Goals

Enhance vector, based upon our past experience and feedback from the field, to:

• to more gracefully handle backpressure due to slow or blocked sinks
• more efficiently discover container logs in order to improve throughtput
• make better use of the resources (e.g. CPU, memory) its given

Motivation

There are several limitations of vectors current design that results in higher cpu and memory consumption or decreased throughput. We strive to improve its capabilities based upon our past experiences with deploying both fluentd and the current design of vector.

Acceptance Criteria

Measured improvement (or no regression) on:

• CPU use
• Memory use
• Log collection rate

These should be measured in several different tests.

• Many containers (100s or 1000s), long-lived (>10 min), moderate log rates (10 - 100 lines/sec)
• Many containers, short-lived (< 1min), high log rates (> 100 lines/sec)
• Many containers, mixed lifespan/rate.
• Single container, logging at max speed.

Risk and Assumptions

Documentation Considerations

None, performance only.