update_s3_acls silently skipped instrumentals/vocals because no s3_key
resolution existed for those file keys. Added generic fallback to
construct s3_path/file_key for any unhandled key.
Added 11 tests: unit (ACL parity across visibility levels, file_keys
membership, update_s3_acls call count), integration (visibility change
propagates to karaoke ACLs), functional (watch JSON and content page
karaoke URL embedding).
The controls bar was overflowing — "Next ▶" wrapped vertically. Switch
from single-row grid-auto-flow to a 3-column grid so items split into
two rows: toggles (Fresh/Reverse/Autoplay) on top, action buttons
(Karaoke/Random/Next) below. Reduce gap from 16px to 6×12px.
Moves vocal isolation track switching from the pop-out player into the
ring header controls so users can cycle Original/Instrumentals/Vocals
while watching. Preserves playback position on track switch, resets to
Original on SPA navigation, and adds K keyboard shortcut.
The upload endpoint streams the request body in chunks without buffering,
so raw_body stays empty on the server side. The HMAC must sign against an
empty body string to match. _sign_from_disk (which hashed file contents)
produced signatures the server could never verify — every upload got 401.
The inline path loaded entire base64 payloads into API memory via
Plug.Parsers, then decoded them again for UploadStore — 4 concurrent
45MB files consumed ~420MB on a 768MB droplet, pushing BEAM into swap
and timing out all RPCs.
Now uses POST /upload (streaming 64KB chunks, constant memory) then
references upload_ids in the execute call. The JSON body drops from
~60MB to ~500 bytes. Zero base64 encoding on the request side.
process_karaoke held 4+ copies of every media file in memory
simultaneously (raw bytes, base64, JSON serialization). With 14-16
concurrent workers on large files, memory exploded past 3.2GB RSS.
Replace all in-memory buffers with a disk-backed pipeline:
- Stream S3 download to tmpfile instead of .read()
- Build JSON payload on disk with streaming base64 encoding
- Incremental HMAC-SHA256 signing from disk
- Stream HTTP request body from file, response to file
- Decode artifacts to tmpfiles one at a time for S3 upload
Peak request-side memory drops from ~1GB/worker to ~64KB/worker.
Function signature unchanged — all callers work without modification.
The grandchild process closes all inherited fds (3..maxfd) to detach
from uWSGI. session_factory().get_bind() returned a session bound to
the now-dead SQLite fd. Capture the DB URL before forking, create a
fresh engine in the child.
4 concurrent workers was overwhelming the unsandbox API (429/502).
Cap at 1 worker for now. Also increase stderr truncation from 500
to 2000 chars to see full ffmpeg errors.
Single fork left the child in the uWSGI process group with inherited
HTTP sockets, causing 502s. Double-fork with setsid() and fd cleanup
so the grandchild is fully orphaned to init.
uWSGI kills worker processes after ~60-90s ("NO MERCY!!!"), which
kills all daemon threads including the backfill. Fork a child process
that's independent of the worker — survives parent SIGKILL.
Replace threading.Lock + set guard with fcntl.flock on a lockfile
(/tmp/karaoke_backfill_{shop_id}.lock) for one-per-shop guard that
auto-releases on crash. Also add error logging for unsandbox execute
failures (was previously silent).
SVG bar charts for 28-day daily views, CSS horizontal bars for
traffic sources and device split, inline mini bars in top products
table. Zero-dep, server-side rendered, theme-aware, no JS.
add is_owner flag to PageSession — set via data attribute when
the viewer can_edit_shop. beacon stores the flag, view_count
skips increment for owners, and all analytics queries filter
owner sessions out of funnel metrics. owner sessions are still
recorded for separate analysis. existing NULL data treated as
non-owner.
Keys can have tiers changed or expire mid-run. Now:
- process_karaoke validates keys before downloading from S3
- backfill aborts cleanly if keys are invalid instead of defaulting to 1 worker
boto3 TransferManager spawns its own ThreadPoolExecutor internally,
which crashes with 'cannot schedule new futures after interpreter
shutdown' when uWSGI recycles the worker process.
Per-shop unsandbox API keys in shop settings. On upload, audio/video
products are processed through voxsplit in a zerotrust container (no
network, destroyed after use) producing both instrumentals and vocals
tracks. Background backfill with ThreadPoolExecutor sized to account
concurrency, exponential backoff retries. Player cycles Original →
Instrumentals → Vocals with toggle button or K shortcut.
Chrome blocks external SVG filter references via CSS url(). Replace with
native CSS filter functions (grayscale + sepia + hue-rotate + saturate)
that work in all browsers. Delete the now-unused SVG file.
Inline SVG filter url(#id) fails when the SVG defs are inside the
filtered element (both on <html> and <body>). Data URI SVG filters
are blocked by Chrome. External SVG file at /static/svg/color-filters.svg
avoids both issues — no DOM circular reference, no security block.
Browser fetches the SVG once and caches it.
Data URI SVG filters don't work in Chrome (blocked for security).
Inline url(#id) on <html> fails due to root-element rendering order.
Fix: keep data-color-filter attribute on <html>, apply filter to <body>
via descendant selector. SVG filter defs in <body> with <defs> wrapper
resolve correctly since DOM is parsed before CSS filter is applied.
CSS filter: url(#id) on <html> fails because the SVG filter definitions
are children of the element being filtered (circular reference). Grayscale
worked because it uses native CSS filter, not SVG. Now all feColorMatrix
filters are self-contained data URIs in CSS — no DOM SVG block needed.
- Add shuffle button next to Next in SPA ring navigation
- Expand grayscale mode to full color filter system (off, grayscale, R, G, B, R+G, R+B, G+B)
using SVG feColorMatrix filters for selective channel isolation
- Move comment email field below textarea, fix email input theming for dark mode
Server renders data-grayscale="on" on <html> so it works without JS.
Toggle in Ribbon & Layout settings, off by default. Original assets
untouched — purely presentational via filter: grayscale(100%).
The .related-content-overflow display:none rule was being overridden by
.related-content-row display:grid later in the file (equal specificity,
cascade order wins). Bump specificity with .related-content-row.related-content-overflow.
