http-logger

http-logger

Builds a JSON access-log entry for each request/response and ships accumulated batches to an arbitrary HTTP endpoint. Entries are queued with a non-blocking, fire-and-forget push; a background task POSTs them once a batch fills, goes idle, or ages out. Place this node in the response pipeline, after the upstream node, so the final status and body size are captured.

Configuration

Key	Type	Default	Description
uri	string	— (required)	HTTP endpoint batches are POSTed to.
method	string	POST	HTTP method for the callout.
headers	object	{}	Extra request headers applied to every batch.
auth_header	string	—	Convenience for a single Authorization header value.
concat_method	string	json	json sends a JSON array (application/json); new_line sends \n-separated JSON objects (text/plain).
ssl_verify	bool	false	Verify TLS certificates for https endpoints.
timeout	integer (s)	3	Whole-call deadline per flush.
log_format	object	—	Custom flat entry of name -> "$var template". When absent, the default structured entry is used.
include_req_body	bool	false	Add the request body to the default entry.
include_resp_body	bool	false	Add the response body to the default entry.
batch_max_size	integer	1000	Flush when the buffer reaches this many entries.
inactive_timeout	integer (s)	5	Flush after this idle period.
buffer_duration	integer (s)	60	Flush when the oldest buffered entry is this old.
max_retry_count	integer	0	Retries after a failed flush before dropping the batch.
retry_delay	integer (s)	1	Delay between retries.
max_pending_entries	integer	10000	Queue capacity; entries are dropped with a warning when full.

- id: access-log
  type: http-logger
  config:
    uri: http://log-collector:3000/logs
    method: POST
    headers:
      X-Token: secret
    concat_method: json
    batch_max_size: 1000
    inactive_timeout: 5

Behavior

The node is a pure passthrough: it never modifies the context and never fails, so only its success port is ever taken. push is fire-and-forget and never blocks the request path — when the queue is full (a slow or retrying endpoint) entries are dropped with a tracing::warn!, which is the operator's signal to raise max_pending_entries or fix the downstream. Delivery, batching, timing, and retries all happen on a background task. On config reload the old sink is drained and flushed before a new one is spawned.