Enable Log Parsing

This how-to turns on pgwatch's CSVLOG parsing so that PostgreSQL server log severities are counted per database and per instance.

Prerequisites

The target PostgreSQL instance is configured to write logs in CSVLOG format (log_destination = 'csvlog' in postgresql.conf).
For local parsing mode: pgwatch runs on the same host as the database server and the OS user running pgwatch has read access to the log directory, plus the pg_read_all_settings role.
For remote parsing mode: the monitoring user has the pg_monitor role and EXECUTE on pg_read_file(text, bigint, bigint).

Activate the server_log_event_counts metric on the source. The simplest way is to pick a preset that includes it (the built-in full preset does) or to add it directly to the source's metrics list:
```
- name: my-postgres
  kind: postgres
  conn_str: postgresql://pgwatch@db1/mydb
  custom_metrics:
    server_log_event_counts: 60
  is_enabled: true
```
Choose the parsing mode. pgwatch picks automatically:
- Local mode — used when the monitoring connection comes through a Unix socket or when the data_directory setting on the server matches what pg_control_system() reports.
- Remote mode — used otherwise; reads log blocks through pg_read_file().
Local mode is preferred where possible — it avoids round-tripping log bytes through the SQL connection.
Tune the fetch interval. Each call reads up to 10 MB from a single log file. With high log volume, raise the interval to keep up; with low volume, lower it for finer-grained event counts.
Verify pgwatch is recording counts:
```
curl -H "Token: $TOKEN" http://localhost:8080/source/my-postgres/metric
```
Look for rows in the server_log_event_counts metric family grouped by severity.

The feature only stores counts, not log lines or usernames — safe for security-sensitive deployments.
For binary PG upgrades via pg_upgrade, helper functions that wrap log access may need to be re-installed afterwards.