MySQL 8.3 Release Notes
Each error log sink (writer) component has a characteristic output format it uses to write messages to its destination, but other factors may influence the content of the messages:
The information available to the log sink. If a log filter component executed prior to execution of the sink component removes a log event field, that field is not available for writing. For information about log filtering, see Section 7.4.2.4, “Types of Error Log Filtering”.
The information relevant to the log sink. Not every sink writes all fields available in error events.
System variables may affect log sinks. See System Variables That Affect Error Log Format.
For names and descriptions of the fields in error events, see Section 7.4.2.3, “Error Event Fields”. For all log sinks, the thread ID included in error log messages is that of the thread within mysqld responsible for writing the message. This ID indicates which part of the server produced the message, and is consistent with general query log and slow query log messages, which include the connection thread ID.
The internal log sink produces traditional error log output. For example:
2020-08-06T14:25:02.835618Z 0 [Note] [MY-012487] [InnoDB] DDL log recovery : begin
2020-08-06T14:25:02.936146Z 0 [Warning] [MY-010068] [Server] CA certificate /var/mysql/sslinfo/cacert.pem is self signed.
2020-08-06T14:25:02.963127Z 0 [Note] [MY-010253] [Server] IPv6 is available.
2020-08-06T14:25:03.109022Z 5 [Note] [MY-010051] [Server] Event Scheduler: scheduler thread started with id 5Traditional-format messages have these fields:
time thread [label] [err_code] [subsystem] msg
The [ and ] square
bracket characters are literal characters in the message
format. They do not indicate that fields are optional.
The label value corresponds to the string
form of the prio error event priority
field.
The [err_code] and
[subsystem] fields were added in MySQL 8.0,
and thus are missing from logs generated by older servers. Log
parsers can treat these fields as parts of the message text
that is present only for logs written by servers recent enough
to include them. Parsers must treat the
err_code part of
[err_code] indicators as a string value,
not a number, because values such as
MY-012487 and MY-010051
contain nonnumeric characters.
The JSON-format log sink produces messages as JSON objects that contain key-value pairs. For example:
{
"prio": 3,
"err_code": 10051,
"source_line": 561,
"source_file": "event_scheduler.cc",
"function": "run",
"msg": "Event Scheduler: scheduler thread started with id 5",
"time": "2020-08-06T14:25:03.109022Z",
"ts": 1596724012005,
"thread": 5,
"err_symbol": "ER_SCHEDULER_STARTED",
"SQL_state": "HY000",
"subsystem": "Server",
"buffered": 1596723903109022,
"label": "Note"
}The message shown is reformatted for readability. Events written to the error log appear one message per line.
The ts (timestamp) key is unique to the
JSON-format log sink. The value is an integer indicating
milliseconds since the epoch ('1970-01-01
00:00:00' UTC).
The ts and buffered
values are Unix timestamp values and can be converted using
FROM_UNIXTIME() and an
appropriate divisor:
mysql> SET time_zone = '+00:00';
mysql> SELECT FROM_UNIXTIME(1596724012005/1000.0);
+-------------------------------------+
| FROM_UNIXTIME(1596724012005/1000.0) |
+-------------------------------------+
| 2020-08-06 14:26:52.0050 |
+-------------------------------------+
mysql> SELECT FROM_UNIXTIME(1596723903109022/1000000.0);
+-------------------------------------------+
| FROM_UNIXTIME(1596723903109022/1000000.0) |
+-------------------------------------------+
| 2020-08-06 14:25:03.1090 |
+-------------------------------------------+The system log sink produces output that conforms to the system log format used on the local platform.
The server generates some error log messages before startup
options have been processed, and thus before it knows error
log settings such as the
log_error_verbosity and
log_timestamps system
variable values, and before it knows which log components are
to be used. The server handles error log messages that are
generated early in the startup process as follows:
The server buffers log events (rather than formatted log messages), which enables it to apply configuration settings to those events retroactively, after the settings are known, with the result that flushed messages use the configured settings, not the defaults. Also, messages are flushed to all configured sinks, not just the default sink.
If a fatal error occurs before log configuration is known and the server must exit, the server formats buffered messages using the logging defaults so they are not lost. If no fatal error occurs but startup is excessively slow prior to processing startup options, the server periodically formats and flushes buffered messages using the logging defaults so as not to appear unresponsive. Although this behavior uses the defaults, it is preferable to losing messages when exceptional conditions occur.
The log_timestamps system
variable controls the time zone of timestamps in messages
written to the error log (as well as to general query log and
slow query log files). The server applies
log_timestamps to error
events before they reach any log sink; it thus affects error
message output from all sinks.
Permitted log_timestamps
values are UTC (the default) and
SYSTEM (the local system time zone).
Timestamps are written using ISO 8601 / RFC 3339 format:
YYYY-MM-DDThh:mm:ss.uuuuuuZ signifying Zulu time
(UTC) or ±hh:mm (an offset that
indicates the local system time zone adjustment relative to
UTC). For example:
2020-08-07T15:02:00.832521Z (UTC)
2020-08-07T10:02:00.832521-05:00 (SYSTEM)