When resolving dependencies like certifi@2025.10.5, the bare version
string "2025.10.5" was being rejected as an invalid SpecifierSet and
falling back to wildcard, which fetched the latest version instead.
Now bare versions starting with a digit are automatically prefixed
with "==" to create an exact match constraint.
The dependency parser was stripping environment markers but not checking
if they indicated optional or platform-specific packages. This caused
packages like jaraco.path to pull in pyobjc (324 sub-packages) even on
non-macOS systems.
Changes:
- Filter dependencies with 'extra ==' markers (optional extras)
- Filter dependencies with 'sys_platform' or 'platform_system' markers
- Add diagnostic logging for depth exceeded errors
- Add unit tests for dependency filtering
Fixes tensorflow dependency resolution exceeding max depth.
Real package managers (pip, npm, Maven) don't have depth limits - they
resolve the full dependency tree. We have other safeguards:
- Loop prevention via fetch_attempted set
- Timeout via auto_fetch_timeout setting
- Dependency trees are finite
Instead of failing with 409 on version conflicts, use "first version wins"
strategy. This allows resolution to succeed for complex dependency trees
like tensorflow where transitive dependencies may have overlapping but
not identical version requirements.
The resolver now:
- Checks if an already-resolved version satisfies a new constraint
- If yes, reuses the existing version
- If no, logs the mismatch and uses the first-encountered version
This matches pip's behavior of picking a working version rather than
failing on theoretical conflicts.
When packages like pytest have extras (e.g., pytest[testing]) that depend
on the base package, the resolution was incorrectly detecting this as a
circular dependency.
Added additional check to skip dependencies that resolve to an artifact
already in the visiting set, preventing the false cycle detection while
still catching real circular dependencies.
Auto-fetching missing dependencies from upstream is the more useful default
behavior. Users who need fast, network-free resolution can explicitly set
auto_fetch=false.
Artifacts are content-addressed by SHA256, so reproducibility concerns don't
apply - the same version always produces the same artifact.
When auto_fetch=true and the root artifact doesn't exist locally in a
system project (_pypi), now attempts to fetch it from upstream before
starting dependency resolution. Also fixed a bug where fetched_artifacts
was being redeclared, which would lose the root artifact from the list.
- Remove unused _get_pypi_upstream_sources_cached function (never called)
- Remove unused CacheService import and get_cache helper
- Remove unused cache parameter from pypi_download_file
- Fix asyncio.get_event_loop() deprecation - use get_running_loop()
- Note: The caching implementation was incomplete but the other
performance improvements (connection pooling, batch DB ops) remain
- Fix import in test_db_utils.py: use app.models instead of backend.app.models
- Update health endpoint test to expect 'ok' status and infrastructure keys
- Add CHANGELOG entries for PyPI proxy performance improvements
Add dependency injection helper functions for HttpClientManager
and CacheService, along with imports for the new infrastructure
modules (http_client, cache_service, db_utils).
Add optimized database operations for artifact storage:
- Atomic upserts using ON CONFLICT for artifact creation
- Batch inserts for dependencies to eliminate N+1 queries
- Joined queries for cached URL lookups
- All methods include comprehensive unit tests
Add HttpClientManager class for managing httpx.AsyncClient pools with
FastAPI lifespan integration. Features include:
- Default shared connection pool for general requests
- Configurable max connections, keep-alive, and timeouts
- Dedicated thread pool for blocking I/O operations
- Graceful startup/shutdown lifecycle management
- Per-upstream client isolation support (for future use)
Includes comprehensive unit tests covering initialization, startup,
shutdown, client retrieval, blocking operations, idempotency, and
error handling.
Comprehensive design for:
- HTTP connection pooling with lifecycle management
- Redis caching layer (TTL for discovery, permanent for immutable)
- Abstract PackageProxyBase for multi-protocol support (npm, Maven)
- Database query optimization with batch operations
- Dependency resolution caching for ensure files
- Observability via health endpoints
Maintains hermetic build guarantees: artifact content and extracted
metadata are immutable, only discovery data uses TTL-based caching.
Wheel METADATA files can list the same dependency multiple times under
different extras (e.g., bokeh appears under [docs] and [bokeh-tests]).
This caused unique constraint violations when storing dependencies.
Fix by deduplicating extracted deps before DB insertion.
Adds the tag removal migration to the inline migrations in database.py:
- Drops tag-related triggers and functions
- Removes tag_constraint column from artifact_dependencies
- Makes version_constraint NOT NULL
- Drops tags and tag_history tables
- Renames uploads.tag_name to version
Re-adds the dependency extraction that was accidentally removed with the
proactive caching feature. Now when a PyPI package is cached:
1. Extract METADATA from wheel or PKG-INFO from sdist
2. Parse Requires-Dist lines for dependencies
3. Store in artifact_dependencies table
This restores the dependency graph functionality for PyPI packages.
Adds ORCHARD_PYPI_DOWNLOAD_MODE setting (default: "redirect"):
- "redirect": Redirect pip to S3 presigned URL - reduces pod bandwidth
- "proxy": Stream through Orchard pod - for environments where clients can't reach S3
In redirect mode, Orchard only handles metadata requests and upstream fetches.
All file transfers go directly from S3 to the client.
The list_package_artifacts endpoint was only querying artifacts via the
Upload table. PyPI proxy creates PackageVersion records but not Upload
records, so cached packages would show stats (size, version count) but
no artifacts in the listing.
Now queries artifacts from both Upload and PackageVersion tables using
a union, so PyPI-cached packages display their artifacts correctly.
Large packages like TensorFlow (~600MB) caused read timeouts because the
entire file was loaded into memory before responding to the client. Now
the file is stored to S3 first, then streamed back using StreamingResponse.
- Add sha256 field to list_package_artifacts response (artifact ID is SHA256)
- Add version field to PackageArtifactResponse schema
- Add version field to frontend PackageArtifact type
- Update getArtifactVersion to prefer direct version field
- Fix upload response to return actual version (not requested version)
when artifact already has a version in the package
- Update ref_count tests to use multiple packages (one version per
artifact per package design constraint)
- Remove allow_public_internet references from upstream caching tests
- Update consistency check test to not assert global system health
- Add versions field to artifact schemas
- Fix dependencies resolution to handle removed tag constraint
- Update CacheRequest test to use version field
- Fix upload_test_file calls that still used tag parameter
- Update artifact history test to check versions instead of tags
- Update artifact stats tests to check versions instead of tags
- Fix garbage collection tests to delete versions instead of tags
- Remove TestGlobalTags class (endpoint removed)
- Update project/package stats tests to check version_count
- Fix upload_test_file fixture in test_download_verification
- Remove Tag/TagHistory model tests from unit tests
- Update CacheSettings tests to remove allow_public_internet field
- Replace tag= with version= in upload_test_file calls
- Update test assertions to use versions instead of tags
- Remove tests for tag: prefix downloads (now uses version:)
- Update dependency tests for version-only schema
Tags were mutable aliases that caused confusion alongside the immutable
version system. This removes tags entirely, keeping only PackageVersion
for artifact references.
Changes:
- Remove tags and tag_history tables (migration 012)
- Remove Tag model, TagRepository, and 6 tag API endpoints
- Update cache system to create versions instead of tags
- Update frontend to display versions instead of tags
- Remove tag-related schemas and types
- Update artifact cleanup service for version-based ref_count
The artifacts endpoint only supports sorting by: created_at, size, original_name
But the frontend was defaulting to 'name' (from the old tags endpoint).
- Change default sort from 'name' to 'created_at'
- Change default order from 'asc' to 'desc' (newest first)
- Remove sortable flag from version/tags columns (not DB fields)
- Add sortable flag to original_name and size columns
The circular dependency error '_pypi/psutil → _pypi/psutil' occurred because
dependencies with extras like 'psutil[test]' weren't being recognized as
self-dependencies. The comparison 'psutil[test] != psutil' failed.
- Add _normalize_pypi_package_name() helper that strips extras brackets
and normalizes separators per PEP 503
- Update _detect_package_cycle to use normalized names for cycle detection
- Update check_circular_dependencies to use normalized initial path
- Simplify self-dependency check in resolve_dependencies to use helper
- Add artifact: prefix handling in resolve_dependencies for direct artifact
ID references, enabling dependency resolution for tagless artifacts
- Refactor PackagePage from tag-based to artifact-based data display
- Add PackageArtifact type with tags array for artifact-centric API responses
- Update download URLs to use artifact:ID prefix when no tags exist
- Conditionally show "View Ensure File" only when artifact has tags
- Add _parse_upstream_error() to extract policy messages from JFrog/Artifactory
- Pass through 403 and other 4xx errors with detailed messages
- Pin babel and electron-to-chromium to older versions for CI compatibility
- Install reactflow and dagre for professional graph visualization
- Use dagre for automatic tree layout (top-to-bottom)
- Custom styled nodes with package name, version, and size
- Built-in zoom/pan controls and minimap
- Click nodes to navigate to package page
- Cleaner, more professional appearance
- Add artifact-level self-dependency check (skip if dep resolves to same artifact)
- Close dependency graph modal if package has no dependencies to show
(only root package with no children and no missing deps)
PyPI packages can have self-referential dependencies for extras
(e.g., pytest[testing] depends on pytest). These were incorrectly
detected as circular dependencies. Now we skip them.
The backend returns detail as an object for some errors (circular dependency,
conflicts, etc.). The API client now JSON.stringifies object details so they
can be properly parsed by error handlers like DependencyGraph.
When dependencies are not cached on the server (common since we removed
proactive caching), the dependency graph now:
- Continues resolving what it can find
- Shows missing dependencies in a separate section with amber styling
- Displays the constraint and which package required them
- Updates the header stats to show "X cached • Y not cached"
This provides a better user experience than showing an error when
some dependencies haven't been downloaded yet.
When a dependency has an invalid version constraint like '>=' (without
a version number), the resolver now treats it as a wildcard and returns
the latest available version instead of failing with 'Dependency not found'.
This handles malformed metadata that may have been stored from PyPI packages.
The background task queue for proactively caching package dependencies was
causing server instability and unnecessary growth. The PyPI proxy now only
caches packages on-demand when users request them.
Removed:
- PyPI cache worker (background task queue and worker pool)
- PyPICacheTask model and related database schema
- Cache management API endpoints (/pypi/cache/*)
- Background Jobs admin dashboard
- Dependency extraction and queueing logic
Kept:
- On-demand package caching (still works when users request packages)
- Async httpx for non-blocking downloads (prevents health check failures)
- URL-based cache lookups for deduplication
The pypi_download_file, pypi_simple_index, and pypi_package_versions endpoints
were using synchronous httpx.Client inside async functions. When upstream PyPI
servers respond slowly, this blocked the entire FastAPI event loop, preventing
health checks from responding. Kubernetes would then kill the pod after the
liveness probe timed out.
Changes:
- httpx.Client → httpx.AsyncClient
- client.get() → await client.get()
- response.iter_bytes() → response.aiter_bytes()
This ensures the event loop remains responsive during slow upstream downloads,
allowing health checks to succeed even when downloads take 20+ seconds.
- Remove "All Jobs" title
- Move Status column to front of table
- Add Cancel button for in-progress jobs
- Add cancel endpoint: POST /pypi/cache/cancel/{package_name}
- Add btn-danger CSS styling
- Download packages in 64KB chunks to temp file instead of loading into memory
- Upload to S3 from temp file (streaming)
- Clean up temp file after processing
- Reduces memory footprint from 2x file size to 1x file size
- Add orchard.pypiCache config section to helm values
- Set default workers to 2 (reduced from 5 to limit memory)
- Bump pod memory from 512Mi to 768Mi (request=limit)
- Add ORCHARD_PYPI_CACHE_* env vars to deployment template
- Add overall progress bar showing completed/active/failed counts
- Unify all job types into single table with Type column
- Simplify status to Working/Pending/Failed badges
- Remove NPM "Coming Soon" section
- Add get_recent_activity() function for future activity feed
- Fix dark mode CSS using CSS variables
Backend:
- Add _recover_stale_tasks() to reset tasks stuck in 'in_progress'
from previous crashes (tasks >5 min old get reset to pending)
- Called automatically on startup
Frontend:
- Fix dark mode colors using CSS variables instead of hardcoded values
- Add elapsed time column showing how long task has been running
- Add spinning indicator next to package name
- Add status badge (Running/Stale?)
- Highlight stale tasks (>5 min) in amber
- Auto-updates every 5 seconds with existing refresh
Shows currently processing cache tasks in a dynamic table with:
- Package name and version constraint being cached
- Recursion depth and attempt number
- Start timestamp
- Pulsing indicator to show live activity
Backend changes:
- Add get_active_tasks() function to pypi_cache_worker.py
- Add GET /pypi/cache/active endpoint to pypi_proxy.py
Frontend changes:
- Add PyPICacheActiveTask type
- Add getPyPICacheActiveTasks() API function
- Add Active Workers section with animated table
- Auto-refreshes every 5 seconds with existing data
When the cache worker downloaded a package through the proxy, dependencies
were always queued with depth=0 instead of depth+1. This meant depth limits
weren't properly enforced for nested dependencies.
Changes:
- Add cache-depth query parameter to pypi_download_file endpoint
- Worker now passes its current depth when fetching packages
- Dependencies are queued at cache_depth+1 instead of hardcoded 0
- Add tests for depth tracking behavior
The dashboard was showing "All jobs completed successfully" whenever
there were no failed tasks, even if there were pending or in-progress
jobs. Now shows:
- "All jobs completed" only when pending=0 and in_progress=0
- "Jobs are processing. No failures yet." when jobs are in queue
New admin page at /admin/jobs showing:
- PyPI cache job status (pending, in-progress, completed, failed)
- Failed task list with error details
- Retry individual packages or retry all failed
- Auto-refresh every 5 seconds (toggleable)
- Placeholder for future NPM cache jobs
Accessible from admin dropdown menu as "Background Jobs".
Replace unbounded thread spawning with managed worker pool:
- New pypi_cache_tasks table tracks caching jobs
- Thread pool with 5 workers (configurable via ORCHARD_PYPI_CACHE_WORKERS)
- Automatic retries with exponential backoff (30s, 60s, then fail)
- Deduplication to prevent duplicate caching attempts
New API endpoints for visibility and control:
- GET /pypi/cache/status - queue health summary
- GET /pypi/cache/failed - list failed tasks with errors
- POST /pypi/cache/retry/{package} - retry single package
- POST /pypi/cache/retry-all - retry all failed packages
This fixes silent failures in background dependency caching where
packages would fail to cache without any tracking or retry mechanism.
When background threads fetch from our own proxy using the request's
base_url, it returns http:// but ingress requires https://. The 308
redirect was dropping trailing slashes, causing requests to hit the
frontend catch-all route instead of /pypi/simple/.
Force HTTPS explicitly in the background caching function to avoid
the redirect entirely.
PEP 503 normalizes package names to use hyphens, but wheel filenames
may use underscores (e.g., typing_extensions-4.0.0-py3-none-any.whl).
Convert the search pattern to match either separator.
The background dependency caching was getting 308 redirects because
request.base_url returns http:// but the ingress redirects to https://.
Enable follow_redirects=True in httpx client to handle this.
When a PyPI package is cached, its dependencies are now automatically
fetched in background threads. This ensures the entire dependency tree
is cached even if pip already has some packages installed locally.
Features:
- Background threads fetch each dependency without blocking the response
- Uses our own proxy endpoint to cache, which recursively caches transitive deps
- Max depth of 10 to prevent infinite loops
- Daemon threads so they don't block process shutdown
- Deduplicate dependencies by package name before inserting
- Some packages (like anyio) list the same dep (trio) multiple times with
different version constraints for different extras
- The unique constraint on (artifact_id, project, package) rejected these
- Also removed debug logging from dependencies.py
- Parse version constraints like >=1.9, <2.0 using packaging library
- Find the latest version that satisfies the constraint
- Support wildcard (*) to get latest version
- Fall back to exact version and tag matching
- Add functions to parse Requires-Dist metadata from wheel and sdist files
- Store extracted dependencies in artifact_dependencies table
- Fix streaming response for cached artifacts (proper tuple unpacking)
- Fix version uniqueness check to use version string instead of artifact_id
- Skip creating versions for .metadata files
- Hide tag count stat for system projects (show "versions" instead of "artifacts")
- Hide "Latest" tag stat for system projects
- Change "Create/Update Tag" to only show for non-system projects
- Add "View Artifact ID" menu option with modal showing the SHA256 hash
- Move dependencies section to a modal (opened via "View Dependencies" menu)
- Add deps-modal and artifact-id-modal CSS styles
- Fix artifact_count and total_size calculation to use Tags instead of
Uploads, so PyPI cached packages show their stats correctly
- Fix PackagePage dropdown menu positioning (use fixed position with backdrop)
- Add system project detection for projects starting with "_"
- Show Version as primary column for system projects, hide Tag column
- Hide upload button for system projects (they're cache-only)
- Rename section header to "Versions" for system projects
- Fix test_projects_sort_by_name to exclude system projects from sort comparison
PyPI proxy improvements:
- Set package format to "pypi" instead of "generic"
- Extract version from filename and create PackageVersion record
- Support .whl, .tar.gz, and .zip filename formats
Package page UX overhaul:
- Move upload to header button with modal
- Simplify table: combine Tag/Version, remove Type and Artifact ID columns
- Add row action menu (⋯) with: Copy ID, Ensure File, Create Tag, Dependencies
- Remove cluttered "Download by Artifact ID" and "Create/Update Tag" sections
- Add modals for upload and create tag actions
- Cleaner, more scannable table layout
Projects owned by teams now display the team name in the Owner column
for better organizational continuity when team members change.
Falls back to created_by if no team is assigned.
- Use storage.get_stream(s3_key) instead of non-existent get_artifact_stream()
- Make _pypi project public (is_public=True) so cached packages are visible
Use correct Artifact model fields:
- original_name instead of filename
- Add required created_by and s3_key fields
- Include checksum fields from storage result
The code was calling storage.store_artifact() which doesn't exist.
Changed to use storage.store() which handles content-addressable
storage with automatic deduplication.
Users can now configure the full path including /simple in their
upstream source URL (e.g., https://example.com/api/pypi/repo/simple)
instead of having the code append /simple/ automatically.
This matches pip's --index-url format, making configuration more
intuitive and copy/paste friendly.
The test was checking for the wrong URL pattern. When urljoin resolves
../../packages/ab/cd/... relative to /api/pypi/pypi-remote/simple/requests/,
it correctly produces /api/pypi/pypi-remote/packages/ab/cd/... (not
/api/pypi/packages/...).
Tests now verify the correct response for each scenario:
- 200: HTML content-type
- 404: "not found" error message
- 503: "No PyPI upstream sources configured" error message
- Tests now accept 200/404/503 responses since upstream sources may or
may not be configured in the test environment
- Added upstream_base_url parameter to _rewrite_package_links test
- Added test for relative URL resolution (Artifactory-style URLs)
When behind a reverse proxy that terminates SSL, the server sees HTTP
requests internally. Added _get_base_url() helper that respects the
X-Forwarded-Proto header to generate correct external HTTPS URLs.
This fixes links in the PyPI simple index showing http:// instead of
https:// when accessed via HTTPS through a load balancer.
Artifactory and other registries may return relative URLs in their
Simple API responses (e.g., ../../packages/...). The proxy now resolves
these to absolute URLs using urljoin() before encoding them in the
upstream parameter.
This fixes package downloads failing when the upstream registry uses
relative URLs in its package index.
- Creates an S3 bucket to be used for anything Orchard
Reference in New Issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
