Audit Log Plugin

Audit Log Plugin

Doris's audit log plugin was developed based on FE's plugin framework. Is an optional plugin. Users can install or uninstall this plugin at runtime.

This plugin can periodically import the FE audit log into the specified system table, so that users can easily view and analyze the audit log through SQL.

Use the audit log plug-in

Starting from Doris version 2.1, the audit log plug-in is directly integrated into the Doris as a built-in plug-in. Users do not need to install additional plug-ins.

After the cluster is started, a system table named audit_log will be created under the __internal_schema database to store audit logs.

1. If you upgrade from an old version, you can continue to use the previous plug-in. You can also uninstall the previous plug-in and use the new built-in plug-in. But note that the built-in plug-in will write the new audit log to a new table instead of the original audit log table.

2. If it is a version before Doris 2.1, please refer to the following Compilation, Configuration and Deployment chapters.

Enable plug-in

The audit log plug-in can be turned on or off at any time through the global variable enable_audit_plugin (the default is off), such as:

set global enable_audit_plugin = true;

After it is enabled, Doris will write the audit log after it is enabled to the audit_log table.

The audit log plugin can be turned off at any time:

set global enable_audit_plugin = false;

After disable, Doris will stop writing to the audit_log table. Audit logs that have been written will not change.

Related configuration

The audit log table is a dynamic partitioned table, partitioned by day, and retains the data of the last 30 days by default.

The following 3 global variables can control some writing behaviors of the audit log table:

• audit_plugin_max_batch_interval_sec: The maximum write interval for the audit log table. Default 60 seconds.
• audit_plugin_max_batch_bytes: The maximum amount of data written in each batch of the audit log table. Default 50MB.
• audit_plugin_max_sql_length: The maximum length of statements recorded in the audit log table. Default 4096.

Can be set via set global xxx=yyy.

Compilation, Configuration and Deployment

FE Configuration

The audit log plug-in framework is enabled by default in Doris and is controlled by the FE configuration plugin_enable

AuditLoader Configuration

1. Download the Audit Loader plugin

   The Audit Loader plug-in is provided by default in the Doris distribution. After downloading the Doris installation package through DOWNLOAD, decompress it and enter the directory, you can find the auditloader.zip file in the extensionsaudit_loader subdirectory.

2. Unzip the installation package

   unzip auditloader.zip

   Unzip and generate the following files:

   • auditloader.jar: plug-in code package.
   • plugin.properties: plugin properties file.
   • plugin.conf: plugin configuration file.

You can place this file on an http download server or copy(or unzip) it to the specified directory of all FEs. Here we use the latter.
The installation of this plugin can be found in INSTALL
After executing install, the AuditLoader directory will be automatically generated.

3. Modify plugin.conf

   The following configurations are available for modification:

   • frontend_host_port: FE node IP address and HTTP port in the format <fe_ip>:<fe_http_port>. The default value is 127.0.0.1:8030.
   • database: Audit log database name.
   • audit_log_table: Audit log table name.
   • slow_log_table: Slow query log table name.
   • enable_slow_log: Whether to enable the slow query log import function. The default value is false. You can set the slow query threshold in the FE configuration item. The parameter is qe_slow_log_ms and the default value is 5s.
   • user: Cluster username. The user must have INSERT permission on the corresponding table.
   • password: Cluster user password.

4. Repackaging the Audit Loader plugin

   zip -r -q -m auditloader.zip auditloader.jar plugin.properties plugin.conf

Create Audit Table

In Doris, you need to create the library and table of the audit log. The table structure is as follows:

If you need to enable the slow query log import function, you need to create an additional slow table doris_slow_log_tbl__, whose table structure is consistent with doris_audit_log_tbl__.

Among them, the dynamic_partition attribute selects the number of days for audit log retention according to your own needs.

create database doris_audit_db__;

create table doris_audit_db__.doris_audit_log_tbl__
(
    query_id varchar(48) comment "Unique query id",
    `time` datetime not null comment "Query start time",
    client_ip varchar(32) comment "Client IP",
    user varchar(64) comment "User name",
    db varchar(96) comment "Database of this query",
    state varchar(8) comment "Query result state. EOF, ERR, OK",
    error_code int comment "Error code of failing query.",
    error_message string comment "Error message of failing query.",
    query_time bigint comment "Query execution time in millisecond",
    scan_bytes bigint comment "Total scan bytes of this query",
    scan_rows bigint comment "Total scan rows of this query",
    return_rows bigint comment "Returned rows of this query",
    stmt_id int comment "An incremental id of statement",
    is_query tinyint comment "Is this statemt a query. 1 or 0",
    frontend_ip varchar(32) comment "Frontend ip of executing this statement",
    cpu_time_ms bigint comment "Total scan cpu time in millisecond of this query",
    sql_hash varchar(48) comment "Hash value for this query",
    sql_digest varchar(48) comment "Sql digest of this query, will be empty if not a slow query",
    peak_memory_bytes bigint comment "Peak memory bytes used on all backends of this query",
    stmt string comment "The original statement, trimed if longer than 2G"
) engine=OLAP
duplicate key(query_id, `time`, client_ip)
partition by range(`time`) ()
distributed by hash(query_id) buckets 1
properties(
    "dynamic_partition.time_unit" = "DAY",
    "dynamic_partition.start" = "-30",
    "dynamic_partition.end" = "3",
    "dynamic_partition.prefix" = "p",
    "dynamic_partition.buckets" = "1",
    "dynamic_partition.enable" = "true",
    "replication_num" = "3"
);

create table doris_audit_db__.doris_slow_log_tbl__
(
    query_id varchar(48) comment "Unique query id",
    `time` datetime not null comment "Query start time",
    client_ip varchar(32) comment "Client IP",
    user varchar(64) comment "User name",
    db varchar(96) comment "Database of this query",
    state varchar(8) comment "Query result state. EOF, ERR, OK",
    error_code int comment "Error code of failing query.",
    error_message string comment "Error message of failing query.",
    query_time bigint comment "Query execution time in millisecond",
    scan_bytes bigint comment "Total scan bytes of this query",
    scan_rows bigint comment "Total scan rows of this query",
    return_rows bigint comment "Returned rows of this query",
    stmt_id int comment "An incremental id of statement",
    is_query tinyint comment "Is this statemt a query. 1 or 0",
    frontend_ip varchar(32) comment "Frontend ip of executing this statement",
    cpu_time_ms bigint comment "Total scan cpu time in millisecond of this query",
    sql_hash varchar(48) comment "Hash value for this query",
    sql_digest varchar(48) comment "Sql digest of a slow query",
    peak_memory_bytes bigint comment "Peak memory bytes used on all backends of this query",
    stmt string comment "The original statement, trimed if longer than 2G "
) engine=OLAP
duplicate key(query_id, `time`, client_ip)
partition by range(`time`) ()
distributed by hash(query_id) buckets 1
properties(
    "dynamic_partition.time_unit" = "DAY",
    "dynamic_partition.start" = "-30",
    "dynamic_partition.end" = "3",
    "dynamic_partition.prefix" = "p",
    "dynamic_partition.buckets" = "1",
    "dynamic_partition.enable" = "true",
    "replication_num" = "3"
);

Notice

In the above table structure: stmt string, this can only be used in 0.15 and later versions, in previous versions, the field type used varchar

Deployment

You can place the packaged auditloader.zip on an http server, or copy auditloader.zip to the same specified directory in all FEs.

Installation

Install the audit loader plugin:

INSTALL PLUGIN FROM [source] [PROPERTIES ("key"="value", ...)]

Detailed command reference: INSTALL-PLUGIN.md

After successful installation, you can see the installed plug-ins through SHOW PLUGINS, and the status is INSTALLED.

After completion, the plugin will continuously insert audit logs into this table at specified intervals.

FAQ

1. There is no data in the audit log table, or no new data is imported after running for a period of time.

   You can check by following these steps:

   • Check whether the partition was created normally

     The audit log table is a dynamic partition table, partitioned by day. By default, partitions for the next 3 days will be created and partitions for the past 30 days will be retained. Only after the partition is created correctly can the audit log be imported normally.

     You can use show dynamic partition tables from __internal_schema to check the scheduling of dynamic partitions and troubleshoot according to the cause of the error. Possible reasons for the error include:

     • The number of BE nodes is less than the required number of replicas: the audit log table has 3 replicas by default, so at least 3 BE nodes are required. Or modify the number of replicas through the alter table statement, such as:

       alter table __internal_schema.audit_log set ("dynamic_partition.replication_num" = "2")

     • No suitable storage medium: You can view the storage_medium attribute through show create table __internal_schema.audit_log. If BE does not have a corresponding storage medium, the partition creation may fail.

     • No suitable resource group: The audit log table defaults to the default resource group. You can use the show backends command to check whether the resource has sufficient node resources.

   • Search for the word AuditLoad in Master FE's fe.log to see if there are related error logs

     The audit log is imported into the table through the internal stream load operation. There may be problems with the import process. These problems will print error logs in fe.log.