A collection of command line tools for Crate developers (and maybe users as well).
- To quickly produce sample data. Often if someone reports an issue sample data is required to be able to reproduce it. insert-fake-data and insert-json address this problem.
- To benchmark queries & compare runtime across Crate versions. timeit 🕐, run-spec and run-track can be used to get runtime statistics of queries. These tools focus on response latencies. Being able to benchmark throughput is NOT a goal of cr8. Similarly, being able to simulate real-world use cases is also NOT a goal of cr8.
Note
Although most commands output text by default. Most take a --output-fmt
json argument to output JSON.
This is useful if used together with jq to post-process the output
Python >= 3.7 is required to use the command line tools.
Install them using pip:
python3 -m venv venv venv/bin/python -m pip install cr8
This will install cr8 into venv/bin
An alternative is to download a single zipapp file from the releases page.
The main binary is called cr8 which contains a couple of sub-commands.
Use cr8 -h or cr8 <subcommand> -h to get a more detailed usage
description.
The included sub-commands are described in more detail below:
Tip:
Any <subcommand> with --hosts argument supports password authentication
like this:
cr8 <subcommand> --hosts http://username:password@localhost:4200 <remaining args>
cr8 supports command completion in both bash and zsh via argcomplete.
- Install
argcomplete - Run
activate-global-python-argcomplete
Make sure you're using the argcomplete > 3.0.
In older versions of argcomplete it would be necessary to use the
bashcompinit compatibility layer in zsh and register the application
via eval "$(register-python-argcomplete cr8)". See the upstream
documentation for details.
A tool that can be used to measure the runtime of a given SQL statement on a cluster:
>>> echo "select name from sys.cluster" | cr8 timeit --hosts localhost:4200
Runtime (in ms):
mean: ... ± ...
min/max: ... → ...
Percentile:
50: ... ± ... (stdev)
95: ...
99.9: ...
A tool that can be used to fill a table with random data. The script will generate the records using faker.
For example given the table as follows:
create table x.demo (
id int,
name text,
country text
);
The following command can be used to insert 1000 records:
>>> cr8 insert-fake-data --hosts localhost:4200 --table x.demo --num-records 200
Found schema:
{
"country": "text",
"id": "integer",
"name": "text"
}
Using insert statement:
insert into "x"."demo" ("id", "name", "country") values ($1, $2, $3)
Will make 1 requests with a bulk size of 200
Generating fake data and executing inserts
<BLANKLINE>
It will automatically read the schema from the table and map the columns to faker providers and insert the give number of records.
(Currently only top-level columns are supported)
An alternative way to generate random records is mkjson which can be used together with
insert-json.
insert-json can be used to insert records from a JSON file:
>>> cat tests/demo.json | cr8 insert-json --table x.demo --hosts localhost:4200
Executing inserts: bulk_size=1000 concurrency=25
Runtime (in ms):
mean: ... ± 0.000
Or simply print the insert statement generated from a JSON string:
>>> echo '{"name": "Arthur"}' | cr8 insert-json --table mytable
('insert into mytable ("name") values ($1)', ['Arthur'])
...
Copies data from one CrateDB cluster or PostgreSQL server to another.
>>> cr8 insert-from-sql \
... --src-uri "postgresql://crate@localhost:5432/doc" \
... --query "SELECT name FROM x.demo" \
... --hosts localhost:4200 \
... --table y.demo \
INSERT INTO y.demo ("name") VALUES ($1)
Runtime (in ms):
...
The concurrency option of the command only affects the number of concurrent
write operations that will be made. There will always be a single read
operation, so copy operations may be bound by the read performance.
A tool to run benchmarks against a cluster and store the result in another cluster. The benchmark itself is defined in a spec file which defines setup, benchmark and teardown instructions.
The instructions itself are just SQL statements (or files containing SQL statements).
In the specs folder is an example spec file.
Usage:
>>> cr8 run-spec specs/sample.toml localhost:4200 -r localhost:4200
# Running setUp
# Running benchmark
<BLANKLINE>
## Running Query:
Name: count countries
Statement: select count(*) from countries
Concurrency: 2
Duration: 1
Runtime (in ms):
mean: ... ± ...
min/max: ... → ...
Percentile:
50: ... ± ... (stdev)
95: ...
99.9: ...
...
## Skipping (Version ...
Statement: ...
# Running tearDown
<BLANKLINE>
-r is optional and can be used to save the benchmark result into a cluster. A table named benchmarks will be created if it doesn't exist.
Writing spec files in python is also supported:
>>> cr8 run-spec specs/sample.py localhost:4200 # Running setUp # Running benchmark ...
Launch a Crate instance:
> cr8 run-crate 0.55.0
This requires Java 8.
run-crate supports chaining of additional commands using --. Under the
context of run-crate any host urls can be formatted using the
{node.http_url} format string:
>>> cr8 run-crate latest-stable -- timeit -s "select 1" --hosts '{node.http_url}'
# run-crate
===========
<BLANKLINE>
...
Starting Crate process
CrateDB launching:
PID: ...
Logs: ...
Data: ...
<BLANKLINE>
...
Cluster ready to process requests
<BLANKLINE>
<BLANKLINE>
# timeit
========
<BLANKLINE>
<BLANKLINE>
<BLANKLINE>
<BLANKLINE>
In the above example timeit is a cr8 specific sub-command. But it's
also possible to use arbitrary commands by prefixing them with @:
cr8 run-crate latest-nightly -- @http '{node.http_url}'
One common use of this feature is to quickly reproduce bug reports:
cr8 run-crate latest-nightly -- @crash --hosts {node.http_url} <<EOF
create table mytable (x int);
insert into mytable (x) values (1);
refresh mytable;
...
EOF
Another use case is to use run-crate in combination with run-spec and
git bisect:
git bisect run cr8 run-crate path/to/crate/src \
-- run-spec path/to/spec.toml '{node.http_url}' --fail-if '{runtime_stats.mean} > 15'
This could also be combined with timeout.
This can also be used in combination with the Java flight recorder to do profiling:
cr8 run-crate latest-nightly \
-e CRATE_HEAP_SIZE=4g \
-e CRATE_JAVA_OPTS="-Dcrate.signal_handler.disabled=true -XX:+UnlockDiagnosticVMOptions -XX:+DebugNonSafepoints -XX:+UnlockCommercialFeatures -XX:+FlightRecorder" \
-s discovery.type=single-node \
-- run-spec path/to/specs/example.toml {node.http_url} --action setup \
-- @jcmd {node.process.pid} JFR.start duration=60s filename=myrecording.jfr \
-- run-spec path/to/specs/example.toml {node.http_url} --action queries \
-- @jcmd {node.process.pid} JFR.stop
cr8 doesn't contain a dedicated command to spawn a CrateDB cluster. But you
can run cr8 run-crate <version> -s cluster.name=<name> to launch multiple
nodes. If the cluster name matches, it will form a cluster.
A tool to run .toml track files.
A track is a matrix definition of node version, configurations and spec files.
For each version and configuration a Crate node will be launched and all specs will be executed:
>>> cr8 run-track tracks/sample.toml # Version: latest-testing ## Starting Crate latest-testing, configuration: default.toml ### Running spec file: sample.toml # Running setUp # Running benchmark ...
A command to re-index all tables on a cluster which have been created in the previous major versions. So if you're running a 3.x CrateDB cluster, all tables from 2.x would be re-created:
>>> cr8 reindex --help usage: cr8 reindex [-h] --hosts HOSTS ...
cr8 supports using HTTP or the postgres protocol.
Note that using the postgres protocol will cause cr8 to measure the
round-trip time instead of the service time. So measurements will be different.
To use the postgres protocol, the asyncpg scheme must be used inside hosts URIs:
>>> echo "select 1" | cr8 timeit --hosts asyncpg://localhost:5432 Runtime (in ms): ...
To get a sandboxed environment with all dependencies installed use venv:
python -m venv .venv source .venv/bin/activate
Install the cr8 package using pip:
python -m pip install -e .
Run cr8:
cr8 -h
Tests are run with python -m unittest