Skip to main content
Version: v4.24

buffer

An in-memory buffer to improve handling of data spikes in upstream operators.

buffer [capacity:int, policy=string]

Description​

The buffer operator buffers up to the specified number of events or bytes in an in-memory buffer.

By default, operators in a pipeline run only when their downstream operators want to receive input. This mechanism is called back pressure. The buffer operator effectively breaks back pressure by storing up to the specified number of events in memory, always requesting more input, which allows upstream operators to run uninterruptedly even in case the downstream operators of the buffer are unable to keep up. This allows pipelines to handle data spikes more easily.

capacity: int (optional)​

The number of events or bytes that may be kept at most in the buffer.

Note that every operator already buffers up to 254Ki events before it starts applying back pressure. Smaller buffers may decrease performance.

policy = string (optional)​

Specifies what the operator does when the buffer runs full.

  • "drop": Drop events that do not fit into the buffer. This policy is not supported for bytes inputs.
  • "block": Use back pressure to slow down upstream operators.

When buffering events, this option defaults to "block" for pipelines visible on the overview page on app.tenzir.com, and to "drop" otherwise. When buffering bytes, this option always defaults to "block".

Examples​

Buffer up to 10 million events or bytes​

Buffer and drop events if downstream cannot keep up:

buffer 10M, policy="drop"