Collecting slow queries
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 DSE 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.
By default, collection is enabled for statements that are issued when the query exceeds a specified time threshold.
Permanently configure slow query logs
In the dse.yaml file, uncomment and set values for cql_slow_log_options.
The location of the dse.yaml file depends on the type of installation:
-
Package installations and Installer-Services installations:
/etc/dse/dse.yaml -
Tarball installations and Installer-No Services installations:
<installation_location>/resources/dse/conf/dse.yaml
cql_slow_log_options:
enabled: true
threshold: 200.0
minimum_samples: 100
ttl_seconds: 259200
skip_writing_to_db: true
num_slowest_queries: 5
|
By default, To store the slow query information in the |
Temporarily configure slow query logs
To temporarily change the cqlslowlog settings without changing dse.yaml or restarting DSE, use the dsetool perf subcommands:
-
Disable slow query logs:
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 -
Change the percentile threshold for slow query logs.
The following example collects the slowest 5 percent of queries:
dsetool perf cqlslowlog 95.0
Retrieve slow query logs
After modifying the slow query configuration and allowing queries to run, get the slow query logs.
- Read from node_slow_log table
-
To read from the
node_slow_logtable, logs must be written to the database. Indse.yaml, theskip_writing_to_dbsetting must befalse, or you must rundsetool perf cqlslowlogwith thewrite_to_dboption.-
Retrieve all slow query logs from the
node_slow_logtable usingcqlsh:SELECT * FROM dse_perf.node_slow_log; -
Retrieve recent slow queries only:
dsetool perf cqlslowlog recent_slowest_queries
-
- Read from memory
-
Slow query logs are stored in memory if
skip_writing_to_dbistrueindse.yamlor you rundsetool perf cqlslowlogwith theskip_writing_to_dboption.To read slow query logs from memory, access the MBean managed Java object named
com.datastax.bdp.performance objects.CqlSlowLogusing the operationretrieveRecentSlowestCqlQueries. For more information about accessing MBeans, see Monitor DataStax Enterprise (DSE) clusters.
Export slow query logs
You can export slow queries using the CQL shell COPY TO command:
COPY dse_perf.node_slow_log ( date, commands, duration )
TO 'slow_queries.csv' WITH HEADER = true;