InfluxDB line protocol ingestion makes it easy for existing InfluxDB users to try QuestDB by only changing the address they send data to.
It is not necessary to create a table schema beforehand: the table will be created on the fly. If new columns are added, the table is automatically updated to reflect the new structure.
Using line protocol
|Name of the table where QuestDB will write data.|
|Array of string key-value pairs separated by commas that represent the reading's associated metadata|
|Array of key-value pairs separated by commas that represent the readings. The keys are string, values can be numeric or boolean|
|UNIX timestamp. By default in microseconds. Can be changed in the configuration|
- When the
table_namedoes not correspond to an existing table, QuestDB will create the table on the fly using the name provided. Column types will be automatically recognized and assigned based on the data.
timestampcolumn is automatically created as designated timestamp with the partition strategy set to
NONE. If you would like to define a partition stragey, you should CREATE the table beforehand.
- When the timestamp is empty, QuestDB will use the server timestamp.
Let's assume the following data:
Line protocol to insert this data in the
readings table would look like this:
There are only 2 spaces in each line. First between the
Dealing with irregularly-structured data
QuestDB can support on-the-fly data structure changes with minimal overhead. Should users decide to send varying quantities of readings or metadata tags for different entries, QuestDB will adapt on the fly.
InfluxDB line protocol makes it possible to send data under different shapes. Each new entry may contain certain metadata tags or readings, and others not. Whilst the example just above highlights structured data, it is possible for InfluxDB line protocol users to send data as follows.
Note that on the third line,
- a new
tagis added: "make"
- a new
fieldis added: "humidity"
After writing two entries, the data would look like this
The third entry would result in the following table
Adding columns on the fly is no issue for QuestDB. New columns will be created in the affected partitions, and only populated if they contain values. Whilst we offer this function for flexibility. However, we recommend that users try to minimise structural changes to maintain operational simplicity.
The TCP receiver can handle both single and multi-row write requests. It is fully multi-threaded and customizable. It can work from the common worker pool or out of dedicated threads. A load balancing mechanism dynamically assigns work between the threads.
By default, QuestDB listens to line protocol packets over TCP on
If you are running QuestDB with Docker, you will need to map port 9009 using
-p 9009:9009 --net=host. This port can be customised.
The TCP receiver uses two types of threads.
- Worker threads - write data to the different tables. Each worker is writing to designated tables. The worker-table repartition is modified over time by the load balancing jobs.
- Network IO thread - receives messages from the network and adds them in a queue for the writers. The network IO thread does not have a dedicated thread. Instead, it shares a common thread with the least busy worker.
The workflow is as follows.
The network IO thread receives write requests and sets up a queue for the workers. Workers pick up write requests for their assigned tables and insert the data.
A load balancing job reassigns work between threads in order to relieve the busiest threads and maintain high ingestion speed. It can be triggered in two ways.
- After a certain number of updates per table
- After a certain amount of time has passed
Once either is met, QuestDB will calculate a load ratio as the number of writes by the busiest thread divided by the number of writes in the least busy thread. If this ratio is above the threshold, the table with the least writes in the busiest worker thread will be reassigned to the least busy worker thread.
Uncommitted rows are committed either:
line.tcp.maintenance.job.hysterisis.in.msmilliseconds have passed
- once reaching
The TCP receiver configuration can be completely customised using configuration keys. You can use this to configure the tread pool, buffer and queue sizes, receiver IP address and port, load balancing etc.
The UDP receiver can handle both single and multi row write requests. It is currently single-threaded, and performs both network IO and write jobs out of one thread. The UDP worker thread can work either on its own thread or use the common thread pool. It supports both multicast and unicast.
Find an example of how to use this in the InfluxDB sender library section.
By default, QuestDB listens for
multicast line protocol packets over UDP on
22.214.171.124:9009. If you are running QuestDB with Docker, you will need to map
port 9009 using
-p 9009:9009 --net=host and publish multicast packets with TTL
of at least 2. This port can be customised, and you can also configure QuestDB
to listen for
Uncommitted rows are committed either:
- after receiving a number of continuous messages equal to
- when messages are no longer being received
The UDP receiver configuration can be completely customised using configuration keys. You can use this to configure the IP address and port the receiver binds to, commit rates, buffer size, whether it should run on a separate thread etc.