mirror of
https://github.com/hanzoai/insights.git
synced 2026-10-30 16:15:31 +00:00
zeekay
0e85299fb4
Two things, both found by following /_health returning 500 while the site
served 200.
SUBSCRIPTIONS. The exports delivery path imported ee.tasks.subscriptions,
which this fork does not carry, so scheduled subscription reports could not
import at all. Ported the five modules into products/exports/backend/
subscriptions, where the product that owns subscriptions lives.
It is a debrand, not a rewrite: upstream already reaches Slack through
SlackIntegration from the models module — the same wrapper stamphog,
customer_analytics and conversations use — so this rides our own Slack app the
moment the import resolves. build_explore_hint was already ported natively;
the rest was posthog.* -> insights.* and the ee path -> the new home. Every
import in the package resolves, and the seven third-party ones are present in
the shipped image (checked against a running pod).
MIGRATIONS. /_health builds the migration graph, and the graph could not be
built: insights.0002_managed_tables depended on ("workflows", "0001_initial"),
a node that does not exist — workflows names its first migration for what it
created. Fixing that exposed the real one: insights was squashed to four
migrations with no `replaces`, so 213 dependencies across 212 product
migrations still named 1xxx_ nodes that are gone.
Each now depends on 0001_initial, which is the squash that contains the schema
they were waiting for, so ordering is unchanged. Zero dangling nodes remain in
any app. This was not only a broken health check: with an unbuildable graph no
future migration could be applied at all.