Freegle / Iznik / 24536

72%

push

circleci

fix(digest): bound getPostsForUser load so the daily digest can't OOM (DIGEST_LOAD_CAP)

The daily digest (mail:digest:unified --mode=daily) streams recipients with lazyById(500),
but getPostsForUser() loaded the ENTIRE backlog since a member's cursor in one unbounded
->get() with attachments/fromUser/groups eager-loaded, then filtered/scored/dedup'd in PHP
down to the 65 it renders. When the run fell behind, each stuck member's cursor went stale,
so the window grew to days × groups (worse after the rippling full-enable added groups per
member), the eager-loaded Collection blew the 1GB PHP memory_limit
("Allowed memory size ... exhausted" at Connection.php fetchAll), and the run died part-way
— leaving higher-id members unreached, their cursors stale, and the next window even bigger
(self-amplifying). Prod symptom: daily sends collapsed 38.6k/day (06-28) -> 12 (07-01);
~44k of 78k daily members stuck with lastsent before 06-29.

Cap the load at DIGEST_LOAD_CAP (500). Oldest-first + limit means a backlogged member drains
500 posts per run (updateDigestTracker already advances the cursor past exactly what's loaded)
until caught up — bounded per-member memory, no OOM, no posts lost. Normal daily volume is
well under the cap so steady-state digests are unchanged. Not a sharding problem; the per-
member load itself is now bounded. Verified end-to-end on a stuck member (1 sent, 0 errors).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_015DQRL5iGD71jNTgyokMpt9

12262 of 16215 branches covered (75.62%)

Branch coverage included in aggregate %.

1 of 1 new or added line in 1 file covered. (100.0%)

130608 of 183099 relevant lines covered (71.33%)

37.82 hits per line