Logging

On a services.frappe host everything goes to the journal and nothing writes a log file. That covers Frappe (FRAPPE_STREAM_LOGGING=1, and a caller asking for a file is overridden), bench (bench/logs/bench.log is no longer written) and nginx (no /var/log/nginx/access.log).

The exceptions are Frappe features that write a file as their product rather than as logging, and only when used: the request monitor (monitor in site_config, logs/monitor.json.log) and the setup wizard's logs/setup-wizard.log.

A host that ships the journal to a log store filters by the fields below and by PRIORITY, so both are part of the module's contract. It is the same contract odoo-nix and wordpress-nix follow.

Reading the journal

With the unified runtime, the site's unit is frappe-<SITE>:

journalctl -u frappe-site1.example.com -f
journalctl -u frappe-init-site1.example.com

Because every unit carries journald fields, you can also filter across units:

journalctl APP_SERVICE=migrate APP_SITE=site1.example.com
journalctl APP_SERVICE=runtime -p warning
journalctl SYSLOG_IDENTIFIER=nginx_access -o cat

Fields

Every unit the module defines carries LogExtraFields, which journald attaches to everything the unit's processes log, whether through stdout, stderr or syslog.

The shared units serve every site on the host, so they carry no APP_SITE. Each access-log entry names its site in its JSON instead.

Priorities

The bench's virtualenv carries frappe_journald (lib/journald, grafted like frappe_unixsock into both virtualenvs). When stderr is the journal, it reformats Frappe's loggers and bench's own as <N>{module} [{site}] {message}. It checks JOURNAL_STREAM against stderr's device and inode, so a terminal, the dev shell, the OCI images and a captured subprocess keep stock output.

journald reads the <N> as the line's syslog priority (debug 7, info 6, warning 4, error 3, critical 2) and stamps the time itself, so the timestamp and level name are gone from the message. The site is added because Frappe's own format has none. Every line of a multi-line record carries the prefix, so a traceback stays at error from its first line to its last instead of dropping to info after the first.

frappe-runtime formats its root logger the same way, and the migrate unit's failure lines are at error, so journalctl -u frappe-migrate-<SITE> -p err finds a failed migration. -p filters work as you would expect on Python log lines.

Log level

services.frappe.logging.level is the threshold for frappe.logger() and for bench's log, passed as FRAPPE_LOG_LEVEL. It accepts debug, info, warning and error, and defaults to warning. Frappe's own production default is error, which silently drops every frappe.logger().warning().

An explicit set_log_level() in code still wins. The runtime's lifecycle lines stay at info, and gunicorn, Node, MariaDB and Redis keep their own levels. The units also set PYTHONUNBUFFERED=1, so print() output is neither held back nor lost in a crash.

The nginx access log

services.frappe.logging.accessLog (default true) sends nginx's access log to the journal over /dev/log, one JSON object per request, under SYSLOG_IDENTIFIER=nginx_access. nginx writes each object on a single line. It is shown formatted here:

{
  "time": "2026-09-29T12:00:00+00:00",
  "site": "erp.example.com",
  "method": "GET",
  "uri": "/app",
  "status": 200,
  "bytes": 5120,
  "request_time": 0.042,
  "upstream_time": "0.041",
  "remote_addr": "203.0.113.7",
  "user_agent": "…",
  "referer": "…"
}

upstream_time is a string. It is - for a request with no upstream, such as /assets/, and a comma-separated list when nginx retried one.

nginx's error log goes to syslog too (services.nginx.logError, overridable), so its [error] and [crit] lines keep their severity.