Freegle / Iznik / 24186

72%

Build Type

push

circleci

Committed by invalid-email-address

Commit Message

feat(rippling): ripple:backfill command + --recompute upgrade with co-located reach reuse

Seed rippling reach for the ~34k live posts that predate go-live (their
messages_spatial.arrival is before freegle.ripple.enabled_at, so initialiseNew
never picked them up and they would never ripple).

ripple:backfill (new command): drains the anti-join (messages_spatial with no
rippling_reach row) by reusing ExpandService::initialiseNew with the arrival
cutoff lifted, so a backfilled row is identical to a natively-initialised one.
Progress bar, --dry-run, --limit, idempotent/resumable, one row per write
(Galera-safe). Shardable (--shards N --shard i on msgid % N), each shard on its
OWN Cache lock (never ripple:expand:run) so parallel instances drain disjoint
slices; total routing load ~= shards x compute_concurrency.

Because seeding computes reach inline on the routing server (dense isochrones
14-29s), a full drain is many hours. So this supports a TWO-PHASE backfill:
  1. a cheap geometry pass seeds every post a placeholder ("DPA") reach = its
     group area (status='stopped', schedule NULL) so nothing is reply-dark now;
  2. --recompute upgrades those placeholders to real routed reach IN PLACE via
     upsert (INSERT ... ON DUPLICATE KEY UPDATE) — the post is never momentarily
     without a reach row. Recompute selects the placeholders (status='stopped'
     AND schedule NULL) instead of the anti-join, skips the arrival cutoff and
     the reply-saturation stop (so the drain terminates), and retires genuinely
     off-graph placeholders to 'done' (keeping their DPA polygon) so the loop
     cannot re-select them forever.

Reach reuse (freegle.ripple.reuse_reach, default on): reach is a deterministic
function of the blurred origin (4dp) + global ripple config, so a post at the
same blurred origin as an already-computed reach copies its schedule instead of
re-routing. Covers "same user posting again from home" and any co-located posts;
removes the bulk of r... (continued)

Coverage Stats

11339 of 13061 branches covered (86.82%)

Branch coverage included in aggregate %.

180 of 204 new or added lines in 2 files covered. (88.24%)

1330 existing lines in 41 files now uncovered.

128495 of 181491 relevant lines covered (70.8%)

36.58 hits per line