clickhouse-flamegraph

ClickHouse flamegraph

command line utility for visualizing clickhouse system.trace_log as flamegraph, thanks https://gist.github.com/alexey-milovidov/92758583dd41c24c360fdb8d6a4da194 for original idea

Prepare Clickhouse server

• install clickhouse-server package version 20.6 or higher how described in documentation
• enable query_log and sampling profiling in settings on each server in your cluster for example add following files:

create /etc/clickhouse-server/config.d/profiling.xml (need server restart to apply changes)

<yandex>
    <!-- Simple server-wide memory profiler. Collect a stack trace at every peak allocation step (in bytes).
         Data will be stored in system.trace_log table with query_id = empty string.
         Zero means disabled. Minimal effective value is 4 MiB.
         Data will dump with 'Memory' trace_type
      -->
    <total_memory_profiler_step>4194304</total_memory_profiler_step>
    <!-- Collect random allocations and deallocations and write them into system.trace_log with 'MemorySample' trace_type.
            The probability is for every alloc/free regardless to the size of the allocation.
            Note that sampling happens only when the amount of untracked memory exceeds the untracked memory limit,
             which is 4 MiB by default but can be lowered if 'total_memory_profiler_step' is lowered.
            You may want to set 'total_memory_profiler_step' to 1 for extra fine grained sampling.
         -->
    <total_memory_tracker_sample_probability>0.01</total_memory_tracker_sample_probability>
</yandex>

create /etc/clickhouse-server/users.d/profiling.xml (config reloaded every 1sec or via SYSTEM CONFIG RELOAD)

<yandex>
    <profiles>
        <default>
            <log_queries>1</log_queries>
            <allow_introspection_functions>1</allow_introspection_functions>
            <!-- 25 times per second sampling profiler -->
            <query_profiler_real_time_period_ns>40000000</query_profiler_real_time_period_ns>
            <query_profiler_cpu_time_period_ns>40000000</query_profiler_cpu_time_period_ns>

            <!-- memory profiling for each query, dump stack trace when 1MiB allocation with query_id not empty
            Whenever query memory usage becomes larger than every next step in number of bytes the memory profiler 
            will collect the allocating stack trace. 
            Zero means disabled memory profiler. 
            Values lower than a few megabytes will slow down query processing. 
            -->
            <memory_profiler_step>1048576</memory_profiler_step>
            <!-- Small allocations and deallocations are grouped in thread local variable and tracked or profiled only 
                when amount (in absolute value) becomes larger than specified value. 
                If the value is higher than 'memory_profiler_step' it will be effectively lowered to 'memory_profiler_step'.
            -->
            <max_untracked_memory>1048576</max_untracked_memory>            
            <!-- Collect random allocations and deallocations and write them into system.trace_log with 'MemorySample' trace_type. 
                 The probability is for every alloc/free regardless to the size of the allocation. 
                 Note that sampling happens only when the amount of untracked memory exceeds 'max_untracked_memory'. 
                 You may want to set 'max_untracked_memory' to 0 for extra fine grained sampling. -->
            <memory_profiler_sample_probability>0.01</memory_profiler_sample_probability>    

        </default>
    </profiles>
</yandex>

Installation

currently clickhouse-flamegraph required perl and flamegraph.pl for correct work, you can just download latest packages from https://github.com/Slach/clickhouse-flamegraph/releases

Simplest way, but you should skip it, if you care about security ;-)

curl -sL https://raw.githubusercontent.com/Slach/clickhouse-flamegraph/master/install.sh | sudo bash

Linux deb or rpm based distributive, amd64 architecture

PKG_MANAGER=$(command -v dpkg || command -v rpm)
PKG_EXT=$(if [[ "${PKG_MANAGER}" == "/usr/bin/dpkg" ]]; then echo "deb"; else echo "rpm"; fi)
cd $TEMP
echo "$(curl -sL https://github.com/Slach/clickhouse-flamegraph/releases/latest | grep href | grep -E "\\.rpm|\\.deb|\\.txt" | cut -d '"' -f 2)" | sed -e "s/^\\/Slach/https:\\/\\/github.com\\/Slach/" | wget -nv -c -i -
grep $PKG_EXT clickhouse-flamegraph_checksums.txt | sha256sum
${PKG_MANAGER} -i clickhouse-flamegraph*.${PKG_EXT}

MacOS 64bit

brew install wget
cd $TEMP
echo "$(curl -sL https://github.com/Slach/clickhouse-flamegraph/releases/latest | grep href | grep -E "darwin_amd64\\.tar\\.gz|\\.txt" | cut -d '"' -f 2)" | sed -e "s/^\\/Slach/https:\\/\\/github.com\\/Slach/" | wget -nv -c -i -
grep darwin_amd64.tar.gz clickhouse-flamegraph_checksums.txt | sha256sum
tar -xvfz -C /usr/bin clickhouse-flamegraph*darwin_amd64.tar.gz

Windows 64bit

install CYGWIN https://cygwin.com/install.html from setup.exe install following packages:

• wget
• sha256sum
• bash run following script

cd $TEMP
echo "$(curl -sL https://github.com/Slach/clickhouse-flamegraph/releases/latest | grep href | grep -E "windows_amd64\\.tar\\.gz|\\.txt" | cut -d '"' -f 2)" | sed -e "s/^\\/Slach/https:\\/\\/github.com\\/Slach/" | wget -nv -c -i -
grep windows_amd64.tar.gz clickhouse-flamegraph_checksums.txt | sha256sum
tar -xvfz -C /usr/bin clickhouse-flamegraph*windows_amd64.tar.gz

Usage

USAGE:
   clickhouse-flamegraph [global options] command [command options] [arguments...]

GLOBAL OPTIONS:
   --width value                                width of image (default 1200) (default: 1200)
   --height value                               height of each frame (default 16) (default: 16)
   --flamegraph-script value                    path of flamegraph.pl. if not given, find the script from $PATH [%CH_FLAME_FLAMEGRAPH_SCRIPT%]
   --output-dir value, -o value                 destination path of generated flamegraphs files (default: "./clickhouse-flamegraphs/") [%CH_FLAME_OUTPUT_DIR%]
   --date-from value, --from value              filter system.trace_log from date in any parsable format, see https://github.com/araddon/dateparse (default: "2020-10-13 09:55:00 +0500") [%CH_FLAME_DATE_FROM%]
   --date-to value, --to value                  filter system.trace_log to date in any parsable format, see https://github.com/araddon/dateparse (default: "2020-10-13 10:00:00 +0500") [%CH_FLAME_DATE_TO%]
   --query-filter value, --query-regexp value   filter system.query_log by any regexp, see https://github.com/google/re2/wiki/Syntax [%CH_FLAME_QUERY_FILTER%]
   --query-ids value, --query-id value          filter system.query_log by query_id field, comma separated list [%CH_FLAME_QUERY_IDS%]
   --trace-types value, --trace-type value      filter system.trace_log by trace_type field, comma separated list (default: "Real", "CPU", "Memory", "MemorySample") [%CH_FLAME_TRACE_TYPES%]
   --clickhouse-dsn value, --dsn value          clickhouse connection string, see https://github.com/mailru/go-clickhouse#dsn (default: "http://localhost:8123/default") [%CH_FLAME_CLICKHOUSE_DSN%]
   --clickhouse-cluster value, --cluster value  clickhouse cluster name from system.clusters, all flame graphs will get from cluster() function, see https://clickhouse.com/docs/en/sql-reference/table-functions/cluster [%CH_FLAME_CLICKHOUSE_CLUSTER%]
   --tls-certificate value                      X509 *.cer, *.crt or *.pem file for https connection, use only if tls_config exists in --dsn, see https://clickhouse.com/docs/en/operations/server-configuration-parameters/settings/#server_configuration_parameters-openssl for details [%CH_FLAME_TLS_CERT%]
   --tls-key value                              X509 *.key file for https connection, use only if tls_config exists in --dsn [%CH_FLAME_TLS_KEY%]
   --tls-ca value                               X509 *.cer, *.crt or *.pem file used with https connection for self-signed certificate, use only if tls_config exists in --dsn, see https://clickhouse.com/docs/en/operations/server-configuration-parameters/settings/#server_configuration_parameters-openssl for details [%CH_FLAME_TLS_CA%]
   --output-format value, --format value        accept values: svg, txt (see https://github.com/brendangregg/FlameGraph#2-fold-stacks), json (see https://github.com/spiermar/d3-flame-graph/#input-format,  (default: "svg") [%CH_FLAME_OUTPUT_FORMAT%]
   --normalize-query, --normalize               group stack by normalized queries, instead of query_id, see https://clickhouse.com/docs/en/sql-reference/functions/string-functions/#normalized-query (default: false) [%CH_FLAME_NORMALIZE_QUERY%]
   --debug, --verbose                           show debug log (default: false) [%CH_FLAME_DEBUG%]
   --console                                    output logs to console format instead of json (default: false) [%CH_FLAME_LOG_TO_CONSOLE%]
   --help, -h                                   show help (default: false)
   --version, -v                                print the version (default: false)

Tips&Tricks

• When you can't change /etc/clickhouse-server/*.xml files on server, just add SETTINGS query_profiler_real_time_period_ns=40000000, query_profiler_cpu_time_period_ns=40000000 to end of your SQL query. And run following command

clickhouse-flamegraph --dsn=tcp://clickhouse-server:9000/?pool_size=1 

• For check all settings in server set properly run following SQL query on your ClickHouse server

SELECT * FROM system.settings WHERE match(name,'introspection|log_queries|profiler|sample') FORMAT Vertical

TODO

• implement json format and webhooks
• try implement interactive dashboard with http://dash.plot.ly
• try integrate with https://github.com/samber/grafana-flamegraph-panel