ammocpr/lib/ammoprices/scraping/scrape_job.ex
Graham McIntire 2efb35dc48
Stop scrape chain from dying on a single bad product
A scraped product with nil price_per_round_cents made Prices.create_snapshot
return {:error, changeset}, which blew up the {:ok, _} match in the runner,
crashed the whole perform, and after 3 retries discarded the job. Nothing
reseeds the chain once the job is discarded, so prod scrapes silently died
on Mar 17 and stayed dead until the app was restarted.

- Runner handles snapshot insert errors by logging and continuing instead
  of raising.
- ScrapeJob wraps each Runner.run in try/rescue so one retailer crashing
  doesn't take down the whole cycle.
- ScrapeJob moves schedule_next into an after block so the chain is always
  reseeded, even if the scrape loop crashes.
2026-04-22 15:51:15 -05:00

82 lines
2.2 KiB
Elixir

defmodule Ammoprices.Scraping.ScrapeJob do
@moduledoc false
use Oban.Worker,
queue: :scraping,
max_attempts: 3,
unique: [period: :infinity, states: [:available, :scheduled, :executing, :retryable]]
alias Ammoprices.Catalog
alias Ammoprices.Scraping.Runner
require Logger
@scrapers [
Ammoprices.Scraping.Retailers.LuckyGunner,
Ammoprices.Scraping.Retailers.SgAmmo,
Ammoprices.Scraping.Retailers.TargetSportsUsa,
Ammoprices.Scraping.Retailers.TrueShotAmmo,
Ammoprices.Scraping.Retailers.PalmettoStateArmory,
Ammoprices.Scraping.Retailers.BulkAmmo,
Ammoprices.Scraping.Retailers.Natchez
]
# Products not seen for this many days are hard-deleted after each scrape cycle.
@stale_product_days 30
@impl Oban.Worker
def perform(_job) do
calibers = Catalog.list_calibers()
for scraper <- @scrapers, caliber <- calibers do
scrape_delay()
safe_run(scraper, caliber)
end
prune_stale_products()
Phoenix.PubSub.broadcast(Ammoprices.PubSub, "prices:updated", {:prices_updated, %{}})
:ok
after
# Always reseed the chain, even if something above crashed. Without this,
# a single unhandled error kills the chain until the app is restarted.
schedule_next()
end
defp safe_run(scraper, caliber) do
Runner.run(scraper, caliber)
rescue
error ->
Logger.error(
"Scrape crashed for #{inspect(scraper)} / #{caliber.slug}: " <>
Exception.format(:error, error, __STACKTRACE__)
)
{:error, :crashed}
end
defp prune_stale_products do
cutoff = DateTime.add(DateTime.utc_now(), -@stale_product_days, :day)
Catalog.delete_stale_products(cutoff)
end
@doc false
def schedule_next do
{min, max} = Application.get_env(:ammoprices, :scrape_interval_seconds, {25_200, 43_200})
delay_seconds = Enum.random(min..max)
%{}
|> new(scheduled_at: DateTime.add(DateTime.utc_now(), delay_seconds, :second))
|> Oban.insert()
end
# Random delay between requests: 10-20 minutes in prod, 0 in test
defp scrape_delay do
{min, max} = Application.get_env(:ammoprices, :scrape_delay_ms, {600_000, 1_200_000})
delay = Enum.random(min..max)
if delay > 0 do
Process.sleep(delay)
end
end
end