- Removed deprecated functions related to document parsing and blob storage in blobstore.ts.
- Introduced new PDF rendering logic in pdf-preview-renderer.ts and pdf-preview-pdfjs-runtime.ts.
- Updated previews-render.ts to utilize the new PDF rendering functions.
- Refactored user-whisper-align-job.ts to use the compute-worker client for alignment requests.
- Enhanced artifact.ts and operation.ts to validate parsed PDF artifacts and resolve current PDF parses.
- Updated snapshot.ts to align with new worker operation types.
- Adjusted runtime-config.ts to check for compute-worker availability.
- Modified types in parsed-pdf.ts and tts.ts to reflect changes in the compute-worker protocol.
- Added unit tests for PDF artifact validation and compute-worker client contract.
- Removed obsolete pdf-op-key.vitest.spec.ts test file.
* refactor(ui): replace anchor tags with next/link in SidebarNavLink and UserMenu
Update SidebarNavLink to use next/link for navigation instead of anchor tags,
ensuring proper routing and improved accessibility in Next.js. Refactor UserMenu
to remove legacy Link wrappers and directly use SidebarNavLink for signin and
signup links. This streamlines navigation components and aligns with Next.js
best practices.
* fix(pdf): handle non-zero viewport origins and improve layout model ordering
Update normalizeTextItemsForLayout to correctly apply viewport transforms,
including non-zero page origins, ensuring accurate mapping of PDF text items
to top-left coordinates. Refactor runLayoutModel to implement a custom order
sequence builder for layout regions using model order logits, improving
region ordering consistency with model semantics. Update related tests to
cover viewport transform edge cases.
* refactor(pdf): improve text normalization with font ascent and vertical overlap logic
Enhance text normalization by incorporating font ascent and descent data to more accurately position glyphs, especially for decorative initials. Update merge logic to better detect line membership using vertical overlap, ensuring drop caps and overlapping glyphs are merged correctly. Extend tests to cover these layout scenarios.
* style(range): redesign slider with precision gauge and ruler ticks
Revamp the range input to feature a minimalist "precision gauge" style.
Introduce a hairline rail, ruler notches for discrete steps, and a slim
needle thumb. Add CSS variables and logic for per-instance tick sizing
and coloring. Remove bulky inline class-based styling in favor of
centralized CSS for improved maintainability and visual clarity.
* refactor(ui): modularize PDF loader and range slider visuals
Move PDF layout scan visualization and range slider styles into dedicated CSS
modules, isolating their styles from the global scope. Integrate PdfLayoutScan
component into the PDF viewer loader UI for animated parse progress. Refactor
progress bars to use a reusable progress-fill class with animated sheen effect.
Update range input to use CSS module for precision gauge styling.
* style(reader): remove grid overlay from PdfLayoutScan visualization
* refactor(api): add staleness detection for inflight worker operation states
Integrate isWorkerOperationStateStale checks into document parse API endpoints to ensure inflight worker operation states are not reused if stale. Introduce helper for staleness detection and corresponding unit tests. Enhance SSE event streaming with keepalive intervals and improve progress acknowledgment error handling.
* feat(worker): recover and fail stale in-flight pdf ops on startup
Add orphaned operation recovery logic to detect and mark stale in-flight pdf_layout jobs as failed during worker startup. Extend OperationStateStore with listOpStates for state enumeration. Update tests and documentation to cover recovery behavior and new environment variable COMPUTE_PDF_JOB_ATTEMPTS.
* refactor(worker): distinguish staleness thresholds for running and queued pdf ops
Update orphan recovery logic to apply separate timeouts for 'running' and
'queued' pdf_layout operations. Adjust tests to verify that only stale
'running' operations are failed, while stale 'queued' operations remain
untouched.
* feat(worker): extend orphan recovery to handle whisper_align ops and improve logging
Update orphan recovery to detect and fail stale 'running' whisper_align operations
in addition to pdf_layout. Refactor recovery logic to generalize staleness checks
across operation kinds and enhance log output with detailed operation info.
Expand tests to verify correct handling of both whisper_align and pdf_layout
operations in running and queued states.
* refactor(control-plane): introduce revision-based CAS for operation state updates
Add revision tracking and compare-and-set (CAS) semantics to operation state
stores, enabling atomic state transitions and preventing lost updates. Extend
the OperationStateStore interface with getOpStateRecord and compareAndSetOpState
methods. Update orchestrator and worker runtime to utilize CAS for marking
operations as failed only if the state is unchanged. Enhance in-memory,
JetStream, and test control plane implementations to support revision logic.
This change improves concurrency safety and correctness of operation state
management across distributed components.
* feat(ui): add parse failure state to PDF layout scan animation
Display a distinct "parse halted" visual state in the PDF layout scan
component and PDF viewer page when parsing fails. The loader animation
is replaced by a static, dimmed page with an alert glyph and updated
styling, ensuring users are not misled by an active animation after a
failure. CSS and component logic updated to support the new state.
* feat(worker): extract orphaned operation recovery to module with periodic sweep
Move orphaned operation recovery logic into a dedicated orphan-recovery module,
introducing a periodic sweep timer that triggers recovery every 15 seconds while
the worker is connected. Refactor runtime to delegate orphan detection and
handling to the new module, improving modularity and maintainability. Add
unit tests for orphan-recovery to ensure correctness.
* fix(ui): adjust PDF viewer layout and update parse loader description
* refactor(pdf): streamline layout model region extraction and update test coverage
- Replace custom order sequence logic with softmax-based class selection in runLayoutModel
- Remove unused sigmoid and buildOrderSequence functions
- Simplify detection loop to filter and map regions directly
- Add targeted tests for layout model extraction logic
- Update CSS animation naming for consistency
- Clarify test description and add inline comments for orphan recovery scenario
* fix(pdf): add strict validation for layout model output shapes and extend test coverage
Add explicit error handling for invalid or inconsistent pred_boxes and logits array lengths in runLayoutModel to prevent silent failures. Expand test suite to verify correct region filtering and error scenarios, ensuring only labeled regions are returned and malformed outputs are handled robustly.
* docker fixes
* fix(worker): default COMPUTE_PREWARM_MODELS to false and update docs, env, and compose
Set COMPUTE_PREWARM_MODELS to false by default in runtime, docker-compose,
and .env.example to avoid pre-downloading ONNX models during worker startup.
Update Dockerfile and deployment documentation to reflect this change and
clarify configuration for model prewarming.
* fix(docker): downgrade pnpm version to 10.33.4 for compatibility
---------
Co-authored-by: Richard R <me@richardr.dev>
Revise the compute worker deployment documentation to clarify usage
scenarios, streamline environment variable instructions, and improve
separation between embedded and standalone deployment modes. Remove
outdated references and reformat for better readability.
Introduce `serverLogger` utility based on pino for consistent, structured logging across all server and API modules. Replace direct console logging with `serverLogger` and add request-scoped logging helpers. Update environment variable handling, documentation, and deployment guides to reflect new logging configuration (`LOG_FORMAT`, `LOG_LEVEL`, `COMPUTE_LOG_LEVEL`). Enforce no-console in server code via ESLint and add pino/pino-pretty dependencies.
This change standardizes log output, improves observability, and prepares the codebase for ingestion by log platforms.
Expand documentation and example env files to explicitly distinguish which
environment variables belong in the app/root `.env` versus the `compute/worker/.env*`
files when deploying in external worker mode. This clarifies routing/auth
settings versus worker runtime configuration, reducing misconfiguration risk.
Update all configuration, manifest, and documentation references to use the q4
Whisper ONNX model files instead of int8. Adjust expected file names, hashes,
and environment variable descriptions to reflect this new default. This aligns
runtime, deployment, and developer documentation with the updated model artifact
expectations for improved consistency.
Introduce COMPUTE_NATS_REPLICAS environment variable to control the number of
replicas for JetStream streams and key-value buckets. Only values 1, 3, or 5
are accepted, defaulting to 1 for other inputs. Update worker server logic and
deployment documentation to describe and support this configuration.
Add lazy NATS connection lifecycle and idle disconnect logic to the worker,
enabling automatic disconnection after 120s of inactivity to support Railway
container sleep. Update deployment docs to describe this behavior, including
caveats and reconnection flow.
Introduce automatic healing for stale PDF parse states in API routes. Add
`isDocumentParseStateStale` utility and use a shared `COMPUTE_OP_STALE_MS`
environment variable (with config getter) to control the stale window for both
worker op replacement and app-side parse-state healing. Update documentation and
environment variable references to reflect new config.
Enables automatic marking of stuck parses as failed, improving reliability and
retry behavior in distributed compute environments.
Align COMPUTE_PDF_JOB_ATTEMPTS default value with new ONNX-based PDF layout
parsing approach by setting it to 1 in environment examples, worker config,
and deployment documentation. This reflects improved reliability and reduces
unnecessary retries for PDF processing jobs.
Standardize compute job concurrency configuration by introducing a shared
COMPUTE_JOB_CONCURRENCY setting, replacing separate PDF and Whisper concurrency
controls. Add cross-platform CPU core detection and thread budgeting utilities,
and update both ONNX model execution and concurrency limiters to use dynamic
thread allocation per job. Refactor environment variable docs and examples to
reflect unified concurrency management. Streamline PDF.js font path resolution
for improved reliability across environments.
Update compute worker to use separate environment variable for PDF job attempts
(`COMPUTE_PDF_JOB_ATTEMPTS`) and set Whisper job max deliveries to 1, clarifying
and separating retry logic between job types. Adjust `.env.example` and deployment
docs to comment out defaults and document new/renamed variables for advanced tuning.
In TTS segment ensure route, introduce explicit timeouts for Whisper alignment
operations, with configurable durations for local and worker modes, improving
robustness and error handling for long-running alignments.
Migrate worker and backend contract to an operation-centric model, introducing opKey-based deduplication for Whisper alignment and PDF layout parsing. Add new operation request/state types and update worker endpoints to accept generic operation requests and expose operation status via SSE. Refactor client and server logic to support operation flows. Update deployment docs, environment examples, and Dockerfile to reflect new architecture and dependency management.
Align compute worker to use PORT environment variable consistently across
Dockerfile, docker-compose, and server code, replacing COMPUTE_WORKER_PORT.
Update .env.example and deployment documentation to clarify platform-specific
behavior, especially for Railway and similar environments. This simplifies
configuration and reduces platform-specific issues.
Add `COMPUTE_JOBS_STREAM_MAX_BYTES` and `COMPUTE_JOB_STATES_MAX_BYTES` to
environment example, worker server configuration, and deployment docs. These
variables allow tuning JetStream resource caps for job queue and state storage
to better fit deployment requirements.
Add a comprehensive Synadia Cloud and Railway setup guide to the compute worker
deployment docs, including step-by-step environment variable configuration and
example image tags. Update Vercel deployment instructions with a quick start for
external compute worker integration via Railway and Synadia NGS. Document new
`.env.prod` pattern in .gitignore for production environment management.
Expand compute worker documentation to explain NATS authentication using
either a credentials file or raw credentials string, including Synadia Cloud
integration steps. Update environment variable examples and server connection
logic to support both `NATS_CREDS_FILE` and `NATS_CREDS`. Add `.creds` to
.gitignore to avoid accidental credential leaks.
This improves deployment flexibility for cloud and on-prem environments by
supporting multiple authentication methods for NATS JetStream.
Replace Redis and BullMQ with NATS JetStream and KV for job queuing and state management
in the compute worker service. Update environment variables, Docker Compose, and documentation
to reflect the new message queue backend. Adjust dependencies and code to use NATS-based
work queue and key-value storage for durable job processing.
This change improves scalability and reliability of distributed compute workloads by leveraging
NATS JetStream's work queue and persistence features. Documentation and configuration examples
are updated to guide deployments using the new backend.
Introduce support for external compute worker mode (`COMPUTE_MODE=worker`) using a new `WorkerComputeBackend`. This enables offloading heavy ONNX Whisper alignment and PDF layout parsing to a standalone worker service (Redis + BullMQ), improving scalability and compatibility with serverless/limited environments.
- Add `@openreader/compute-core` as a shared package for ONNX inference and PDF parsing logic.
- Implement `WorkerComputeBackend` and worker contract/types for remote job execution.
- Update compute backend selection logic and remove previous worker mode guards.
- Extend `WhisperAlignInput` and `PdfLayoutInput` types to support object keys for remote data access.
- Refactor local compute backend to use `@openreader/compute-core` and support both buffer and object key inputs.
- Update job runner, TTS segment alignment, and PDF layout parsing flows to use new compute backend APIs.
- Add scripts, Docker workflow, and documentation for deploying and running the compute worker.
- Update environment variable docs and examples for worker mode, including storage requirements and configuration.
- Document published images and stack changes to reflect the new compute worker architecture.
BREAKING CHANGE: `COMPUTE_MODE=worker` now requires an external compute worker service and S3-compatible object storage. Embedded SeaweedFS (`weed mini`) is not supported in worker mode. See the new documentation for deployment and configuration details.
