); requires an https:// url
  #     tls:                                        # for https:// targets with a private CA / mTLS
  #       ca_file: /etc/ssl/certs/internal-ca.pem

feeds:
  - url: https://example.com/blog/rss.xml
    interval: 5m
    sinks: [pretty]                               # explicit: publish to the "pretty" sink only
  - url: https://other.example/atom.xml
    interval: 15m                                 # implicit: no sinks list → falls back to "default"
    http:
      timeout: 10s
      headers:
        Authorization: "Bearer ${OTHER_FEED_TOKEN}"
      retry:             # per-feed override; omitted fields inherit http.retry above
        max_attempts: 5

# Dynamic feed sources (optional). Ordered list; earlier entries win on URL
# collision. Omit entirely to use only the static feeds: block above. The serve
# daemon reloads feeds from these at runtime (file watch + SIGHUP) without a
# restart. Only "static", "file", "postgres", and "http" types are implemented today.
# feed_sources:
#   - type: file          # JSON array of feed specs; watched, reloads on change
#     name: control-plane
#     path: /etc/rss2msg/feeds.json
#   - type: postgres      # reads the feed list from a Postgres table; polled
#     name: db-feeds
#     interval: 30s        # how often to re-read the table (min 1s)
#     postgres:
#       dsn: "postgres://user:pass@db:5432/rss2msg?sslmode=require"
#       table: feeds        # default "feeds"; columns: url (required), interval, sinks (text[])
#       # query: "SELECT url, interval, sinks FROM feeds WHERE enabled"  # raw SQL override (mutually exclusive with table)
#       # tls:
#       #   ca_file: /etc/rss2msg/pg-ca.pem
#   - type: http          # fetches the feed list from an HTTP endpoint; polled
#     name: control-plane
#     interval: 30s        # how often to re-fetch (min 1s)
#     http:
#       url: https://cp.example/feeds        # required; returns {"feeds":[...]}
#       timeout: 10s
#       headers:
#         Authorization: "Bearer ${CP_TOKEN}"   # or Basic <base64>, or X-API-Key
#       # tls:
#       #   ca_file: /etc/rss2msg/cp-ca.pem
#       #   cert_file: /etc/rss2msg/client.pem   # cert_file + key_file: both or neither
#       #   key_file: /etc/rss2msg/client-key.pem
#   - type: static        # injects the feeds: block at this precedence position