Collecting slow queries

Steps to identify slow queries on a cluster to easily find and tune poorly performing queries using the DataStax Enterprise Performance Service.

The node_slow_log table collects information about slow queries on a node and retains query information of long-running CQL statements to help you identify slow queries on a cluster to find and tune poorly performing queries.

You can also use OpsCenter to view, troubleshoot, and trace Slow Queries .

Pending inserts into the node_slow_log might still be processed after the service has been disabled. You can enable and disable the service. After the service is disabled, the logging of queries that take longer than the specified threshold is stopped. However, disabling the logging does not flush the pending write queue, a background thread eventually processes everything.

dse.yaml

The location of the dse.yaml file depends on the type of installation:

Package installations
Installer-Services installations

/etc/dse/dse.yaml

Tarball installations
Installer-No Services installations

installation_location/resources/dse/conf/dse.yaml

Procedure

  1. By default, collection is enabled for statements that are issued when the query exceeds a specified time threshold.
    • To permanently enable collecting information on slow queries, edit the dse.yaml file. Uncomment and define values for cql_slow_log_options as shown in the following listing. Notice the default skip_writing_to_db: true setting.
      cql_slow_log_options:
          enabled: true
          threshold: 200.0
          minimum_samples: 100
          ttl_seconds: 259200
          skip_writing_to_db: true
          num_slowest_queries: 5

      If you keep the default skip_writing_to_db: true setting then the slow query information is stored in memory, not in the node_slow_log table shown later in this section.

      To store the slow query information in the node_slow_log table, set skip_writing_to_db to false in the dse.yaml file.

      If you must store the slow query information in memory, then the information is accessed through the MBean managed Java object named com.datastax.bdp.performance objects.CqlSlowLog using the operation retrieveRecentSlowestCqlQueries.

    • To temporarily change the cqlslowlog setting without changing dse.yaml or restarting DSE, use the dsetool perf subcommands:
      • Disable collecting information on slow queries that exceeded the threshold:
        dsetool perf cqlslowlog disable
      • Keep slow queries in memory only:
        dsetool perf cqlslowlog skip_writing_to_db
      • Write slow queries to the database:
        dsetool perf cqlslowlog write_to_db
    • Set the number of slow queries to keep in memory. For example, 5 queries:
      dsetool perf cqlslowlog set_num_slowest_queries 5
      Retrieve the most recent slow queries:
      dsetool perf cqlslowlog recent_slowest_queries
    • To temporarily change the threshold to collect information on 5% of the slowest queries:
      dsetool perf cqlslowlog 95.0
      After you collect information using this temporarily set threshold, you can run a script to view queries that took longer with this threshold than the previously set threshold. For example:
      $ cqlsh 
      cqlsh> USE dse_perf; 
      cqlsh:dse_perf> SELECT * FROM node_slow_log;
      ...
  2. You can export slow queries using the CQL copy command:
    cqlsh:dse_perf> COPY node_slow_log ( date, commands, duration )
    TO 'slow_queries.csv' WITH HEADER = true;