Logging

The operator is designed to log in JSON format directly to standard output, including PostgreSQL logs.

Each log entry has the following fields:

  • level: log level (info, notice, ...)
  • ts: the timestamp (epoch with microseconds)
  • logger: the type of the record (e.g. postgres or pg_controldata)
  • msg: the actual message or the keyword record in case the message is parsed in JSON format
  • record: the actual record (with structure that varies depending on the logger type)
  • logging_podName: the pod where the log was created generated

Warning

Long-term storage and management of logs is outside the operator's purview, and needs to be provided at the level of the Kubernetes installation. Please refer to the Kubernetes Logging Architecture documentation.

Info

The level and ts field names can be renamed via the log-field-level and log-field-timestamp flags of the operator controller, should your log ingestion system require it. All you have to do is edit the Deployment definition of the cloudnative-pg operator.

Operator log

A log level can be specified in the cluster spec with the option logLevel and can be set to any of error, warning, info(default), debug or trace.

At the moment, the log level can only be set when an instance starts and can not be changed at runtime. If the value is changed in the cluster spec after the cluster was started, this will take effect only in the new pods and not the old ones.

PostgreSQL log

Each entry in the PostgreSQL log is a JSON object having the logger key set to postgres and the structure described in the following example:

{
  "level": "info",
  "ts": 1619781249.7188137,
  "logger": "postgres",
  "msg": "record",
  "record": {
    "log_time": "2021-04-30 11:14:09.718 UTC",
    "user_name": "",
    "database_name": "",
    "process_id": "25",
    "connection_from": "",
    "session_id": "608be681.19",
    "session_line_num": "1",
    "command_tag": "",
    "session_start_time": "2021-04-30 11:14:09 UTC",
    "virtual_transaction_id": "",
    "transaction_id": "0",
    "error_severity": "LOG",
    "sql_state_code": "00000",
    "message": "database system was interrupted; last known up at 2021-04-30 11:14:07 UTC",
    "detail": "",
    "hint": "",
    "internal_query": "",
    "internal_query_pos": "",
    "context": "",
    "query": "",
    "query_pos": "",
    "location": "",
    "application_name": "",
    "backend_type": "startup"
  },
  "logging_pod": "cluster-example-1",
}

Internally, the operator relies on the PostgreSQL CSV log format. Please refer to the PostgreSQL documentation for more information about the CSV log format.

PGAudit logs

CloudNativePG has transparent and native support for PGAudit on PostgreSQL clusters.

All you need to do is add the required pgaudit parameters to the postgresql section in the configuration of the cluster.

Important

It is unnecessary to add the PGAudit library to shared_preload_libraries. The library will be added automatically by CloudNativePG based on the presence of pgaudit.* parameters in the postgresql configuration. The operator will detect and manage the addition and removal of the library from shared_preload_libraries.

The operator also takes care of creating and removing the extension from all the available databases in the cluster.

Important

CloudNativePG runs the CREATE EXTENSION and DROP EXTENSION command in all databases in the cluster that accept connections.

Here is an example of a PostgreSQL 13 Cluster deployment which will result in pgaudit being enabled with the requested configuration:

apiVersion: postgresql.cnpg.io/v1
kind: Cluster
metadata:
  name: cluster-example
spec:
  instances: 3
  imageName: ghcr.io/cloudnative-pg/postgresql:13

  postgresql:
    parameters:
      "pgaudit.log": "all, -misc"
      "pgaudit.log_catalog": "off"
      "pgaudit.log_parameter": "on"
      "pgaudit.log_relation": "on"

  storage:
    size: 1Gi

The audit CSV logs entries returned by PGAudit are then parsed and routed to stdout in JSON format, similarly to all the remaining logs:

  • .logger is set to pgaudit
  • .msg is set to record
  • .record contains the whole parsed record as a JSON object, similar to logging_collector logs - except for .record.audit, which contains the PGAudit CSV message formatted as a JSON object

See the example below:

{
  "level": "info",
  "ts": 1627394507.8814096,
  "logger": "pgaudit",
  "msg": "record",
  "record": {
    "log_time": "2021-07-27 14:01:47.881 UTC",
    "user_name": "postgres",
    "database_name": "postgres",
    "process_id": "203",
    "connection_from": "[local]",
    "session_id": "610011cb.cb",
    "session_line_num": "1",
    "command_tag": "SELECT",
    "session_start_time": "2021-07-27 14:01:47 UTC",
    "virtual_transaction_id": "3/336",
    "transaction_id": "0",
    "error_severity": "LOG",
    "sql_state_code": "00000",
    "backend_type": "client backend",
    "audit": {
      "audit_type": "SESSION",
      "statement_id": "1",
      "substatement_id": "1",
      "class": "READ",
      "command": "SELECT FOR KEY SHARE",
      "statement": "SELECT pg_current_wal_lsn()",
      "parameter": "<none>"
    }
  },
  "logging_pod": "cluster-example-1",
}

Please refer to the PGAudit documentation for more details about each field in a record.

Other logs

All logs that are produced by the operator and its instances are in JSON format, with logger set accordingly to the process that produced them. Therefore, all the possible logger values are the following ones:

  • barman-cloud-wal-archive: from barman-cloud-wal-archive directly
  • barman-cloud-wal-restore: from barman-cloud-wal-restore directly
  • initdb: from running initdb
  • pg_basebackup: from running pg_basebackup
  • pg_controldata: from running pg_controldata
  • pg_ctl: from running any pg_ctl subcommand
  • pg_rewind: from running pg_rewind
  • pgaudit: from PGAudit extension
  • postgres: from the postgres instance (having msg different than record)
  • wal-archive: from the wal-archive subcommand of the instance manager
  • wal-restore: from the wal-restore subcommand of the instance manager

Except for postgres that has the aforementioned structures, all other possible values just have msg set to the escaped message that is logged.