What an operator has to do after deploying a change, and why. Newest first. Code-only changes are not listed — only the ones that need action outside the repository.
Real account handles, hostnames and tokens belong in the mounted
riptide.json / team-keys Secret, never in this file: the repository is public.
Add the accounts nobody declares to the production riptide.json.
Detection now runs in this order: configured automation authors → the git
host's own verdict (type: SERVICE) → the *-bot name shape → identities a
sender declares about itself. Only accounts that reach none of those need a
config entry.
- Renovate (or any dependency bot) — needs an entry. Its account is an
ordinary user to Bitbucket, and it POSTs nothing to riptide, so nothing
declares it. Add both the login handle and the display name to
automation.renovate.authors; matching is case-insensitive and covers either field. Without it, only itsrenovate/branches are recognised and its pushes to the integration branch count as human activity. - The Bitbucket system user — no entry needed. The server marks it
type: SERVICEand riptide tags itservice-accounton its own. - The CI account — no entry needed once the pipeline sender reports
actor_handle; see below.
Rows already ingested keep the is_automated they were written with — the
tables are append-only. Read-time queries filter historical rows through the
non_human_identities CTE in the README's pickup-time query.
Redeploy the collector. Anything older than the draft→ready parser fix
stores raw pr:modified rows and emits no pr:ready_for_review, so PRs opened
as drafts have no clock-start for review pickup time. Verify after rollout:
pr:ready_for_review rows appear and no-op pr:modified rows stop.
Have CI senders report image_ref. The full image reference
(registry/path:tag), which Argo CD stores verbatim in payload->'images'.
This is the only exact link from a deploy back to its build and commit: image
tags are versions, not commit SHAs, and argocd_events.revision is the
GitOps-repo SHA. Until senders report it, lead time and per-ticket flow run on
the release-note fallback, whose coverage decays as releases move to direct
version-bump commits. See
Correlating deploys back to commits.
Have CI senders report actor_handle (+ actor_account_kind: service) —
the git account the CI system commits and pushes as. A CI account can author a
third of all repository events; declared, they leave human-activity metrics
without anyone naming the account in config.
Both fields are optional and additive; senders that omit them keep working.
Deploy riptide before noergler. The rollup schema rejects unknown fields,
so a collector without reviewer_handle answers every rollup with HTTP 422,
not just the ones carrying a new field — and a rejected rollup is never
retried, because the PR is marked as emitted when it is claimed. Anything
closed in that window is lost.
Once both sides are deployed, noergler declares the account it reviews under, and riptide filters its comments out of code-review pickup time without any per-installation configuration.
0003 adds nullable columns (author_display_name, image_ref,
actor_handle / actor_account_kind, reviewer_handle /
reviewer_account_kind); 0004 adds the lead-time views and one index. Both
are applied by the init container on deploy — no manual step, no downtime, and
downgrade is clean if a rollback is needed.