Files
meta-feeder-arxiv/feeder-plugin/arxiv-feeder
Manielle ebc7e1da65
Publish image / publish (push) Failing after 13s
give every record a url-locator CID so the gateway can persist it
A record with no CID cannot be persisted by the gateway ("search-persist: record
carries no CID; not persisted"), yet the gateway still stamps the query as
covered for an hour (META_GATEWAY_SEARCH_COVERAGE_TTL_SECS) and, inside that
window, skips this feeder and answers from what it persisted. So the first query
returned papers and every identical one for the next hour returned none.
Measured on the live box: 40 hits, then 0 hits in 1.5 s with no fan-out.

Each record now carries exactly one `cids/` member: a 0x1006 url-locator over the
canonical, versioned PDF URL (https://arxiv.org/pdf/<id>.pdf). This is the fleet
convention (Gutenberg, internetarchive, jamendo): it is a pure function of the
URL, so it costs no I/O, and it doubles as the byte path that meta-share
redeems. The versioned id means v2 of a paper is a different CID. Built from the
canonical URL rather than the feed's link, which arXiv emits as http:// without
.pdf.

Verified live after deploying: 40 hits each carrying a CID; the same query again
inside its coverage window returns 40 (was 0); meta-share serves the locator
with 206 Partial Content and the bytes are a real PDF (%PDF-1.7, 1,182,901 B).
2026-10-11 17:02:17 +02:00
..
2026-08-24 19:13:08 +02:00