Two jobs on the same shared asset (same video+format queued twice, or by two
users) starved the worker pool: worker A downloads the asset; worker B claims the
sibling job, finds the asset `downloading`, requeues it, sleep(2)s, and re-claims
it on the next tick — burning a whole worker slot behind one in-flight asset for
the entire download (at concurrency=2, the queue stalls).
Fix: the claim SQL now excludes any job whose asset is already `downloading`
(`NOT EXISTS … media_assets.status='downloading'`), so a competing sibling simply
waits until the asset settles and the workers pick up unrelated jobs meanwhile.
The `_process_job` busy/lost-race branches drop their `time.sleep(2)` — the
exclusion keeps a requeued job out of the claim window, so there's no spin.
TDD: the new test failed on the old claim (claimed the sibling), passes now.
Full lane green, ruff clean.
- Floor the effective max_concurrent at 1 on the read path (quota.resolve)
and in the claim SQL (GREATEST(..., 1)). admin_set_quota already floors on
write and the sysconfig default's min is 1, but a stray stored 0 (direct DB
edit) would diverge the two "at limit" checks — the SQL claim-window
exclusion can't see a user "capped" with zero running jobs, while
at_concurrency_limit can — reopening the starvation the fairness fix closes.
Keeping the effective limit >= 1 makes the two checks provably agree.
- shared_with_me now orders by (created_at DESC, id DESC) to match
list_downloads, so the newly-added 500 cap has a stable boundary among
same-timestamp shares instead of a non-deterministic one.
- 0062 upgrade loop: rename the unused loop var "table" to "_table"
(matching the underscore-marked "_col").
Test: a stored max_concurrent=0 flooder no longer wedges the claim loop.
#2 — _claim_job excluded users at their concurrency limit only in Python,
AFTER a fixed LIMIT-50 fetch window. One user flooding the queue with >50
queued jobs while at their cap filled the whole window, chosen came back None,
and every other user's work past position 50 was never seen. Exclude capped
users IN THE SQL (per-user max_concurrent from the quota row, else the
sysconfig default; unlimited users never capped), keeping the Python re-check
as a race guard that can only reject, never over-claim.
#3 — list_downloads and shared_with_me loaded every job for a user with no
bound. Cap both at 500 newest-first (a shared _MAX_LIST, matching the admin
listing's existing cap); the Download Center renders newest-first and 500 is
far beyond any real manual-download history, so no frontend change is needed.
Tests: DB-backed — a capped user's 60-job flood no longer starves a user
behind the window; oldest-by-queue_pos claim, all-capped->None, unlimited
never excluded; and the list cap bounds the result set.
peter