ClickHouse datasource for Grafana 4.6+
ClickHouse datasource plugin provides a support for ClickHouse as a backend database.
Grafana 7.x setup notes
Grafana team currently doesn't provide worked signing method for community plugins https://community.grafana.com/t/how-to-create-a-signed-backend-plugin/30068/2
so, for properly setup you need change configuration option
or setup environment variable
You can install plugin from grafana.net
Copy files to your Grafana plugin directory. Restart Grafana, check data sources list at http://your.grafana.instance/datasources/new, choose ClickHouse option.
- Access to CH via HTTP
- Query setup
- Raw SQL editor
- Query formatting
- Macros support
- Additional functions
- Table view
- SingleStat view
- Ad-hoc filters
Access to CH via HTTP
Page configuration is standard
There is a small feature - ClickHouse treats HTTP Basic Authentication credentials as a database user and will try to run queries using its name.
Using of CHProxy will bring additional features:
- Easily setup
HTTPSaccess to ClickHouse as shown here to provide secure access.
- Limit concurrency and execution time for requests from
Grafanaas shown here to prevent
- Protection against request bursts for dashboards with numerous graphs.
CHProxyallows queueing requests and execute them sequentially. To learn more - read about params
max_queue_timeat CHProxy page.
- Response caching for the most frequent queries as shown here.
ClickHousefrom excessive refreshes and will be optimal option for popular dashboards.
Hint - if you need to cache requests like
last 24hwhere timestamp changes constantly then try to use
Query setup interface:
FROM contains two options: database and table. Table values depends on a selected database.
Second row contains selectors for time filtering:
Plugin will try to detect date columns automatically
Column:TimeStampare required for time-based macros and functions because all analytics based on these values
Go to Query is just a toggler to Raw SQL Editor
Raw SQL Editor
Raw Editor allows custom SQL queries to be written:
Raw Editor allows typing queries, get info about functions and macros, format queries as Clickhouse do. Under the Editor you can find a raw query (all macros and functions have already been replaced) which will be sent directly to ClickHouse.
Plugin supports the following marcos:
- $table - replaced with selected table name from Query Builder
- $dateCol - replaced with
Column:Datevalue from Query Builder
- $dateTimeCol - replaced with
Column:TimeStampvalue from Query Builder
- $from - replaced with (timestamp with ms)/1000 value of UI selected "Time Range:From"
- $to - replaced with (timestamp with ms)/1000 value of UI selected "Time Range:To"
- $interval - replaced with selected "Group by a time interval" value (as a number of seconds)
- $timeFilter - replaced with currently selected "Time Range". Requires Column:Date and Column:DateTime or Column:TimeStamp to be selected.
- $timeFilterByColumn($column) - replaced with currently selected "Time Range" for a column passed as
$columnargument. Use it in queries or query variables as
- $timeSeries - replaced with special ClickHouse construction to convert results as time-series data. Use it as "SELECT $timeSeries...".
- $unescape - unescapes variable value by removing single quotes. Used for multiple-value string variables: "SELECT $unescape($column) FROM requests WHERE $unescape($column) = 5"
- $adhoc - replaced with a rendered ad-hoc filter expression, or "1" if no ad-hoc filters exist. Since ad-hoc applies automatically only to outer queries the macros can be used for filtering in inner queries.
A description of macros is available by typing their names in Raw Editor
Functions are just templates of SQL queries, and you can check the final query at Raw SQL Editor mode. If you need some additional complexity - just copy raw sql into Raw Editor and make according changes. Remember that macros are still available to use.
There are some limits in function use because of poor query analysis:
- Column:Date and Column:DateTime or Column:TimeStamp must be set in Query Builder
- Query must begin from function name
- Only one function can be used per query
Plugin supports the following functions:
$rate(cols...) - converts query results as "change rate per interval"
$rate(countIf(Type = 200) AS good, countIf(Type != 200) AS bad) FROM requests
Query will be transformed into:
SELECT t, good / runningDifference(t / 1000) AS goodRate, bad / runningDifference(t / 1000) AS badRate FROM ( SELECT (intDiv(toUInt32(EventTime), 60)) * 1000 AS t, countIf(Type = 200) AS good, countIf(Type != 200) AS bad FROM requests WHERE ((EventDate >= toDate(1482796747)) AND (EventDate <= toDate(1482853383))) AND ((EventTime >= toDateTime(1482796747)) AND (EventTime <= toDateTime(1482853383))) GROUP BY t ORDER BY t )
$columns(key, value) - query values as array of [key, value], where key will be used as label
$columns(OSName, count(*) c) FROM requests ANY INNER JOIN oses USING (OS)
Query will be transformed into:
SELECT t, groupArray((OSName, c)) AS groupArr FROM ( SELECT (intDiv(toUInt32(EventTime), 60) * 60) * 1000 AS t, OSName, count(*) AS c FROM requests ANY INNER JOIN oses USING (OS) WHERE ((EventDate >= toDate(1482796627)) AND (EventDate <= toDate(1482853383))) AND ((EventTime >= toDateTime(1482796627)) AND (EventTime <= toDateTime(1482853383))) GROUP BY t, OSName ORDER BY t, OSName ) GROUP BY t ORDER BY t
This will help to build the next graph:
$rateColumns(key, value) - is a combination of $columns and $rate
$rateColumns(OS, count(*) c) FROM requests
Query will be transformed into:
SELECT t, arrayMap(lambda(tuple(a), (a.1, a.2 / runningDifference(t / 1000))), groupArr) FROM ( SELECT t, groupArray((OS, c)) AS groupArr FROM ( SELECT (intDiv(toUInt32(EventTime), 60) * 60) * 1000 AS t, OS, count(*) AS c FROM requests WHERE ((EventDate >= toDate(1482796867)) AND (EventDate <= toDate(1482853383))) AND ((EventTime >= toDateTime(1482796867)) AND (EventTime <= toDateTime(1482853383))) GROUP BY t, OS ORDER BY t, OS ) GROUP BY t ORDER BY t )
$perSecond(cols...) - converts query results as "change rate per interval" for Counter-like(growing only) metrics
$perSecond(Requests) FROM requests
Query will be transformed into:
SELECT t, if(runningDifference(max_0) < 0, nan, runningDifference(max_0) / runningDifference(t / 1000)) AS max_0_Rate FROM ( SELECT (intDiv(toUInt32(EventTime), 60) * 60) * 1000 AS t, max(Requests) AS max_0 FROM requests WHERE ((EventDate >= toDate(1535711819)) AND (EventDate <= toDate(1535714715))) AND ((EventTime >= toDateTime(1535711819)) AND (EventTime <= toDateTime(1535714715))) GROUP BY t ORDER BY t )
// see issue 78 for the background
$perSecondColumns(key, value) - is a combination of $columns and $perSecond for Counter-like metrics
$perSecondColumns(Protocol, Requests) FROM requests WHERE Protocol in ('udp','tcp')
Query will be transformed into:
SELECT t, groupArray((Protocol, max_0_Rate)) AS groupArr FROM ( SELECT t, Protocol, if(runningDifference(max_0) < 0, nan, runningDifference(max_0) / runningDifference(t / 1000)) AS max_0_Rate FROM ( SELECT (intDiv(toUInt32(EventTime), 60) * 60) * 1000 AS t, Protocol, max(Requests) AS max_0 FROM requests WHERE ((EventDate >= toDate(1535711819)) AND (EventDate <= toDate(1535714715))) AND ((EventTime >= toDateTime(1535711819)) AND (EventTime <= toDateTime(1535714715))) AND (Protocol IN ('udp', 'tcp')) GROUP BY t, Protocol ORDER BY t, Protocol ) ) GROUP BY t ORDER BY t
// see issue 80 for the background
If you add a template variable of the type
Query, you can write a ClickHouse query that can
return things like measurement names, key names or key values that are shown as a dropdown select box.
For example, you can have a variable that contains all values for the
hostname column in a table if you specify a query like this in the templating variable Query setting.
SELECT hostname FROM host
To use time range dependent macros like
timeFilterByColumn($column) in your query the refresh mode of the template variable needs to be set to On Time Range Change.
SELECT event_name FROM event_log WHERE $timeFilterByColumn(time_column)
Another option is a query that can create a key/value variable. The query should return two columns that are named
__text column value should be unique (if it is not unique then the first value will use). The options in the dropdown will have a text and value that allows you to have a friendly name as text and an id as the value. An example query with
hostname as the text and
id as the value:
SELECT hostname AS __text, id AS __value FROM host
You can also create nested variables. For example if you had another variable named
region. Then you could have the hosts variable only show hosts from the current selected region with a query like this (if
region is a multi-value variable then use the
IN comparison operator rather than
= to match against multiple values):
SELECT hostname FROM host WHERE region IN ($region)
If you are using templating to feed your predicate, you will face performance degradation when everything will select as the predicate, and it's not necessary. It's also true for textbox when nothing is enter, you have to write specific sql code to handle that.
To workaround this issue a new macro $conditionalTest(SQL Predicate,$variable) can be used to remove some part of the query. If the variable is type query with all selected or if the variable is a textbox with nothing enter, then the SQL Predicate is not include in the generated query.
To give an example: with 2 variables $var query with include All option $text textbox
The following query
SELECT $timeSeries as t, count() FROM $table WHERE $timeFilter $conditionalTest(AND toLowerCase(column) in ($var),$var) $conditionalTest(AND toLowerCase(column2) like '%$text%',$text) GROUP BY t ORDER BY t
$var is selected as "All" value, and the
$text variable is empty, the query will be converted into:
SELECT $timeSeries as t, count() FROM $table WHERE $timeFilter GROUP BY t ORDER BY t
$var template variable have select some elements, and the
$text template variable has at least one char, the query will be converted into:
SELECT $timeSeries as t, count() FROM $table WHERE $timeFilter AND toLowerCase(column) in ($var) AND toLowerCase(column2) like '%$text%' GROUP BY t ORDER BY t
Working with panels
Pie Chart (https://grafana.com/plugins/grafana-piechart-panel)
Remember that pie chart plugin is not welcome for using in grafana - see https://grafana.com/blog/2015/12/04/friends-dont-let-friends-abuse-pie-charts
To create "Top 5" diagram we will need two queries: one for 'Top 5' rows and one for 'Other' row.
SELECT 1 AS t, /* fake timestamp value */ UserName, sum(Requests) AS Reqs FROM requests GROUP BY t, UserName ORDER BY Reqs DESC LIMIT 5
SELECT 1 AS t, /* fake timestamp value */ UserName, sum(Requests) AS Reqs FROM requests GROUP BY t, UserName ORDER BY Reqs LIMIT 5,10000000000000 /* select some ridiculous number after first 5 */
There are don't contain any tricks in displaying time-series data. To print summary data, omit time column, and format the result as "Table".
SELECT UserName, sum(Requests) as Reqs FROM requests GROUP BY UserName ORDER BY Reqs
Vertical histogram (https://grafana.com/plugins/graph)
To make the vertical histogram from graph panel we will need to edit some settings:
- Display -> Draw Modes -> Bars
- Axes -> X-Axis -> Mode -> Series
You can use next query:
$columns( Size, sum(Items) Items) FROM some_table
// It is also possible to use query without macros
Worldmap panel (https://github.com/grafana/worldmap-panel)
If you have a table with country/city codes:
SELECT 1, Country AS c, sum(Requests) AS Reqs FROM requests GLOBAL ANY INNER JOIN ( SELECT Country, CountryCode FROM countries ) USING (CountryCode) WHERE $timeFilter GROUP BY c ORDER BY Reqs DESC
If you are using geohash set following options:
You can make following query with
If there is an Ad-hoc variable, plugin will fetch all columns of all tables of all databases (except system database) as tags.
So in dropdown menu will be options like
database.table.column. If you specify the default database it will only fetch tables and columns from that database, and the dropdown menu will have an option like
If there are ENUM columns, the plugin will fetch their options and use them as tag values.
Also, plugin will fetch 300 unique values for fields with other types.
Plugin will apply Ad-hoc filters to all queries on the dashboard if their settings
$table are the same
database.table specified in Ad-hoc control. If the ad-hoc filter doesn't specify a table, it will apply to all queries regardless of the table.
This is useful if the dashboard contains queries to multiple different tables.
There are no option to apply OR operator for multiple Ad-hoc filters - see grafana/grafana#10918
There are no option to use IN operator for Ad-hoc filters due to Grafana limitations
There may be cases when CH contains too many tables and columns so their fetching could take notably amount of time. So, if you need
to have multiple dashboards with different databases using of
default database won't help. The best way to solve this will be to have parametrized
ad-hoc variable in dashboard settings. Currently, it's not supported by Grafana interface (see issue).
As a temporary workaround, plugin will try to look for variable with name
adhoc_query_filter and if it exists will use its value as query to fetch columns.
For this purpose we recommend creating some variable
constant with the name
adhoc_query_filter and set the value similar to the following one:
SELECT database, table, name, type FROM system.columns WHERE table='myTable' ORDER BY database, table
That should help to control data fetching by ad-hoc queries.
To use time range dependent macros like
$to in your query the refresh mode of the template variable needs to be set to On Time Range Change.
SELECT ClientID FROM events WHERE EventTime > toDateTime($from) AND EventTime < toDateTime($to)
Configure the Datasource with Provisioning
It’s now possible to configure datasources using config files with Grafana’s provisioning system. You can read more about how it works and all the settings you can set for datasources on the provisioning docs page.
Here are some provisioning example:
apiVersion: 1 datasources: - name: Clickhouse type: vertamedia-clickhouse-datasource access: proxy url: http://localhost:8123 # <bool> enable/disable basic auth basicAuth: # <string> basic auth username basicAuthUser: # <string> basic auth password basicAuthPassword: # <bool> enable/disable with credentials headers withCredentials: # <bool> mark as default datasource. Max one per org isDefault: # <map> fields that will be converted to json and stored in json_data jsonData: # <bool> enable/disable sending 'add_http_cors_header=1' parameter addCorsHeader: # <bool> enable/disable using POST method for sending queries usePOST: # <string> default database name defaultDatabase:
Some settings and security params are the same for all datasources. You can find them here.
Why time series last point is not the real last point?
Plugin extrapolates last datapoint if time range is
last N to avoid displaying of constantly decreasing graphs
when timestamp in a table is rounded to minute or bigger.
If it so then in 99% cases last datapoint will be much less than previous one, because last minute is not finished yet.
That's why plugin checks prev datapoints and tries to predict last datapoint value just as it was already written into db.
This behavior could be turned off via "Extrapolation" checkbox in query editor.
Which table schema used in SQL query examples?
All examples in this plugin use following table schema:
CREATE TABLE IF NOT EXISTS countries( Country LowCardinality(String), CountryCode LowCardinality(String) ) ENGINE MergeTree() ORDER BY (CountryCode, Country); CREATE TABLE IF NOT EXISTS oses ( OSName LowCardinality(String), OS LowCardinality(String) ) ENGINE MergeTree() ORDER BY (OS); CREATE TABLE IF NOT EXISTS requests( EventTime DateTime, EventDate Date, Protocol LowCardinality(String), UserName LowCardinality(String), OS LowCardinality(String), CountryCode LowCardinality(String), Type UInt8, Requests UInt32 ) ENGINE=MergeTree() ORDER BY (EventDate, EventTime, Type, OS, Protocol, UserName) PARTITION BY toYYYYMM(EventDate);
What about alerts support?
Alerts feature requires changes in
Grafana's backend, which can be extended only for Grafana 6.5+.
Grafana's maintainers are working on this feature.
Current alerts support for
clickhouse-grafana datasource plugin in beta and support only for amd64 architecture for Linux, MacOSX, Windows.
The resulting alerts should look like this
If you have any idea for an improvement or found a bug do not hesitate to open an issue or submit a pull request. We will appreciate any help from the community which will make working with such amazing products as ClickHouse and Grafana more convenient.
see CONTRIBUTING.md for Development and Pull request Contributing instructions
MIT License, please see LICENSE for details.