InfluxDB driver for Elixir
This module has experimental parts that may change unexpectedly.
Tested influxdb version: 1.0.0
(see
.travis.yml
to be sure)
Add Instream as a dependency to your mix.exs
file:
defp deps do
[{ :instream, "~> 0.13" }]
end
You should also update your applications to include all necessary projects:
def application do
[ applications: [ :instream ]]
end
To run the tests you need to have the http-authentication enabled.
Using the statements from the .travis.yml
you can generate all necessary
users for the tests with their proper privileges.
Defining a connection requires defining a module:
defmodule MyApp.MyConnection do
use Instream.Connection, otp_app: :my_app
end
The :otp_app
name and the name of the module can be freely chosen.
They only need to be linked to an entry in your config.exs
:
config :my_app, MyApp.MyConnection,
host: "localhost",
http_opts: [ insecure: true, proxy: "http://company.proxy" ],
pool: [ max_overflow: 0, size: 1 ],
port: 8086,
scheme: "http",
writer: Instream.Writer.Line
Configuration can be done statically (as shown above) or by referencing your system environment:
config :my_app, MyApp.MyConnection,
port: { :system, "MY_ENV_VARIABLE" }
You now have a connection definition you can hook into your supervision tree:
Supervisor.start_link(
[ MyApp.MyConnection.child_spec ],
strategy: :one_for_one
)
The full connection configuration is split into two parts, compile time and runtime configuration.
Compile time configuration are, as the name implies, used during compilation
for the connection module. Currently the only key in this category is
:loggers
.
All other values are runtime configuration values that are directly accessed
from the application environment using
Application.get_env(connection_otp_app, connection_module)
and therefore can be changed without recompilation:
old_config = MyConnection.config()
new_config = Keyword.put(old_config, :host, "changed.host")
:ok = Application.put_env(:my_otp_app, MyConnection, new_config)
Only the host
key is mandatory for a connection configuration. The following
values will be used as defaults if no other value is set:
config :my_app, MyApp.MyConnection,
pool: [ max_overflow: 10, size: 5 ],
port: 8086,
scheme: "http",
writer: Instream.Writer.Line
This also means that per default the connection uses no authentication.
Internally all requests are done using
:hackney
.
The configuration key :http_opts
is directly passed to the client process.
Please see
:hackney.request/5
for a complete list of available options.
To connect to an influxdb instance with http_auth enabled you have to configure your credentials:
config :my_app, MyApp.MyConnection,
auth: [ method: :basic, username: "root", password: "root" ]
For method
you can choose between header authentication (basic auth) using
:basic
or query parameters using :query
. If nothing or an invalid value
is given the connection will be made using :basic
authentication.
If you are using the regular line protocol writer Instream.Writer.Line
you
are done without having anything to configure. It is used by default and
connects to the port you have configured for connection.
To write points over UDP you can adjust your configuration:
config :my_app, MyApp.MyConnection,
host: "localhost",
port_udp: 8089,
writer: Instream.Writer.UDP
The connection will then write using UDP and connecting to the port :port_udp
.
All non-write queries will be send to the regular :port
you have configured.
All queries are (by default) logged using Logger.debug/1
via the default
logging module Instream.Log.DefaultLogger
. To customize logging you have to
alter the configuration of your connection:
config :my_app, MyApp.MyConnection,
loggers: [
{ FirstLogger, :log_fun, [] },
{ SecondLogger, :log_fun, [ :additional, :args ]}
]
This configuration replaces the default logging module.
Configuration is given as a tuple of { module, function, arguments }
. The log
entry will be inserted as the first argument of the method call. It will be one
of Instream.Log.PingEntry
, Instream.Log.QueryEntry
,
Instream.Log.StatusEntry
or Instream.Log.WriteEntry
, depending on what type
of request should be logged.
Please be aware that every logger has to return the entry it received in order to allow combining multiple loggers.
In addition to query specific information every entry carries metadata around:
:query_time
: milliseconds it took to send request and receive the responseresponse_status
: status code or0
if not applicable/available
When using the default logger you have to re-configure :logger
to be able to
get them printed:
config :logger, :console,
format: "\n$time $metadata[$level] $levelpad$message\n",
metadata: [ :application, :pid, :query_time, :response_status ]
Warning: In order to log the :pid
(provided by :logger
) used to send the
queries you need to have at least elixir ~> 1.1.0
. Any earlier version will
fail because the String.Chars
protocol was not implemented for pids at that
time.
To prevent a query from logging you can pass an option to the execute call:
query |> MyApp.MyConnection.execute(log: false)
# also works with convenience methods
MyApp.MyConnection.ping(log: false)
To validate a connection you can send ping requests to the server:
MyApp.MyConnection.ping()
The response will be :pong
on success or :error
on any failure.
To ping "a host other than the first in your configuration" you can pass it explicitly:
MyApp.MyConnection.ping("some.host.name")
All values necessary to ping the host (scheme, port, ...) will be taken from the connection used. It does not matter whether the host is configured in that connection or not.
To get InfluxDB to verify status of your server you can send a status call:
MyApp.MyConnection.status()
As with ping requests you can target a specific host:
MyApp.MyConnection.status("some.host.name")
Every query can be executed asynchronously by passing [async: true]
to
MyApp.MyConnection.execute()
. The result will then always be an immediate
:ok
without waiting for the query to be actually executed.
To access the raw (undecoded) response of a query you can pass
[result_as: :raw]
to MyApp.MyConnection.execute()
.
Managing Databases:
# create "my_database"
"my_database"
|> Instream.Admin.Database.create()
|> MyApp.MyConnection.execute()
# drop "my_database"
"my_database"
|> Instream.Admin.Database.drop()
|> MyApp.MyConnection.execute()
Managing Retention Policies:
# create "my_rp" retention policy
# argument order: policy, database, duration, replication, default
Instream.Admin.RetentionPolicy.create(
"my_rp", "my_database", "1h", 3, true
)
|> MyApp.MyConnection.execute()
# drop "my_rp" retention policy
Instream.Admin.RetentionPolicy.drop("my_rp", "my_database")
|> MyApp.MyConnection.execute()
Please see the point "Series Definitions" on how to write data to your InfluxDB database.
Reading data:
# passing database to execute/1
"SELECT * FROM some_measurement"
|> MyApp.MyConnection.query(database: "my_database")
# defining database in the query
"SELECT * FROM \"my_database\".\"default\".\"some_measurement\""
|> MyApp.MyConnection.query()
# passing precision (= epoch) for query results
"SELECT * FROM some_measurement"
|> MyApp.MyConnection.query(precision: :minutes)
Some queries require you to switch from the regular read only context
(all GET requets) to a write context
(all POST requests).
When not using the query build you have to pass that information manually
to execute/2
:
"CREATE DATABASE create_in_write_mode"
|> MyApp.MyConnection.execute(method: :post)
Using all default values and no specific parameters each query is allowed to
take up to 5000 milliseconds (GenServer.call/2
timeout) to complete.
That may be too long or not long enough in some cases.
To change that timeout you can configure your connection:
# lowering timeout to 500 ms
config :my_app,
MyApp.MyConnection,
query_timeout: 500
or pass an individual timeout for a single query:
MyApp.MyConnection.execute(query, timeout: 250)
A passed or connection wide timeout configuration override any :recv_timeout
of your :hackney
(HTTP client) configuration.
This does not apply to write requests. They are currently only affected by
configured :recv_timeout
values. Setting a connection timeout enables you to
have a different timeout for read and write requests.
Note: You will probably see some MatchError
messages. These are related
to the current pool clients not matching for timeouts returned by :hackney
.
This behaviour will change "soon-ish".
Experimental definition! Will change often and unexpected! (or may disappear...)
Using the query builder you can avoid writing your select statements by hand:
import Instream.Query.Builder
# SELECT one, or, more, fields FROM some_measurement
from(MySeries)
|> select([ "one", "or", "more", "fields" ])
|> MyApp.MyConnection.query()
# SELECT * FROM some_measurement WHERE binary = 'foo' AND numeric = 42
from("some_measurement")
|> where(%{ binary: "foo", numeric: 42 })
|> MyApp.MyConnection.query()
If you do not want to define the raw maps for writing data you can pre-define a series for later usage:
defmodule MySeries do
use Instream.Series
series do
database "my_database"
measurement "my_measurement"
tag :bar
tag :foo
field :value
end
end
You can include a default value for tags and fields in your series definition:
series do
measurement "my_measurement"
tag :host, default: "www"
field :value, default: 100
end
These values will be pre-assigned when using the data struct.
All fields or tags without a default value will be set to nil
.
You can then use this module to assemble a data point (one at a time) for writing:
data = %MySeries{}
data = %{ data | fields: %{ data.fields | value: 17 }}
data = %{ data | tags: %{ data.tags | bar: "bar", foo: "foo" }}
And then write one or many at once:
data
|> MyApp.MyConnection.write()
# write the point asynchronously
data
|> MyApp.MyConnection.write(async: true)
# write multiple points at once
[ point_1, point_2, point_3 ]
|> MyApp.MyConnection.write()
If you want to pass an explicit timestamp to the database you can use the key
:timestamp
:
data = %MySeries{}
data = %{ data | timestamp: 1439587926000000000 }
The timestamp is (by default) expected to be a nanosecond unix timestamp. To use different precision (for all points in this write operation!) you can change this value by modifying your write call:
data = %MySeries{}
data = %{ data | timestamp: 1439587926 }
data
|> MyApp.MyConnection.write([ async: true, precision: :second ])
Supported precision types are:
:hour
:minute
:second
:millisecond
:microsecond
:nanosecond
Please be aware that the UDP protocol writer does not support custom timestamp precisions. All UDP timestamps are implicitly expected to already be at nanosecond precision.