attemptCount tracks how many times the worker has tried each failed
source message. Combined with WORKER_MAX_SKIP_ATTEMPTS (default 5), the
worker will auto-retry across cycles but eventually let the watermark
advance past a chronically failing file so cycles aren't pinned forever.
The SkippedPackage row stays so the user can manually retry via the UI.
CHANNEL_ACCESS_LOST is a new notification type the worker emits when a
source channel becomes inaccessible (account got removed, channel
deleted, etc.) — surfaces the issue instead of silently failing every
cycle as we've been doing with "Iridium 2 w/ Add-ons [Completed]".
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
sendMessage resolves with the temporary message ID inside a .then()
microtask. If TDLib emits updateMessageSendSucceeded synchronously
(cached file, already-known media), the event handler fires while
tempMsgId is still null — the success is dropped and the promise hangs
until the 15-min upload timeout fires.
Buffer success/failure events that arrive before tempMsgId is known,
then replay them in the .then() callback once tempMsgId is set.
Extract completeWithSuccess / completeWithFailure helpers so the
resolution path is shared between live events and replayed events.
This race matters more now that stalls fail fast — without the buffer,
a fast-completing upload could still hang for 15 min before recovery.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Previously a single TDLib event-stream degradation cost ~45 minutes
per archive: 3 retries x 15-min minimum timeout, all on the same
broken client. The retries had no chance of succeeding because the
underlying issue (missing updateMessageSendSucceeded events) is a
client-level problem, not a transient send failure.
Now the first stall throws UploadStallError immediately. The caller
in processArchiveSets already recreates the TDLib client on
UploadStallError, so we drop from ~45 min recovery to ~15 min
(one timeout cycle) per stalled archive.
The stalled set is recorded in SkippedPackage; with the watermark
cap from d99a506 it gets retried on the next ingestion cycle with
a fresh client.
FLOOD_WAIT retries inside sendWithRetry are unchanged — those handle
legitimate rate limiting, not stalls.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Previously the channel/topic watermark could advance past failed
archive sets in two ways:
1. A later successful set raised maxProcessedId past a failed earlier
set within the same scan.
2. scanResult.maxScannedMessageId was used as fallback even when
archives in the scan had failed (added in 77c26ad to prevent
re-scanning empty channels).
Both paths buried failed archives below the watermark on the next
cycle — they sat permanently in SkippedPackage with no auto-recovery.
Now processArchiveSets returns the lowest failed source message ID
alongside the highest processed one. The caller caps the watermark at
(minFailedId - 1n) so the next scan re-includes the failed messages
and processOneArchiveSet retries them. Successful sets above the
failure boundary are not re-uploaded — packageExistsBySourceMessage
early-skips them on the second pass.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The pg pool had max=5 connections shared between Prisma operations and
advisory locks. With 2 account locks held permanently and hash locks
from timed-out (but still running) background work, pool.connect()
would block forever — causing the Turnbase.7z stall.
- Increase pool max from 5 to 15 for headroom
- Add 30s connectionTimeoutMillis so pool.connect() throws instead of
hanging forever when the pool is exhausted
- On startup, terminate zombie PostgreSQL sessions from previous worker
instances that hold stale advisory locks
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Previously, channels/topics with no new archives never had their
watermark updated. This meant every cycle re-scanned all messages from
scratch just to discover nothing new — especially costly for the 1079-
topic Model Printing Emporium forum.
- Add maxScannedMessageId to ChannelScanResult (highest msg ID seen)
- Set channel watermark to scan boundary when no archives are found
- Set topic watermark to scan boundary when no archives are found
- Fall back to scan watermark when archive processing doesn't advance it
After one full cycle, subsequent cycles will skip already-scanned
messages via the early-exit boundary check, dramatically reducing
TDLib API calls on channels with mostly non-archive content.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
searchChatMessages returns newest-first. Once the oldest message on a
page is at or below the lastProcessedMessageId boundary, all remaining
pages are even older. Stop scanning immediately instead of reading every
message in the channel.
This was already implemented for topic scans but missing from channel
scans. On a test run, total messages scanned dropped from 3805 to 1615
(57% reduction) for an account with no new archives.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
When TDLib's event stream degrades, uploads complete (bytes sent) but
confirmations never arrive. Previously the worker retried 3x with the
same broken client, wasting 60+ min per archive and holding the mutex.
- Add UploadStallError class to distinguish stalls from other failures
- Reduce stall detection timeout from 5min to 3min (faster detection)
- Recreate TDLib client after consecutive upload stalls instead of
retrying on the same degraded connection
- Add forceReleaseMutex() to prevent cascade failures when one account
blocks others via stuck mutex after cycle timeout
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
After TDLib login completes, calls getMe() to detect isPremium, persists
it to DB via updateAccountPremiumStatus, and returns { client, isPremium }
from createTdlibClient. All callers updated to destructure accordingly.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The FK referenced "users" but the actual table is "User" (no @@map in Prisma schema).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Manual file upload:
- Upload dialog in STL page with drag-and-drop file picker
- Files saved to shared Docker volume (/data/uploads)
- Worker processes via pg_notify('manual_upload') channel
- Hashes, reads metadata, splits >2GB, uploads to Telegram
- Multiple files automatically grouped
- Status polling shows upload/processing/complete states
Notification fixes:
- Add dismiss (X) button on each notification
- Add "Clear" button to remove all notifications
- Fix false positive MISSING_PART alerts from legacy packages
(only flag when >1 destMessageIds stored but count wrong,
not when only 1 ID from backfill)
Infrastructure:
- ManualUpload + ManualUploadFile schema + migration
- Shared manual_uploads Docker volume between app and worker
- Upload API routes (POST /api/uploads, GET /api/uploads/[id])
- Worker manual-upload processor with full pipeline
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Manual override training (GroupingRule):
- Learn patterns from manual group creation (common filename prefix or creator)
- Apply learned rules as first auto-grouping pass (highest confidence after albums)
- GroupingRule model stores pattern, channel, signal type, confidence
Hash verification after upload:
- Re-hash upload files on disk before indexing to catch disk corruption
- Creates HASH_MISMATCH notification on discrepancy
Grouping conflict detection:
- After all grouping passes, check if grouped packages match rules from different groups
- Creates GROUPING_CONFLICT notification for manual review
Per-channel grouping flags:
- Add autoGroupEnabled boolean to TelegramChannel (default true)
- Auto-grouping passes (all except album) gated behind this flag
- Album grouping always runs as it reflects Telegram's native behavior
Full-text search (tsvector):
- Add searchVector tsvector column with GIN index and auto-update trigger
- Backfill 1870 existing packages
- FTS with ts_rank for ranked results, ILIKE fallback for short/failed queries
- Applied to both web app and bot search
Bot group awareness:
- /group <query> — view group info or search groups by name
- /sendgroup <id> — send all packages in a group to linked Telegram account
Bulk repair:
- repairPackageAction clears dest info and resets watermark for re-processing
- Repair button in notification bell for MISSING_PART and HASH_MISMATCH alerts
- /api/notifications/repair endpoint
Retroactive category re-tagging:
- When channel category changes, auto-update tags on all existing packages
- Removes old category tag, adds new one
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Group merge UI:
- Add mergeGroups query and mergeGroupsAction server action
- Add "Start Merge" / "Merge Here" buttons to group row actions
- Two-step UX: click Start on source, click Merge Here on target
ZIP path prefix grouping (Signal 7):
- Compare PackageFile.path root folders across ungrouped packages
- Auto-group if 2+ packages share the same dominant root folder
Reply chain grouping (Signal 6):
- Capture reply_to_message_id during channel scanning
- Group archives that reply to the same root message
- Add replyToMessageId field to Package schema
Caption fuzzy match grouping (Signal 8):
- Capture source caption during channel scanning
- Normalize captions (strip extensions, extract significant words)
- Group packages with matching normalized caption keys
- Add sourceCaption field to Package schema
Periodic integrity audit:
- Check multipart packages for completeness (parts vs destMessageIds)
- Detect orphaned indexes (destChannelId set but no destMessageId)
- Runs after each ingestion cycle, deduplicates notifications
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Pattern grouping (Signal 3):
- Extract YYYY-MM dates, month names, and project prefixes from filenames
- Auto-group packages sharing the same pattern within a channel
- Groups created with groupingSource=AUTO_PATTERN
Creator grouping (Signal 4):
- Auto-group 3+ ungrouped packages from the same creator within a channel
- Runs after pattern grouping as lowest-priority automatic signal
Notification UI:
- Add NotificationBell component to header with unread badge
- Popover panel shows recent notifications with severity icons
- Mark individual or all notifications as read
- Polls every 30 seconds for updates
Failure notifications:
- Upload/download failures now create SystemNotification records
- Visible in the notification bell alongside hash mismatch alerts
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Schema:
- Add GroupingSource enum (ALBUM, MANUAL, AUTO_TIME, AUTO_PATTERN, etc.)
- Add groupingSource field to PackageGroup with backfill
- Add SystemNotification model for persistent alerts
- Add NotificationType and NotificationSeverity enums
Ungrouped staging tab:
- Add listUngroupedPackages/countUngroupedPackages queries
- Add "Ungrouped" tab to STL page showing packages without a group
Time-window auto-grouping:
- After album grouping, cluster ungrouped packages within configurable
time window (default 5 min, AUTO_GROUP_TIME_WINDOW_MINUTES env var)
- Groups named from common filename prefix
- Groups created with groupingSource=AUTO_TIME
Hash verification after split:
- Re-hash split parts and compare to original contentHash
- Log error and create SystemNotification on mismatch
- Prevents silently corrupted split uploads
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Raise WORKER_MAX_ZIP_SIZE_MB from 4GB to 200GB (production .env)
- Make MAX_PART_SIZE configurable via MAX_PART_SIZE_MB env var
(default 1950 MiB, set to 3900 for Premium accounts)
- Remove hardcoded 1950 MiB constants in split.ts and worker.ts
- Add grouping system audit report with real-world failure cases
10 archives were blocked by the 4GB limit (up to 70.5GB).
They will be retried on next ingestion cycle.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Multi-part send fix:
- Add destMessageIds BigInt[] to Package schema with backfill migration
- Worker uploadToChannel now returns all message IDs, stored in DB
- Bot forwards all parts of multi-part archives (not just the first)
- Add retry logic for upload rate limits (429) and download stalls
Kickstarter package linking:
- Add package search/linking queries and API routes
- Add PackageLinkerDialog with search + checkbox selection
- Add "Link Packages" and "Send All" actions to kickstarter table
- Add sendAllKickstarterPackages server action
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Groups related packages posted together in Telegram channels.
Auto-detects albums via media_album_id, supports manual grouping
from UI. Groups appear as collapsible rows in STL files table.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- createOrFindPackageGroup: catch unique constraint violation from
concurrent creates and fall back to findFirst
- createManualGroup: guard against empty package results before
accessing first element
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Update STL page to use listDisplayItems query for mixed package/group display
- Rewrite package-columns to handle StlTableRow union type (group headers + packages)
- Add group expand/collapse with chevron toggle and indented member rows
- Add checkbox selection with "Group N Selected" toolbar button and dialog
- Add inline group actions: rename, dissolve, send all, remove member
- Add clickable group preview thumbnail with file upload for preview images
- Extend DataTable with optional rowClassName prop for group row styling
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Add downloadStarted flag to prevent false "stopped unexpectedly" errors
when TDLib emits initial updateFile before download is active
- Add 5-minute stall detection for both downloads and uploads
- Reduce max split part size from 2GiB to 1950MiB to stay under
Telegram's internal upload part count limits
- Increase timeouts from max(10min, 15min/GB) to max(15min, 20min/GB)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
1. Distinguish failure reasons: inspect error messages to label skipped
packages as DOWNLOAD_FAILED, UPLOAD_FAILED, or EXTRACT_FAILED
instead of catch-all DOWNLOAD_FAILED.
2. Detect orphaned uploads: before uploading, check if the same content
hash already has a successful upload on the destination channel. Reuse
the existing message ID instead of re-uploading (prevents duplicates
when worker crashed between upload and DB write).
3. Increase timeouts: download from max(5min, GB*10min) to
max(10min, GB*15min), upload from GB*10min to GB*15min.
Prevents premature timeouts on slow connections.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Hand-written migration matching the SkippedPackage model in schema.prisma.
Will be applied automatically by docker-entrypoint.sh via `prisma migrate deploy`.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Replace BigInt literal `1n` with `BigInt(1)` for ES target compatibility
- Add default matchedFileCount/matchedByContent to getPackageById return
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
