Skip to main content

Customize Monitoring

Configure Monitoring using SQL

Notice

Using SingleStoreDB Toolbox to configure cluster monitoring is the easier, preferred, and default method.

Enable the Exporter Process

The memsql-exporter process (or simply the exporter) collects data about a running cluster. The user that starts the exporter (other than the SingleStoreDB root user) must have the following permissions at a minimum:

GRANT CLUSTER on *.* to <user>
GRANT SHOW METADATA on *.* to <user>
GRANT SELECT on *.* to <user>
GRANT PROCESS on *.* to <user>

Start the Exporter Process

Perform the following steps on the Source cluster Master Aggregator.

Note that the exporter process is restarted when the exporter_port variable is set to any non-zero value, including the current value. The exporter_port variable can be set by running SET GLOBAL exporter_port = <port>;.

HTTP Connections

You may start the exporter in the SingleStoreDB engine by passing a user, port, and password to the exporter. The following example shows the exporter being started with the SingleStoreDB root user:

SET GLOBAL exporter_user = root;
SET GLOBAL exporter_password = '<secure-password>';
SET GLOBAL exporter_port= 9104;

HTTPS Connections

  1. Copy the server certificate and key to the Master Aggregator host of the Source cluster. This certificate will be used by the exporter process and must be readable by the user under which the nodes are running on the host (typically the memsql user).

  2. Use the following SQL statements to start the exporter in the SingleStoreDB engine.

    SET GLOBAL exporter_user = root;
    SET GLOBAL exporter_password = '<secure-password>';
    SET GLOBAL exporter_use_https= true;
    SET GLOBAL exporter_ssl_cert = '/path/to/server-cert.pem';
    SET GLOBAL exporter_ssl_key = '/path/to/server-key.pem';
    SET GLOBAL exporter_ssl_key_passphrase= '<passphrase>';

    Note: The exporter_ssl_key_passphrase variable should only be included if the server key has a passphrase.

How to Stop the Exporter Process

You may stop the exporter at any time by running the following command on the Source cluster Master Aggregator.

Use an engine variable to stop the exporter process by setting the port to 0.

SET GLOBAL exporter_port = 0;

Configure the metrics Database

The metrics database can either reside in the Source cluster, in a dedicated Metrics cluster, or in a remote SingleStoreDB cluster that the Metrics cluster can access.

Note that the user that creates the metrics database may differ from the one that started the exporter. This user must have the ability to create tables, databases, and pipelines. At minimum, it requires the following permissions:

metrics Database DDL

You may run the following SQL scripts to configure the database.

Please note the following:

  • The database that is created to store monitoring data is named metrics by default. You must edit these SQL statements manually to change the database name.

  • The database user is root by default. When creating the metrics database with a database user other than root, the SUPER privilege must be granted to this user. For example, for a dbmon user:

    GRANT SUPER ON *.* TO 'dbmon'@'%';

  • These commands must be run on the Master Aggregator of the Metrics cluster.

Create the metrics database and associated tables by copying, pasting, and executing each set of SQL statements below. This can be done via the SQL Editor in Studio, or your favored SQL client.

You may also download the latest metrics database DDL SQL file from SingleStore.

create database if not exists metrics;

CREATE TABLE IF NOT EXISTS metrics.config (
    `hostport`      TINYBLOB NOT NULL,
    `retention_period_minutes` INT NOT NULL,
    `purge_frequency_minutes` INT NOT NULL,
    `purge_limit` INT NOT NULL,
    PRIMARY KEY(`hostport`)
);

CREATE TABLE IF NOT EXISTS metrics.`metrics` (
    `labels`                                           JSON,
    `name`                                             TINYBLOB NOT NULL,
    `memsql_tags`                                      JSON COLLATE utf8_bin,
    `cluster` AS `memsql_tags`::$cluster               PERSISTED TINYBLOB,
    `host` AS `labels`::$host                          PERSISTED TINYBLOB,
    `port` as `labels`::$port                          PERSISTED SMALLINT(6),
    `role` as `labels`::$role                          PERSISTED TINYBLOB,
    `extractor` as substring_index(`name`, '_', 1)     PERSISTED TINYBLOB,
    `subsystem` as substring_index(substring_index(`name`, '_', 2), '_', -1) PERSISTED TINYBLOB,
    `job` AS `memsql_tags`::$push_job                  PERSISTED TINYBLOB,
    `value`                                            DOUBLE NOT NULL,
    `intval` AS cast(floor(`value`) AS signed integer) PERSISTED BIGINT(20),
    `time_sec`                                         BIGINT(20) NOT NULL,
    KEY `name` (`cluster`,`extractor`,`subsystem`,`host`,`role`,`name`,`time_sec`) USING CLUSTERED COLUMNSTORE,
    SHARD KEY()
);

CREATE TABLE IF NOT EXISTS metrics.act_samples (
    `cluster` varchar(512) CHARACTER SET utf8 COLLATE utf8_bin NOT NULL,
    `memsql_tags` JSON NOT NULL,
    `ts` timestamp NOT NULL DEFAULT CURRENT_TIMESTAMP,
    `ACTIVITY_TYPE` varchar(512) CHARACTER SET utf8 COLLATE utf8_bin NOT NULL,
    `ACTIVITY_NAME` varchar(512) CHARACTER SET utf8 COLLATE utf8_bin NOT NULL,
    `AGGREGATOR_ACTIVITY_NAME` varchar(512) CHARACTER SET utf8 COLLATE utf8_bin DEFAULT NULL,
    `DATABASE_NAME` varchar(512) CHARACTER SET utf8 COLLATE utf8_bin DEFAULT NULL,
    `NODE_ID` smallint(6) DEFAULT NULL,
    `PARTITION_ID` smallint(6) DEFAULT NULL,
    `CPU_TIME_MS` bigint(4) DEFAULT NULL,
    `CPU_WAIT_TIME_MS` bigint(4) DEFAULT NULL,
    `ELAPSED_TIME_MS` bigint(4) DEFAULT NULL,
    `LOCK_TIME_MS` bigint(4)  DEFAULT NULL,
    `NETWORK_TIME_MS` bigint(4)  DEFAULT NULL,
    `DISK_TIME_MS` bigint(4) DEFAULT NULL,
    `DISK_B` bigint(4) DEFAULT NULL,
    `DISK_LOGICAL_READ_B` bigint(4) DEFAULT NULL,
    `DISK_LOGICAL_WRITE_B` bigint(4) DEFAULT NULL,
    `DISK_PHYSICAL_READ_B` bigint(4) DEFAULT NULL,
    `DISK_PHYSICAL_WRITE_B` bigint(4) DEFAULT NULL,
    `LOCK_ROW_TIME_MS` bigint(4) DEFAULT NULL,
    `LOG_FLUSH_TIME_MS` bigint(4) DEFAULT NULL,
    `LOG_BUFFER_TIME_MS` bigint(4) DEFAULT NULL,
    `LOG_BUFFER_WRITE_B` bigint(4) DEFAULT NULL,
    `NETWORK_B` bigint(4) DEFAULT NULL,
    `NETWORK_LOGICAL_RECV_B` bigint(4) DEFAULT NULL,
    `NETWORK_LOGICAL_SEND_B` bigint(4) DEFAULT NULL,
    `MEMORY_BS` bigint(4) DEFAULT NULL,
    `MEMORY_MAJOR_FAULTS` bigint(4) DEFAULT NULL,
    `RUN_COUNT` bigint(4)  NULL DEFAULT '0',
    `SUCCESS_COUNT` bigint(4)  NULL DEFAULT '0',
    `FAILURE_COUNT` bigint(4)  NULL DEFAULT '0',
    KEY (`cluster`, `ACTIVITY_TYPE`,`ACTIVITY_NAME`,`AGGREGATOR_ACTIVITY_NAME`,`DATABASE_NAME`,`NODE_ID`,`PARTITION_ID`,ts) USING CLUSTERED COLUMNSTORE,
    SHARD KEY()
);

create table IF NOT EXISTS metrics.mv_queries
(
    `cluster` varchar(512) NOT NULL,
    `memsql_tags` JSON NOT NULL,
    `ts` timestamp NOT NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,
    `ACTIVITY_NAME` varchar(512),
    `QUERY_TEXT` longtext,
    `PLAN_WARNINGS` varchar(8192),
    KEY (`cluster`, `ACTIVITY_NAME`) USING CLUSTERED COLUMNSTORE,
    SHARD KEY()
);

create table IF NOT EXISTS metrics.cluster_info
(
    `cluster` varchar(512) NOT NULL,
    `memsql_tags` JSON NOT NULL,
    `ts` timestamp NOT NULL DEFAULT CURRENT_TIMESTAMP,
    `status` JSON NOT NULL,
    KEY (cluster, ts) USING CLUSTERED COLUMNSTORE,
    SHARD KEY()
);

create table IF NOT EXISTS metrics.db_status
(
    `memsql_tags` JSON NOT NULL,
    `keys` JSON NOT NULL,
    `values` JSON NOT NULL,
    `ts` timestamp NOT NULL DEFAULT CURRENT_TIMESTAMP,
    `cluster` AS `memsql_tags`::$cluster          PERSISTED TINYBLOB,
    database_name AS `keys`::$database_name       PERSISTED TINYBLOB,
    `role` as `keys`::$role                       PERSISTED tinyblob,
    num_partitions as `values`::$num_partitions   PERSISTED SMALLINT(6),
    `online` as `values`::$online                 PERSISTED SMALLINT(6),
    `offline` as `values`::$offline               PERSISTED SMALLINT(6),
    replicating as `values`::$replicating         PERSISTED SMALLINT(6),
    recovering as `values`::$recovering           PERSISTED SMALLINT(6),
    pending as `values`::$pending                 PERSISTED SMALLINT(6),
    transition as `values`::$transition           PERSISTED SMALLINT(6),
    unrecoverable as `values`::$unrecoverable     PERSISTED SMALLINT(6),
    ref_db_state as `values`::$ref_db_state       PERSISTED TINYBLOB,
    summary as `values`::$summary                 PERSISTED TINYBLOB,
    KEY (cluster, database_name, ts) USING CLUSTERED COLUMNSTORE,
    SHARD KEY()
);

create table if not exists metrics.mv_events
(
    `memsql_tags` JSON NOT NULL,
    `keys` JSON NOT NULL,
    `values` JSON NOT NULL,
    ts TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
    event_ts TIMESTAMP,
    `cluster` AS `memsql_tags`::$cluster          PERSISTED TINYBLOB,
    `event_type` AS `keys`::$event_type           PERSISTED TINYBLOB,
    `origin_node_id` AS `keys`::$origin_node_id   PERSISTED SMALLINT,
    `ip_addr` AS `values`::$ip_addr               PERSISTED TINYBLOB,
    `port` AS `values`::$port                     PERSISTED SMALLINT,
    `type` AS `values`::$type                     PERSISTED TINYBLOB,
    `severity` as `keys`::$severity               PERSISTED TINYBLOB,
    `details` as `values`::$details               PERSISTED TINYBLOB,
    KEY (cluster, origin_node_id, event_ts) USING CLUSTERED COLUMNSTORE,
    SHARD KEY()
);

create table if not exists metrics.mv_nodes
(
    `memsql_tags` JSON NOT NULL,
    `keys` JSON NOT NULL,
    `values` JSON NOT NULL,
    `ts` TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
    `cluster`                AS `memsql_tags`::$cluster            PERSISTED TINYBLOB,
    `node_id`                AS `keys`::$node_id                   PERSISTED SMALLINT,
    `ip_addr`                AS `keys`::$ip_addr                   PERSISTED TINYBLOB,
    `port`                   AS `keys`::$port                      PERSISTED SMALLINT,
    `role`                   AS `keys`::$role                      PERSISTED TINYBLOB,
    `external_host`          AS `keys`::$external_host             PERSISTED TINYBLOB,
    `external_port`          AS `keys`::$external_port             PERSISTED SMALLINT,
    `state`                  AS `values`::$state                   PERSISTED TINYBLOB,
    `availability_group`     AS `values`::$availability_group      PERSISTED SMALLINT,
    `num_cpus`               AS `values`::$num_cpus                PERSISTED BIGINT,
    `max_memory_mb`          AS `values`::$max_memory_mb           PERSISTED BIGINT,
    `max_table_memory_mb`    AS `values`::$max_table_memory_mb     PERSISTED BIGINT,
    `memory_used_mb`         AS `values`::$memory_used_mb          PERSISTED BIGINT,
    `table_memory_used_mb`   AS `values`::$table_memory_used_mb    PERSISTED BIGINT,
    `total_data_disk_mb`     AS `values`::$total_data_disk_mb      PERSISTED BIGINT,
    `available_data_disk_mb` AS `values`::$available_data_disk_mb  PERSISTED BIGINT,
    `uptime`                 AS `values`::$uptime                  PERSISTED BIGINT,
    `version`                AS `values`::$version                 PERSISTED TINYBLOB,
    KEY(`cluster`, `node_id`, `ts`) USING CLUSTERED COLUMNSTORE,
    SHARD KEY()
);

create table if not exists metrics.purge_log
(
    `ts`           TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
    `hostport`         TINYBLOB NOT NULL,
    `rows_deleted` BIGINT NOT NULL,
    `status`       TINYBLOB NOT NULL,
    KEY(`ts`, `hostport`),
    SHARD KEY()
);

delimiter //
create or replace procedure metrics.purge_metrics(current_ts TIMESTAMP, current_hostport TINYBLOB,
    retention_period INT) AS
declare rc int = 1;
declare earliest_ts INT = UNIX_TIMESTAMP(TIMESTAMPADD(MINUTE, -retention_period, current_ts));
begin
    delete from metrics where ifnull(`memsql_tags`::$host, "") = current_hostport and time_sec < earliest_ts;
    rc = row_count();
    update purge_log set rows_deleted = rows_deleted + rc where `hostport` = current_hostport and `ts` = current_ts;
end //
delimiter ;

delimiter //
create or replace procedure metrics.purge_blobs(current_ts TIMESTAMP, current_hostport TINYBLOB,
    retention_period INT) AS
declare rc int = 1;
declare earliest_ts TIMESTAMP = TIMESTAMPADD(MINUTE, -retention_period, current_ts);
begin
    delete from act_samples where ifnull(`memsql_tags`::$host, "") = current_hostport and ts < earliest_ts;
    rc = row_count();
    delete from mv_events where ifnull(`memsql_tags`::$host, "") = current_hostport and ts < earliest_ts;
    rc += row_count();
    delete from mv_queries where ifnull(`memsql_tags`::$host, "") = current_hostport and ts < earliest_ts;
    rc += row_count();
    delete from cluster_info where ifnull(`memsql_tags`::$host, "") = current_hostport and ts < earliest_ts;
    rc += row_count();
    delete from db_status where ifnull(`memsql_tags`::$host, "") = current_hostport and ts < earliest_ts;
    rc += row_count();
    delete from mv_nodes where ifnull(`memsql_tags`::$host, "") = current_hostport and ts < earliest_ts;
    rc += row_count();
    update purge_log set rows_deleted = rows_deleted + rc where `hostport` = current_hostport and `ts` = current_ts;
end //
delimiter ;

-- This procedure purges monitoring data that is older than the retention period set in the config table.
-- To prevent overrunning this we use purge_log table in order to track when the last purge
-- was completed. New purge process won't start unless at least some time passed since previous purge,
-- controlled by `purge_frequency_minutes` in the config table.
delimiter //
CREATE or replace PROCEDURE metrics.purge_data(current_hostport TINYBLOB) AS
declare
    need_purge_qry QUERY(need_purge BOOLEAN) =
        select (TIMESTAMPDIFF(
            MINUTE,
            (select ifnull(max(ts), from_unixtime(0)) from purge_log
              where hostport = current_hostport),
            CURRENT_TIMESTAMP()
        )>=(select ifnull(ANY_VALUE(purge_frequency_minutes), 60)
              from config
             where hostport = current_hostport) );
    retention_period_qry QUERY(retention_period INT) =
        select ifnull(ANY_VALUE(retention_period_minutes), 10080)
        from config
        where hostport = current_hostport;
    retention_period INT = SCALAR(retention_period_qry);
    current_ts TIMESTAMP;
begin
    IF SCALAR(need_purge_qry) THEN
        BEGIN
            select current_timestamp() into current_ts;
            insert into purge_log values (current_ts, current_hostport, 0, "RUNNING");
            call purge_metrics(current_ts, current_hostport, retention_period);
            call purge_blobs(current_ts, current_hostport, retention_period);
            update purge_log set status = "FINISHED" where `ts` = current_ts;
            EXCEPTION
                WHEN OTHERS THEN
                    update purge_log set status = "FAILED" where `ts` = current_ts and hostport = current_hostport;
            delete from purge_log where TIMESTAMPDIFF(MINUTE, ts, CURRENT_TIMESTAMP()) > 525600;
        END;
    END IF;
end //
delimiter ;

delimiter //
CREATE or replace PROCEDURE metrics.load_metrics (metrics_pipe query(`labels` JSON, `name` TINYBLOB, `memsql_tags` JSON COLLATE utf8_bin,`value` DOUBLE NOT NULL,`time_sec` BIGINT(20) NOT NULL)) AS
DECLARE hostport TINYBLOB;
BEGIN
    INSERT INTO metrics (
        `labels`,
        `name`,
        `memsql_tags`,
        `value`,
        `time_sec`
    ) SELECT
        `labels`,
        `name`,
        `memsql_tags`,
        `value`,
        `time_sec`
        from metrics_pipe;
    SELECT IFNULL(ANY_VALUE(`memsql_tags`::$host), "") INTO hostport FROM metrics_pipe;
    CALL purge_data(hostport);
END //
delimiter ;


delimiter //
create or replace procedure metrics.load_blobs (blobs_pipe query(type text, `keys` JSON, `vals` JSON, time_sec bigint, `memsql_tags` JSON NULL)) AS
DECLARE hostport TINYBLOB;
begin
    insert into act_samples (
        `cluster`,
        `memsql_tags`,
        `ts`,
        `ACTIVITY_TYPE`,
        `ACTIVITY_NAME`,
        `AGGREGATOR_ACTIVITY_NAME`,
        `NODE_ID`,
        `PARTITION_ID`,
        `DATABASE_NAME`,
        `CPU_TIME_MS`,
        `CPU_WAIT_TIME_MS`,
        `ELAPSED_TIME_MS`,
        `LOCK_TIME_MS` ,
        `NETWORK_TIME_MS` ,
        `DISK_TIME_MS`,
        `DISK_B`,
        `DISK_LOGICAL_READ_B`,
        `DISK_LOGICAL_WRITE_B`,
        `DISK_PHYSICAL_READ_B`,
        `DISK_PHYSICAL_WRITE_B`,
        `LOCK_ROW_TIME_MS`,
        `LOG_FLUSH_TIME_MS`,
        `LOG_BUFFER_TIME_MS`,
        `LOG_BUFFER_WRITE_B`,
        `NETWORK_B`,
        `NETWORK_LOGICAL_RECV_B`,
        `NETWORK_LOGICAL_SEND_B`,
        `MEMORY_BS`,
        `MEMORY_MAJOR_FAULTS`,
        `RUN_COUNT`,
        `SUCCESS_COUNT`,
        `FAILURE_COUNT`
    ) select
    `memsql_tags`::$cluster,
    `memsql_tags`,
    FROM_UNIXTIME(time_sec),
    `keys`::$activity_type,
    `keys`::$activity_name,
    `keys`::$aggregator_activity_name,
    `keys`::node_id,
    `keys`::partition_id,
    `keys`::$db_name,
    `vals`::cpu_time_ms,
    `vals`::cpu_wait_time_ms,
    `vals`::elapsed_time_ms,
    `vals`::lock_time_ms,
    `vals`::network_time_ms,
    `vals`::disk_time_ms,
    `vals`::disk_b,
    `vals`::disk_logical_read_b,
    `vals`::disk_logical_write_b,
    `vals`::disk_physical_read_b,
    `vals`::disk_physical_write_b,
    `vals`::lock_row_time_ms,
    `vals`::log_flush_time_ms,
    `vals`::log_buffer_time_ms,
    `vals`::log_buffer_write_b,
    `vals`::network_b,
    `vals`::network_logical_recv_b,
    `vals`::network_logical_send_b,
    `vals`::memory_bs,
    `vals`::memory_major_faults,
    `vals`::run_count,
    `vals`::success_count,
    `vals`::failure_count
    from blobs_pipe
    where type = 'activity';

    insert into mv_events (
        `memsql_tags`,
        `keys`,
        `values`,
        `ts`,
        `event_ts`
    )
    select
    `memsql_tags`,
    `keys`,
    `vals`,
    FROM_UNIXTIME(time_sec),
    FROM_UNIXTIME(`keys`::$event_time)
    from blobs_pipe b
    left join (SELECT distinct cluster,
        origin_node_id,
        event_ts from mv_events) k
    ON b.`memsql_tags`::$cluster = k.cluster
    and b.`keys`::$origin_node_id = k.origin_node_id
    and b.`keys`::$event_time = k.event_ts
    where type = 'event'
    and k.cluster is null
    and k.origin_node_id is null;

    insert into mv_queries (
        `cluster`,
        `ACTIVITY_NAME`,
        `memsql_tags`,
        `ts`,
        `QUERY_TEXT`,
        `PLAN_WARNINGS`
    )
    select
    `memsql_tags`::$cluster,
    `keys`::$activity_name,
    -- memsql_tags and time_sec are the same for the whole batch
    ANY_VALUE(`memsql_tags`),
    FROM_UNIXTIME(ANY_VALUE(time_sec)),
    ANY_VALUE(`vals`::$query_text),
    ANY_VALUE(`vals`::$plan_warnings)
    from blobs_pipe b
    left join (SELECT distinct cluster, ACTIVITY_NAME from mv_queries) k
    ON b.`memsql_tags`::$cluster = k.cluster and b.`keys`::$activity_name = k.ACTIVITY_NAME
    where type = 'query'
    and k.cluster is null and k.ACTIVITY_NAME is null
    group by 1, 2;

    insert into cluster_info (
        `cluster`,
        `memsql_tags`,
        `ts`,
        `status`
    ) select
    `memsql_tags`::$cluster,
    `memsql_tags`,
    FROM_UNIXTIME(time_sec),
    `vals`
    from blobs_pipe
    where type = 'cluster';

    insert into db_status (
        `memsql_tags`,
        `keys`,
        `values`,
        `ts`
    ) select
    `memsql_tags`, `keys`, `vals`, FROM_UNIXTIME(time_sec)
    from blobs_pipe
    where type = 'db_state';

    insert into mv_nodes (
        `memsql_tags`,
        `keys`,
        `values`,
        `ts`
    ) select
    `memsql_tags`, `keys`, `vals`, FROM_UNIXTIME(time_sec)
    from blobs_pipe
    where type = 'node';

    SELECT IFNULL(ANY_VALUE(`memsql_tags`::$host), "") INTO hostport FROM blobs_pipe;
    CALL purge_data(hostport);
end //
delimiter ;

-- Grafana Stuff

delimiter //
CREATE OR REPLACE FUNCTION metrics.`trim_metric`(input varchar(512) CHARACTER SET utf8 COLLATE utf8_general_ci NULL, prefix varchar(255) CHARACTER SET utf8 COLLATE utf8_general_ci NULL) RETURNS varchar(512) CHARACTER SET utf8 COLLATE utf8_general_ci NULL AS begin
  return
  CASE
  WHEN input is not null and prefix is not null and locate(prefix, input, 1) > 0 THEN
    UPPER(substr(input, locate(prefix, input, 1) + length(prefix), length(input) - length(prefix)))
  ELSE input end;
end //
delimiter ;

delimiter //
CREATE OR REPLACE FUNCTION metrics.`trim_host`(input varchar(512) CHARACTER SET utf8 COLLATE utf8_general_ci NULL, c int default 1) RETURNS varchar(512) CHARACTER SET utf8 COLLATE utf8_general_ci NULL AS begin
 return
 CASE
 WHEN input RLIKE '([[:digit:]]{1,3}\\.){3}[[:digit:]]{1,3}' THEN
  input
 ELSE substring_index(input, '.', c) end;
end //
delimiter ;

-- helper table for integers 1..100
drop table if exists metrics.ints;

create reference table metrics.ints
(
    f int key
);

insert into metrics.ints values
(1), (2), (3), (4), (5), (6), (7), (8), (9), (10),
(11), (12), (13), (14), (15), (16), (17), (18), (19), (20),
(21), (22), (23), (24), (25), (26), (27), (28), (29), (30),
(31), (32), (33), (34), (35), (36), (37), (38), (39), (40),
(41), (42), (43), (44), (45), (46), (47), (48), (49), (50),
(51), (52), (53), (54), (55), (56), (57), (58), (59), (60),
(61), (62), (63), (64), (65), (66), (67), (68), (69), (70),
(71), (72), (73), (74), (75), (76), (77), (78), (79), (80),
(81), (82), (83), (84), (85), (86), (87), (88), (89), (90),
(91), (92), (93), (94), (95), (96), (97), (98), (99), (100);

metrics Database DDL Variables

The following variables may be set via the metrics DDL file instead of using command-line options when starting cluster monitoring.

  • Monitoring database name: Default is metrics.

  • By default, the metrics.config table is empty. The sdb-admin start-monitoring command adds a new row to this table automatically when a new Source cluster is first monitored.

    In the case of multi-cluster monitoring, an existing monitoring configuration can be overridden for a specific Source cluster by manually adding a row to this table and specifying:

    • The <exporter-host><exporter-port> of the Source cluster, in hostport form

    • retention_period_minutes value, in minutes

    • A purge_frequency_minutes value, in minutes

    • A purge_limit value, an integer

  • retention_period_minutes = 10080 minutes (default), which equals 7 days: Period of time to retain monitoring data for a specific Source cluster.

  • purge_frequency_minutes = 60 (default), minutes: Time between purges of expired monitoring data for a specific Source cluster.

  • purge_limit = 10000 (default), integer: Maximum number of entries deleted during a monitoring data purge for a specific Source cluster.

  • Purge log retention period = 525600 minutes (default), which equals 365 days: Period of time to retain purge log entries.

  • batch_interval = 15000 (default), milliseconds: Time interval between data collections.

  • high_cardinality_metrics = false (default). When set to true, additional monitoring data is collected, and a higher load is placed on the Source cluster(s) during collection.

  • monitoring_version: No default. Can be specified as either Major.Minor version (such as 7.3) or Major.Minor.Patch version (such as 7.3.25)

  • sample_queries = true (default): Controls whether the queries run on the Source cluster(s) also include the entire query text (from information_schema.mv_queries). As query sampling is enabled by default, specifying a value of false, or using the --disable-sample-queries option with the sdb-admin start-monitoring command, will not collect query text.

Refer to sdb-admin start-monitoring for more information.

Pipelines DDL

Run the following in the SQL Editor in Studio, or your favored SQL client.

Note: You must edit exporter-host and port in the following SQL statements to align with where your exporter process resides.

  • The exporter-host is typically the host of your Source cluster’s Master Aggregator that’s running the exporter, and must include http://.

  • The default port for the endpoint is 9104.

  • The IP address provided in the exporter-host argument must be resolvable by all hosts in the cluster. If the cluster you are monitoring has hosts that contain leaf nodes (as is common in a multi-host vs. single-host deployment), do not use localhost. Instead, provide the explicit IP address or hostname that the other hosts in the cluster can resolve.

HTTP Connections
  1. Start the metrics pipeline.

    CREATE OR REPLACE PIPELINE metrics.metrics AS
    LOAD DATA prometheus_exporter 
    "http://<exporter-host>:<exporter-port>/cluster-metrics"
    CONFIG '{"is_memsql_internal":true, "high_cardinality_metrics":false, 
    "monitoring_version":"<SingleStoreDB version of the 'metrics' cluster>"}' 
    batch_interval 15000
    INTO PROCEDURE `load_metrics` FORMAT JSON;
    
    START PIPELINE IF NOT RUNNING metrics.metrics;
    
  2. Start the blobs pipeline.

    CREATE OR REPLACE PIPELINE metrics.blobs AS
    LOAD DATA prometheus_exporter 
    "http://<exporter-host>:<exporter-port>/samples"
    CONFIG '{"is_memsql_internal":true,
    "download_type":"samples", "sample_queries":true,
    "monitoring_version":"<SingleStoreDB version of the metrics cluster>",
    "ssl_ca":"path to ca cert" OR "ssl_capath":"path to ca dir"}'
    batch_interval 15000
    INTO PROCEDURE `load_blobs` FORMAT JSON;
    
    START PIPELINE IF NOT RUNNING metrics.blobs;
    
HTTPS Connections
  1. Copy the CA certificate to the same file path on each host of the Metrics cluster.

    Alternatively, a directory containing multiple CA certificates can be provided, and this path must be the same on all hosts. This directory will be used by the monitoring pipelines and must be readable by the user under which the nodes are running on the host (typically the memsql user).

  2. Start the metrics pipeline.

    CREATE OR REPLACE PIPELINE metrics.metrics AS
    LOAD DATA prometheus_exporter 
    "https://<exporter-host>:<exporter-port>/cluster-metrics"
    CONFIG '{"is_memsql_internal":true, "high_cardinality_metrics":false, 
    "monitoring_version":"<SingleStoreDB version of the 'metrics' cluster>"}'
    batch_interval 15000
    INTO PROCEDURE `load_metrics` FORMAT JSON;
    
    START PIPELINE IF NOT RUNNING metrics.metrics;
    
  3. Start the blobs pipeline.

    CREATE OR REPLACE PIPELINE metrics.blobs AS
    LOAD DATA prometheus_exporter 
    "https://<exporter-host>:<exporter-port>/samples"
    CONFIG '{"is_memsql_internal":true,
    "download_type":"samples", "sample_queries":true,
    "monitoring_version":"<SingleStoreDB version of the metrics cluster>",
    "ssl_ca":"path to ca cert" OR "ssl_capath":"path to ca dir"}'
    batch_interval 15000
    INTO PROCEDURE `load_blobs` FORMAT JSON;
    
    START PIPELINE IF NOT RUNNING metrics.blobs;
    

How to Stop the Monitoring Process

You may stop the monitoring process at any time by manually stopping the pipelines created earlier.

STOP PIPELINE metrics;
STOP PIPELINE blobs;