As of the time of this publication, audit logging features are made available and licensed only as part of the MemSQL Advanced Security Option. Before using or implementing this functionality, please consult with your enterprise’s licensing administrator to confirm that your enterprise has purchased the necessary Advanced Security Option license from MemSQL.
Audit logs are written to the directory specified by the auditlogsdir variable in a node’s memsql.cnf file. Each node in a cluster will create its own files in the destination directory using the following naming convention:
auditlog_<hostname>-<port>_<year>-<month>-<day>_<hour>-<minute>-<second>.log
For example, consider the following example log file name:
auditlog_myVmbox-3306_2016-08-30_06-38-46.log
If a log file’s rotation size is very small, an auto-incrementing integer may be appended to the end of the log file. For example:
auditlog_myVmbox-3306_2016-08-30_06-38-46_1.log
auditlog_myVmbox-3306_2016-08-30_06-38-46_2.log
auditlog_myVmbox-3306_2016-08-30_06-38-46_3.log
Data within the log file is written in comma-separated values (CSV) format. Log files are updated immediately as queries are executed; they are not cached locally and then eventually written to the destination directory.
Due to the security requirements for audit logging, every log entry must be flushed to disk before the query executes. This requirement may result in significant performance limitations for a cluster.
There are three types of schemas for log file entries:
- Schema for the first line in a log file
- Schema for database results when queries are executed in
ALL-RESULTSmode - Generic schema for all other entries
Schema for First Line in a Log File
The first line in a log file has a unique schema that differs from the generic case. Consider the following example:
0,2016-08-30 20:05:47.694,INFO: Log rotated. Format version 1
The first line above demonstrates the following schema:
<log-entry-id>,<timestamp>,INFO: <log-file-event-origin>. <log-format-version>
This line’s schema is described in the table below:
| Example Value | Definition |
|---|---|
log-entry-id |
This value is always 0 for the first line in a logfile. |
timestamp |
This value is identical to a normal [timestamp]. |
log-file-event-origin |
Describes the reason or event that caused the creation of the log file. There are currently two possible values: * Log rotated.—Indicates that a new file was required because rotation conditions were met. When a log is rotated, either the auditlog_rotation_size or the auditlog_rotation_time engine variables’ conditions were met.* Log opened on MemSQL startup.— Indicates that a new file was created because the node was either started or restarted. |
Schema for ALL-RESULTS Database Responses
If your node is configured to use the ALL-RESULTS logging mode, the database’s responses are logged uniquely from other entries. Consider the following example result entry for the SHOW VARIABLES LIKE '%master_aggregator%' query:
237,R,master aggregator
The log entry above demonstrates the following schema:
<parent-query-log-entry-id>,R,<data>
Query results share the same entry ID as the query itself. In the example above, the log entry for the query also has an <entry-id> value of 237. The comma-separated R denotes that the log entry only contains query result data. Lastly, the returned data is included as the last value in the entry.
General Log Entry Schema
All entries in a log file adhere to a predefined schema. To help understand the log file schema, consider the following example entry for the statement UPDATE t SET a = 0 WHERE a = 1:
237,2016-08-25 11:29:09,PDT,localhost:3306,agg,1,99999,root,db,temp_1_14_16,9959767090931131257,UPDATE t SET a = 0 WHERE a = 1
The log entry above demonstrates the following schema:
<log-entry-id>,<timestamp>,<time-zone>,<hostname>:<port>,<node-type>,<aggregator-id>,<thread-id>,<username>,<database-name>,<correlation-id>,<query-hash>,<query>
Therefore, each of the 12 values can be mapped in the following way:
| Example Value | Definition |
|---|---|
237 |
log-entry-id |
2016-08-25 11:29:09.875 |
timestamp |
PDT |
time-zone |
localhost:3306 |
hostname:port |
agg |
node-type |
1 |
aggregator-id |
99999 |
thread-id |
root |
username |
db |
database-name |
temp_1_14_16 |
correlation-id |
9959767090931131257 |
query-hash |
UPDATE t SET a = 0 WHERE a = 1 |
query |
log-entry-id
The first value is log-entry-id, which represents the entry’s unique identifier within the scope of a single log file. Every new file always starts with an entry that has an ID of 0, incrementing by one afterwards. A log-entry-id is a 64-bit unsigned integer value.
The entry ID for any given operation is unique to the node, regardless of how a query is processed across the cluster. For example, imagine a three node cluster with a master aggregator and two leaves. When the master aggregator receives a distributed query and executes it on both leaf nodes, each leaf’s entry ID for the operation may be completely different.
If your database is configured to use the ALL-RESULTS logging mode, the results will appear as additional log entries with the same <log-entry-id> as the parent. For more information, see Schema for ALL-RESULTS Database Responses.
timestamp
The second value is timestamp, which specifies the node’s local system time when an operation was executed. The timestamp is printed in %Y-%m-%d %H:%M:%S.%f format.
time-zone
The third value is time-zone, which specifies the local time zone when an operation when executed. The time zone is printed as an initialism, such as GMT (Greenwich Mean Time).
hostname:port
The fourth value is hostname:port, which specifies the node’s hostname and port. The hostname is the same the database’s hostname global variable. Together, these values represent the identifier for a node.
node-type
The fifth value is node-type, which specifies whether the node is an aggregator (agg) or a leaf (leaf).
aggregator-id
The sixth value is aggregator-id, which specifies the ID of the aggregator that issued any given query. An aggregator-id is a 64-bit unsigned integer value.
An aggregator-id may contain an invalid value of 18446744073709551615 if the aggregator has not yet been given an ID.
In a cluster with multiple aggregators, the ID of the master aggregator never changes, but the ID of a child aggregator may change. For example, if a child aggregator temporarily loses network connectivity and then reconnects, it may acquire a new aggregator ID. The child aggregators that remained connected might also have their aggregator IDs altered during the event.
When correlating a master aggregator’s query with its resulting downstream leaf node queries, you can use the combination of an entry’s correlation-id and aggregator-id values as a unique identifier. For example, if a query is executed against a cluster with one master aggregator and one leaf node, the query will be logged in the master aggregator’s log file and the leaf node’s log file. To quickly correlate entries for the same query across large log files, use the combination of the correlation-id and aggregator-id as a key.
thread-id
The seventh value is thread-id, which specifies the ID of the node’s thread that is executing the query. A thread-id is a 32-bit integer value.
username
The eight value is username, which specifies the username of the authenticated user who is executing the query.
database-name
The ninth value is database-name, which specifies the name of the database that is being queried.
This value may appear as [unknown] in a log entry under a few conditions:
- If the query is not related to a specific database (e.g.
SHOW VARIABLES). - If the user does not specify
USE <database-name>before executing their query. However, in this case the query does contain the database, such asSELECT * from <db-name>.t.
correlation-id
The tenth value is correlation-id, which specifies an identifier for the current query. Multiple successive log entries can reference the same correlation-id, especially when a query requires data from more than one leaf partition.
When correlating a master aggregator’s query with its resulting downstream leaf node queries, you can use the combination of an entry’s correlation-id and aggregator-id values as a unique identifier. For example, if a query is executed against a cluster with one master aggregator and one leaf node, the query will be logged in the master aggregator’s log file and the leaf node’s log file. To quickly correlate entries for the same query across large log files, use the combination of the correlation-id and aggregator-id as a key.
Under some circumstances, a correlation-id value may not appear in the log entry. This value is usually missing when a DDL operation is executed.
query-hash
The 11th value is query-hash, which specifies a unique hash of either the query signature or the query text. The query-hash is a 64-bit unsigned integer value.
The full query text will only be hashed if the node’s logging level is set to ALL-QUERIES-PLAINTEXT or ALL-QUERIES-PLAINTEXT-INCLUDING-PARSE-FAILS. Otherwise, just the query signature will be hashed.
query
The 12th value is the query, which specifies either the query signature or the query text.
The full query text will only be written to the log if the node’s logging level is set to ALL-QUERIES-PLAINTEXT or ALL-QUERIES-PLAINTEXT-INCLUDING-PARSE-FAILS. Otherwise, just query signature will be hashed.
