1.0.0-beta.12

Fixed

• Log much less from falconeri_worker by default, and make it configurable.
  This fixes an issue where the newer tracing code was causing the worker to
  log far too much.

1.0.0-beta.10

Fixed

• Prevent key constraint error when retrying failed datums (Issue
  #33). But see Issue
  #36; we still don't do the
  right thing when output files are randomly named.
• Reduce odds of birthday paradox collision when naming jobs (Issue
  #35).

1.0.0-beta.9

Fixed

• Hard-code PostgreSQL version to prevent it from getting accidentally upgraded
  by Kubernetes.

1.0.0-beta.8

Fixed

• Use correct file name to upload release assets (again).

1.0.0-beta.7

Fixed

• Use correct file name to upload release assets.

1.0.0-beta.6

Fixed

• Attempted to fix binary builds on Linux (yet again).

1.0.0-beta.5

Fixed

• Attempted to fix binary builds on Linux (again).

1.0.0-beta.4

Fixed

• Attempted to fix binary builds on Linux. Not even trying on the Mac.

v1.0.0-beta.2: Fix `job_timeout`

Fixed

• Fix job_timeout conversion to ttlActiveSeconds in the Kubernetes YAML.

v1.0.0-beta.1: Add "babysitter" and `datum_tries` support

This release adds a "babysitter" process inside each falconerid. We use this to monitor jobs and datums, and detect and/or recover from various types of errors. Updating an existing cluster should be fine, but it's likely to spend a minute or two detecting and marking problems with old jobs. So please exercise appropriate caution.

We plan to stabilize a falconeri 1.0 with approximately this feature set. It has been in production for years, and the babysitter was the last missing critical feature.

Added

• If worker pod disappears off the cluster while processing a datum, detect this and set the datum to status = Status::Error. This is handled automatically by a "babysitter" thread in falconerid.
• Add support for datum_tries in the pipeline JSON. Set this to 2, 3, etc., to automatically retry failed datums. This is also handled by the babysitter.
• Periodically check to see whether a job has finished without being correctly marked as such. This is mostly intended to clean up existing clusters.
• Periodically check to see whether a Kubernetes job has unexpectedly disappeared, and mark the corresponding falconeri job as having failed.
• Add trace spans for most low-level database access.

Fixed

• We now correctly update updated_at on all tables that have it.