Report query performance issues

Use these guidelines to work with InfluxData engineers to troubleshoot and resolve query performance issues.

1. Send InfluxData output artifacts
2. Document your test process
3. Document your environment
4. Document your data schema
5. Establish query performance degradation conditions
6. Reduce query noise
7. Establish baseline single-query performance
8. Run queries at multiple load scales
9. Gather debug information
   1. Kubernetes-specific information
   2. Clustered-specific information
   3. Query analysis
      1. EXPLAIN
      2. EXPLAIN VERBOSE
      3. EXPLAIN ANALYZE
10. Gather system information
    • Collect table information
    • Collect compaction information for the table
    • Collect partition information for multiple tables

Send InfluxData output artifacts

As you follow these guidelines, package all produced output artifacts in the following form:

Outputs:

• test-artifact-name.tar.gz

Send InfluxData engineers all produced artifacts for analysis.

Document your test process

Currently, InfluxDB Clustered doesn’t provide a standardized performance test suite that you can run in your cluster. Please document your test process so that InfluxData engineers can replicate it–include the following:

• The steps you take when performance testing.
• Timestamps of your test runs, to correlate tests with logs.

Document your environment

Provide as much detail about your environment as your organization allows, including the following:

• Your kubernetes cluster
• The cloud provider where it runs or indicate that it’s “on-prem”
• The hardware it runs on
• The type and size of disk in use–for example: hard disk, SSD, NVMe, etc.
• CPU and memory resources set on each type of InfluxDB pod
• The number of pods in each InfluxDB StatefulSet and Deployment
• The type of object store used and how it is hosted
• How the Catalog (PostgreSQL-compatible database) is hosted
• Indicate if either the Object store or the Catalog is shared by more than one InfluxDB Clustered product
  • If so, describe the network-level topology of your setup

Document your data schema

Document your the data schema to help InfluxData engineers better understand the conditions that reproduce your issue.

Establish query performance degradation conditions

The most effective way to investigate query performance is to have a good understanding of the conditions in which you don’t see the expected performance. Consider the following:

• Does this always happen, or only sometimes?
  • If only sometimes, is it at a consistent time of day or over a consistent period?
• Will a single query execution reproduce the issue, or does it only appear with multiple queries are running at the same time?
• How are you executing the queries? For example:
  • influxctl
  • Client libraries
  • Other environments or tools

Reduce query noise

Test in an environment without periodic or intermittent queries to measure baseline system performance without additional query noise.

When running multiple tests with different queries, allow the system to recover between tests. Wait at least one minute after receiving a query result before executing the next query.

Establish baseline single-query performance

Perform some tests with single queries in isolation to measure baseline performance. This approach may not always reproduce your issue but can provide useful data for analysis by InfluxData engineers.

Run queries at multiple load scales

If the issue isn’t replicated after reducing query noise and establishing baseline single-query performance, systematically increase query concurrency to reproduce the problem and identify the scale at which it occurs–for example, run the following test plan.

1. Turn off intermittent or periodic InfluxDB queries and allow the cluster to recover.
2. Run Query A and allow the cluster to recover for 1 minute.
3. Run 5 concurrent instances of Query A and allow the cluster to recover for 1 minute.
4. Run 10 concurrent instances of Query A and allow the cluster to recover for 1 minute.
5. Run 20 concurrent instances of Query A and allow the cluster to recover for 1 minute.
6. Run 40 concurrent instances of Query A and allow the cluster to recover for 1 minute.
7. Provide InfluxData the debug information associated with each test run.

Your test findings and associated debug information from your Kubernetes environment can help recommend configuration changes to improve query performance as your usage scales.

Gather debug information

Shortly after testing a problematic query against your InfluxDB cluster, collect the following debug information.

Kubernetes-specific information

Outputs:

• ${DATETIME}-cluster-info.tar.gz

DATETIME="$(date -Iminutes)"
kubectl cluster-info dump --namespace influxdb --output-directory "${DATETIME}-cluster-info/"
tar -czf "${DATETIME}-cluster-info.tar.gz" "${DATETIME}-cluster-info/"

Clustered-specific information

Outputs:

• app-instance.yml: Provide a copy of your AppInstance manifest.

Query analysis

Use EXPLAIN commands to output query plan information for a long-running query.

Outputs (InfluxQL):

• explain.csv
• explain-verbose.csv
• explain-analyze.csv

Outputs (SQL):

• explain.txt
• explain-verbose.txt
• explain-analyze.txt

In the examples below, replace the following:

• DATABASE_NAME: The name of the database to query
• DATABASE_TOKEN: A database token with read permissions on the queried database
• YOUR_QUERY: Your long-running query (formatted as a single line with escaped double quotes (\"))

EXPLAIN

influxctl \
  --config config.toml \
    query \
  --database 
DATABASE_NAME
 \
  --format table \
  --token 
DATABASE_TOKEN
 \
  "EXPLAIN 
YOUR_QUERY
;" > explain.txt

curl --get "https://cluster-host.com/query" \
  --output "./explain.csv" \
  --header "Authorization: Bearer 
DATABASE_TOKEN
" \
  --header "Accept: application/csv" \
  --data-urlencode "db=
DATABASE_NAME
" \
  --data-urlencode "q=EXPLAIN 
YOUR_QUERY
"

EXPLAIN VERBOSE

influxctl \
  --config config.toml \
    query \
  --database 
DATABASE_NAME
 \
  --format table \
  --token 
DATABASE_TOKEN
 \
  "EXPLAIN VERBOSE 
YOUR_QUERY
;" > explain-verbose.txt

curl --get "https://cluster-host.com/query" \
  --output "./explain-verbose.csv" \
  --header "Authorization: Bearer 
DATABASE_TOKEN
" \
  --header "Accept: application/csv" \
  --data-urlencode "db=
DATABASE_NAME
" \
  --data-urlencode "q=EXPLAIN VERBOSE 
YOUR_QUERY
"

EXPLAIN ANALYZE

influxctl \
  --config config.toml \
    query \
  --database 
DATABASE_NAME
 \
  --format table \
  --token 
DATABASE_TOKEN
 \
  "EXPLAIN ANALYZE 
YOUR_QUERY
;" > explain-analyze.txt

curl --get "https://cluster-host.com/query" \
  --output "./explain-analyze.csv" \
  --header "Authorization: Bearer 
DATABASE_TOKEN
" \
  --header "Accept: application/csv" \
  --data-urlencode "db=
DATABASE_NAME
" \
  --data-urlencode "q=EXPLAIN ANALYZE 
YOUR_QUERY
"

Gather system information

If queries are slow for a specific table, run the following system queries to collect information for troubleshooting:

• Collect table information
• Collect compaction information for the table
• Collect partition information for multiple tables

To optimize system queries, use table_name, partition_key, and partition_id filters. In your queries, replace the following:

• TABLE_NAME: the table to retrieve partitions for
• PARTITION_ID: a partition ID (int64)
• PARTITION_KEY: a partition key derived from the table’s partition template. The default format is %Y-%m-%d (for example, 2024-01-01).

Collect table information

SELECT *
FROM system.tables
WHERE table_name = '
TABLE_NAME
';

Collect compaction information for the table

Query the system.compactor table to collect compaction information–for example, run one of the following queries:

SELECT * 
FROM system.compactor 
WHERE
  table_name = '
TABLE_NAME
' 
    AND partition_key = '
PARTITION_KEY
';

SELECT * 
FROM system.compactor 
WHERE
  table_name = '
TABLE_NAME
' 
    AND partition_id = '
PARTITION_ID
';

Collect partition information for multiple tables

If the same queries are slow on more than 1 table, also run the following query to collect the size and number of partitions for all tables:

SELECT table_name,
  COUNT(*) as partition_count,
  MAX(last_new_file_created_at) as last_new_file_created_at,
  SUM(total_size_mb) as total_size_mb
FROM system.partitions
WHERE table_name IN ('foo', 'bar', 'baz')
GROUP BY table_name;