Nomad Autoscaler 0.5.0 Released

Nomad
1

Dear All,

We are excited to announce the release of Nomad Autoscaler 0.5.0. This release includes the following changes.

Features:

  • InfluxDB APM plugin — A new plugin lets you use InfluxDB 1.x as a metrics source for autoscaling decisions, with support for InfluxQL queries, basic auth, and JWT Bearer authentication.

  • Scheduled scaling windows — You can now define schedule blocks on policies and checks to control when evaluations take effect, using cron expressions with a start time and end time or duration.

  • Prometheus instant queries — Prometheus checks now support query_window = "instant", allowing point-in-time metric lookups instead of range queries where appropriate.

Improvements:

  • Multi-namespace support — The nomad.namespace config and -nomad-namespace flag now accept multiple namespaces, or * to monitor all namespaces at once.

  • Fewer redundant metric fetches — If multiple checks in a policy share the same APM source and query, the result is now fetched once and reused, reducing load on your metrics backend.

  • Scale through node drains — The autoscaler can now continue making scaling decisions even when nodes are being drained, controlled via node_filter_ignore_drain.

Bug Fixes:

  • AWS ASG premature drain — Fixed a bug where nodes were drained before checking ASG constraints (e.g. minimum size, suspended termination), which could cause unnecessary capacity loss.Fixed AWS-ASG plugin draining nodes before validating ASG constraints.

  • Autoscaler crash on bad strategy config — A misconfigured strategy check no longer crashes the autoscaler; it now correctly follows the on_error / on_check_error behavior.

  • Fixed-value strategy no longer requires APM config — Checks using a fixed target value can now be defined without needing an APM source or query.

  • Scale-in respected requested count — Fixed a bug where scale-in could select more nodes than requested for removal.

  • Ineligible nodes excluded from capacity — Nodes that are ready but marked ineligible for scheduling were being counted as available capacity, causing the autoscaler to underestimate the need to scale out. They are now correctly excluded.

  • (ENT) DAS vertical scaling panic — Fixed a panic in Dynamic Application Sizing vertical checks that occurred when strategy errors were configured to be ignored.

Please report any issues found to GitHub or this list.

As usual, a shout-out to the community for bug reports and PRs.

Thank you,

The Nomad Team

Links:

1 Like