[alex289]

Type of change

• 🐛 Bug fix
• 🚀 New feature
• ❓ Other (please specify)

Description

Part 2/4 of #268. Stores every change of an application's health so availability can be computed over time.

• New application_health_changes table (migration 000036): status, optional reason, the failing services (JSON) and changed_at. It is deleted along with its application.
• A row is only written when the status differs from the application's latest row, using the reports agents already send. Reports for another agent's application are still ignored and not recorded.
• Agent offline: a disconnect records unknown with reason agent_offline. At hub startup each agent is marked offline from its last_seen time, so hub downtime shows up as a gap instead of the last status silently continuing.
• Deploys and image updates don't write history. The previous status holds until the agent reports the result, as agreed in the issue.
• Retention: a daily pruner deletes changes older than 30 days. It keeps the latest change before the cutoff per application, so the 30d range still starts in the right state.
• Timestamps are stored in UTC, because SQLite compares them as text.

[codecov]

Codecov Report

❌ Patch coverage is 82.19178% with 26 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
.../internal/hub/models/application_health_changes.go	0.00%	10 Missing ⚠️
...ackend/internal/hub/healthhistory/healthhistory.go	90.12%	8 Missing ⚠️
backend/internal/hub/websocket/handler.go	84.61%	6 Missing ⚠️
backend/internal/hub/server.go	81.81%	2 Missing ⚠️

📢 Thoughts on this report? Let us know!

[timokoessler]

Interesting query :)
Not sure if this is really easier or which one is better:

Where(`changed_at < ? AND EXISTS (
    SELECT 1 FROM application_health_changes newer
    WHERE newer.application_id = application_health_changes.application_id
      AND newer.changed_at < ?
      AND (newer.changed_at, newer.id) > (application_health_changes.changed_at, application_health_changes.id)
)`, cutoff, cutoff)

[alex289]

Both delete the same rows. I kept the window function because it says directly "keep the latest row before the cutoff per app". Speed doesn't matter much either way: after the first prune, only about a day of rows plus one row per app are older than the cutoff. Happy to switch to the EXISTS version if you prefer it.