Fix camera button permissions & ffmpeg process leak (#550)
Camera button on printer card was clickable without camera:view
permission. ffmpeg processes (~240MB each) accumulated after closing
camera streams because: (1) stop endpoint called terminate() without
wait()/kill(), (2) HTTP disconnect detection only ran between frames
so was blocked when the generator was stuck on stdout read, and
(3) no mechanism caught processes orphaned by generator abandonment
or app restarts.
- Add camera:view permission check + tooltip to camera button
- Fix stop endpoint: terminate() → wait(2s) → kill() → wait()
- Add background disconnect monitor (polls every 2s, kills ffmpeg
directly on disconnect)
- Add periodic /proc scan (every 60s) that SIGKILLs any ffmpeg
with rtsps://bblp: not in an active stream
- Add noCamera i18n key to all 6 locales
- Fix camera API test mocks for async wait() and pid attribute
The H2C dual nozzle variant reports model code O1C2 via MQTT, but only
O1C was recognized. This caused the camera to use the wrong protocol
(chamber image on port 6000 instead of RTSP on port 322), producing a
reconnect loop. Added O1C2 to all model ID maps across 8 files.
Some firmware versions send MQTT payloads with non-UTF-8 bytes.
UnicodeDecodeError was uncaught, silently dropping the entire message
and leaving printer status as "unknown" with 0°C temps and no AMS.
Fall back to decode(errors="replace") to keep JSON parseable.
The start_bambuddy.bat launcher had Unix (LF) line endings. When a
user's git is configured with core.autocrlf=false or input, the file
is checked out with LF endings and cmd.exe fails with "the syntax of
the command is incorrect" before reaching the hash check.
Add .gitattributes forcing CRLF for *.bat files so they always get
correct line endings on checkout regardless of git config.
The purple queue counter badge in the printer card header used the raw
unfiltered queue item count, so it appeared on ALL printers of the same
model when a job was scheduled for "any [model]" — even printers without
the matching filament color. The PrinterQueueWidget below it (which
shows "Clear Plate & Start") already filtered by filament type + color.
Apply the same filament compatibility filter (type check + color override
check) to the badge count so it only shows on printers that can actually
run the queued jobs.
cap_add: NET_BIND_SERVICE in docker-compose.yml doesn't reliably
propagate to the Python process when combined with the user: directive
(depends on ambient capability support in the container runtime).
Set the file capability directly on the Python binary via setcap in
the Dockerfile, which the kernel honors regardless of runtime config.
The chart X axis used dataMin/dataMax domain, compressing the view to
only the data present. A 6h view with 1.5h of data looked like 1.5h.
Pin the X axis to the full selected time range, pad data edges so the
line extends to both ends, and connect through nulls for a continuous
line.
When a print was queued to "any printer" with a filament color override
(e.g., white PETG), the "Clear Plate & Start Next" button appeared on
printers that had the right type but wrong color. The backend scheduler
already rejected color mismatches, but the frontend only checked filament
types. Now passes loaded type+color pairs from AMS status to the widget
and filters against filament_overrides, mirroring the backend's
_count_override_color_matches() logic.
The SQLite WAL was never checkpointed during container shutdown, leaving
uncommitted data (including schema migrations) in the -wal file. On
restart, partial WAL recovery could cause inconsistent query results,
making the queue appear empty until a filter was applied. Add
PRAGMA wal_checkpoint(TRUNCATE) and engine.dispose() to the lifespan
shutdown handler.
Long filament profile names were cut off because inline filament ID
codes consumed horizontal space in the dropdown. Remove the codes from
dropdown items (selected code still shown below the input) and widen
the modal from max-w-lg to max-w-xl.
The print scheduler had two code paths for queue items: model-based
(target_model set) and printer-specific (printer_id set). Only the
model-based path computed ams_mapping before starting the print. VP
queue items assigned to a specific printer went through the
printer-specific path with null ams_mapping, causing the printer to
default to the first AMS slot regardless of what the 3MF required.
Add _compute_ams_mapping_for_printer() call to the printer-specific
path, matching the existing model-based behavior.
When a virtual printer auto-queued a file, PrintQueueItem was created
without plate_id. The scheduler defaulted to plate_id=1, generating
MQTT path "Metadata/plate_1.gcode". For multi-plate 3MF files sliced
on a different plate, the printer couldn't find the gcode and returned
HMS error 0500_4003. Extract plate index from the 3MF's slice_info.config
before creating the queue item.
When scheduling a print to "Any {model}", a redundant "Target Model"
dropdown appeared even though the G-code is already sliced for a
specific printer model. Changing it would lead to print failures.
Hide the dropdown when slicedForModel is known — the tab label already
communicates the target. The dropdown still appears as a fallback for
legacy files without model metadata.
The "Clear Plate & Start Next" button appeared on ALL printers of a
model when a job was queued via "Queue to Any X", even printers without
the required filament loaded. Clicking it started a print that fails.
PrinterQueueWidget now filters queue items by filament compatibility —
each item's required_filament_types must all be present in the printer's
loaded filaments (AMS + external spools). If no compatible items exist,
the widget is hidden entirely. Matching is type-only (not color) and
case-insensitive, consistent with the backend scheduler. When filament
data is unavailable, no filtering is applied (graceful degradation).
Add weight_locked flag to spools that auto-sets when weight_used is
explicitly updated via the API. Both the MQTT AMS remain% auto-sync and
the manual force-sync endpoint skip locked spools. Usage tracker delta
tracking is unaffected. Users can re-enable AMS sync by setting
weight_locked to false.
Queue endpoints now return filament_type, filament_color, layer_height,
nozzle_diameter, and sliced_for_model from the archive or library file.
Previously this data was only available via the archive API.
Virtual printer "print_queue" mode created queue items with no printer
assignment. Now sets target_model from the VP's SSDP model code (e.g.
P1S, X1C) for "Any Printer" scheduling, or uses target_printer_id if
configured.
Settings page auto-save onSuccess overwrote localSettings with the
server response, discarding characters typed during the save request.
Removed the stale state overwrite so in-progress input is preserved.
When scheduling a print to "any printer" with a filament color override,
the scheduler picked a printer with the correct filament type but wrong
color. _find_idle_printer_for_model() used color matching only for
ranking candidates, not filtering — a printer with 0 color matches was
still selected if it had the right types.
Now requires at least 1 color match when filament overrides specify
colors. Printers with 0 matches are skipped and reported in the
"waiting for filament" reason instead of being treated as candidates.
Filenames containing / \ ? or # (e.g. "Abzweigdose/Verteilerdose 70mm")
broke the slicer protocol handler. BambuStudio/OrcaSlicer url_decode()
the entire protocol URL before downloading, which decoded %2F back to /,
creating extra path segments that returned 404. Sanitize these characters
to _ in slicer download URLs — the filename is cosmetic (backend resolves
files by archive ID).
Three independent code paths wrote to archive.cost with conflicting
strategies, causing the same model to produce different prices on each
reprint (e.g. £0.77, £1.54, £2.03).
- Remove add_reprint_cost (Path 3) — redundant cost accumulation that
double-counted on top of usage tracker
- Fix usage tracker (Path 2) — compute cost from current session's
results only, not a SUM of all historical SpoolUsageHistory rows
- Remove _reprint_archives tracking set from main.py
- Update test mocks to match simplified query pattern
archive.cost now always reflects the cost of a single print.
All backend timestamps used datetime.now() (server local time) or the
deprecated datetime.utcnow(). The frontend's parseUTCDate() assumes
timestamps without timezone indicators are UTC and appends 'Z', so
stored timestamps were off by the timezone offset when the container's
timezone wasn't UTC.
Backend: replaced datetime.now() and datetime.utcnow() with
datetime.now(timezone.utc) across 16 files (~80 call sites) for all
database fields and DB comparisons. Cosmetic timestamps (filenames,
user-facing local time formatting) intentionally left as local time.
Frontend: replaced 13 new Date(backendTimestamp) calls with
parseUTCDate() across 8 files to correctly interpret UTC timestamps.
The sidebar hid Settings based on hardcoded role === 'user' instead of
the settings:read permission, and login set user state directly from the
response instead of re-fetching full auth status with permissions.
The "Power off printer when done" checkbox in the print modal and the
auto power off tri-state toggle in the bulk edit modal were accessible
to all users regardless of permissions. Users without printers:control
can now no longer enable auto power off — controls are disabled and
visually dimmed.
The bed cooldown monitor was defined at the end of on_print_complete,
after an early return that exits when no archive is found. Prints
started from BambuStudio or the printer's touchscreen have no archive,
so the function returned before the bed cooldown task was ever created.
Moved the bed cooldown block (function def + task creation) to before
the archive_id early-return so it fires for all completed prints.
Also hardened the temperature dict check from truthiness to isinstance.
Three related fixes in PrintersPage.tsx:
1. External spool (vt_tray) fill level was missing the AMS remain
fallback — only checked Spoolman/inventory, ignoring valid AMS data.
Now uses the same fallback chain as regular and AMS-HT slots.
2. When fill was unknown (null), slot visual showed a full-width gray
bar ("full") while the hover card showed "—" ("empty"). Removed the
misleading gray fallback bar from all three slot types so both views
consistently show "unknown".
3. Fill source priority always preferred AMS remain over Spoolman and
inventory, even when those sources were more accurate (e.g. spools
with usage tracking or migrated from Spoolman). Reversed priority to
Spoolman → Inventory → AMS remain. Fixed fillSource label to reflect
the actual data source (was always 'ams' even when using fallback).
constructor before the try/except, so the FileNotFoundError was unhandled and crashed the entire daemon via asyncio.gather().
Now when the scale hardware isn't available (no /dev/i2c-0), it logs "Scale init failed" and sets self._ok = False, so scale_poll_loop will skip polling with "Scale not
available, skipping scale polling" — same graceful degradation the NFC reader already has.
maziggy