Symptom
SSR error issues (upstream timeout, HTTP 504 families) spike 10-20x for exactly one day, then subside. Looks like a deploy regression or an infra incident.
Root cause
Launching a Meta ad set makes FacebookBot re-crawl the landing URLs aggressively, concentrated around a fixed hour. On a small/cold SSR stack (e.g. Render + SvelteKit SSR proxying an API), that synchronized crawl trips upstream timeouts. The spike is load composition, not code.
One-call diagnosis
Sample 50-100 events from the spike window and Counter two tags:
- url -> concentrated on the exact landing pages of the just-launched ad cells (our case: 45/49 events on the 2 new campaign views)
- browser.name -> dominated by FacebookBot
Cross-reference against the ad-ops record of which cells went live that day. If the burst URLs are the new cells and the UA is FacebookBot, close it as campaign crawl load.
Consequences
- Expect one such spike at every campaign launch; write the SSR family escalation criterion as a sustained rate (e.g. >20/day for 2 consecutive days), never a single-day burst.
- The same worker death can appear in two Sentry projects at once (backend WORKER TIMEOUT and frontend SSR 504 with same-second timestamps) - correlate timestamps before filing two incidents.