Transform data at rest
This feature is currently only available on the command line using the
tenzir-ctl binary. We're working on bringing it back as an operator so that
you can also use it from the app.
Tenzir provdides several features to transform historical data at a node.
Delete old data when reaching storage quota
The disk-monitoring feature enables periodic deletion of events based on utilized disk storage. To limit the disk space used by a node, configure a disk quota:
Whenever a node detects that its database has exceeded the configured quota, it
will erase the oldest data. You can specify a corridor for the disk space usage
by additionally providing the option
--disk-quota-low. This can be used to
avoid running permanently at the upper limit and to instad batch the deletion
The full set of available options looks like this:
# Triggers removal of old data when the DB dir exceeds the disk budget.
# When the DB dir exceeds the budget, Tenzir erases data until the directory
# size falls below this value.
# Seconds between successive disk space checks.
When using this method, we recommend placing the log file outside of the database directory. It counts towards the size calculations, but cannot be automatically deleted during a deletion cycle.
Transform old data when reaching storage quota
Instead of deleting data periodically, a node can also trigger spatial compaction when exceeding a given disk budget. A spatial compaction cycle transforms data until disk usage falls below the budget, e.g., by removing columns or rows from certain events, or by deleting them entirely.
When the disk budget exceeds the configured threshold, the node decides what data to compact. The compaction mode defines how this happens. Currently, there exists only one mode: weighted age.
This compaction mode selects all events according to a weighted age. To compute the weighted age, the node divides the actual age of an event with the weight assigned to this event type. For example, applying a weight of 100 to an event that is 100 days old would yield a weighted age of 1 day. This causes it to be transformed after events that are 50 days old. Conversely, a weights less than one results in an older weighted age, resulting in earlier consideration in a compaction cycle.
The default weight is 1 for all event types. Here is an example configuration that adjusts the weights:
interval: 6 hours
- weight: 0.1
- weight: 100
pipeline key for each type is optional. If present, the corresponding
pipeline processes all matching events. If absent, the nodes deletes matching
Two additional keys are useful to fine-tune the behavior of the compaction plugin:
compaction.space.scan-binary: an absolute path to a binary that should be executed to determine the current disk usage
compaction.space.step-size: adjust how many compaction candidates should be processed before re-checking the size of the database directory
Transform data after exceeding a retention span
A node triggers temporal compaction according to a set of rules that define how to transform events after they reach a specfic age. This declarative specification makes it easy to express fine-grained data retention policies, which is often needed for regulatory requirements and compliance.
For each compaction cycle, the node processes all rules and identifies what subset of the data has become subject to transformation. To this end, each rule defines a minimum age, i.e., a lower bound that must be exceeded before the corresponding events undergo their configured pipeline.
To configure temporal compaction, provide a list of compaction rules under the
plugins.compaction.time in the configuration. A compaction rule defines
the minimum age using key
after, the pipeline to apply with the key
pipeline, the scope in terms of schema using the key
types, and a name to
uniquely refer to the rule. Omitting the
types key causes temporal compaction
rules to be applied to all schemas.
By default, a compaction rule consumes its input, i.e., it erases the original
events from the database and replaces them with the transformed events. The
preserve-input option can be specified on a temporal compaction rule to
override this behavior and to keep the input partitions available.
A node applies each rule only once per partition and stores the applied rule name within the partition meta data. If you rename a rule in the configuration and reload a new compaction configuration, already compacted partitions will undergo another round of compaction.
The pipelines referenced in the compaction configuration must be defined in your configuration.
# How often to check the `after` condition below.
interval: 1 day
- after: 2 days
- after: 7 days
Trigger a compaction cycle manually
You can also interact with the compaction plugin on the command line, through
compaction subcommand. Use the
list subcommand to show all configured
tenzir-ctl compaction list
You can then trigger a compaction manually via
tenzir-ctl compaction run <rule>
status subcommand for an overview of the current status of the
tenzir-ctl status compaction