Logging

Everything goes to the journal, with no log files, in a shape a shipper reading it (Vector to Loki, or journalctl -p warning) can filter without parsing text. It is the same contract as frappe-nix's and wordpress-nix's, so one query spans all three.

journalctl APP_SITE=acme -p warning           # everything wrong with one site
journalctl APP_SERVICE=migrate APP_SITE=acme  # its deploy-time migrations
journalctl SYSLOG_IDENTIFIER=nginx_access -o cat | jq 'select(.status >= 500)'

Fields

Every unit the module defines or turns on carries APP_SERVICE and APP_SITE (systemd LogExtraFields, so they are on everything journald attributes to the unit: its output, its syslog messages and systemd's own lines about it).

APP_SITE is dbName. With no pinned dbName (several databases behind dbFilter) it is left out. The roles are python-devops-nix's one written list (lib.journald.roles).

Priorities

Odoo's stderr format is made for a terminal and lands in the journal at priority 6 whatever its level. Instead, each record is written as <N>odoo.http [acme] …: journald reads <N> as the syslog priority (DEBUG 7, INFO 6, WARNING 4, ERROR 3, CRITICAL 2) and strips it, [acme] is the record's database (left out when there is none), Odoo's perf_info follows a request line's message, and the timestamp, PID and colour codes, which the journal records or does not need, are dropped. journald splits on newlines, so every line of a record carries its <N>: a traceback stays at the priority of its error.

odoo-migrate's own progress lines stay at priority 6; Odoo's warnings and errors during a migration get their priorities too.

This is the odoo_journald graft (lib/journald.nix, over python-devops-nix's lib.journald), in every Python environment odoo-nix builds, production included, loaded by a .pth file so odoo-bin run any way gets it. It only acts when stderr really is the journal stream ($JOURNAL_STREAM), so running the same package from a terminal keeps Odoo's own format. logging.journald and $ODOO_NIX_JOURNALD no longer do anything; both are accepted with a warning through 1.x.

Requests

logging.accessLog (on by default, with nginx.enable) logs one JSON object per request to the journal, as SYSLOG_IDENTIFIER=nginx_access: time, site (the Host), method, uri, status, bytes, request_time, upstream_time, remote_addr (the real client, in socket mode too), user_agent and referer. false turns it off.

nginx's error log goes to the journal over syslog (SYSLOG_IDENTIFIER=nginx, services.nginx.logError, a mkDefault) at error and above, each line at its own priority. It is nginx-wide, as logError is.

PostgreSQL

With database.createLocally, each line is prefixed with the database and role (log_line_prefix = "%d %u ", a mkDefault). database.slowQueryMs = 500; logs every statement that runs that long (log_min_duration_statement; 0 logs them all).

Log levels and files

logging.level, logging.handlers (per-logger overrides such as "werkzeug:WARNING") and logging.db/logging.dbLevel (mirror records into a database table) set Odoo's own log_level, log_handler, log_db and log_db_level.

logging.file still works, but Odoo's file and stderr handlers are mutually exclusive: with a file set, the journal only sees what is printed before Odoo's logging starts, so the module warns. Leave it null under systemd, since the journal already timestamps, rotates and ships, or set logging.journald = false to keep the file without the warning. logging.rotate wires services.logrotate to the file, daily with 14 kept; Odoo's WatchedFileHandler picks up the rotated file by itself.