TokenBlacklist Improvements:
Fixed size limit enforcement to properly maintain max size of 1000 tokens
Improved eviction strategy to remove expired tokens first before removing oldest
Added proper cleanup of tokens during Add operation to prevent size overflow
Fixed oldest token eviction logic to ensure correct token removal
Added proper locking mechanisms to prevent race conditions
Cache Improvements:
Fixed cleanup mechanism to only remove truly expired items
Improved eviction strategy in LRU cache to prioritize expired items
Added smarter eviction in evictOldest to scan for expired items first
Fixed aggressive cleanup that was removing valid items
Maintained proper LRU ordering while handling evictions
MetadataCache:
Verified proper implementation of metadata caching with hourly refresh
Confirmed proper handling of cache extension on fetch failures
Validated thread-safe operations with proper RWMutex usage
* Add todo list.
* fixup! Add todo list.
* fixup! fixup! Add todo list.
* fixup! fixup! fixup! Add todo list.
* Improve the session handling and cache.
* Fix an issue where expired session can cause infinite redirect loop
* fixup! Fix an issue where expired session can cause infinite redirect loop
* Add semver setup for automatic releases.
* fixup! Add semver setup for automatic releases.
* fixup! fixup! Add semver setup for automatic releases.
* fixup! fixup! fixup! Add semver setup for automatic releases.
Reduced connection timeouts from 30s to 15s
Decreased idle connection limits from 100 to 30
Lowered keepalive duration from 90s to 30s
Added MaxConnsPerHost limit of 50 to prevent connection flooding
Optimized TLS handshake timeout to 5s
Cache System Optimization:
Implemented size-limited LRU cache with max 1000 items
Added efficient eviction of least recently used items
Improved cleanup process with batch operations
Reduced lock contention by splitting read/write operations
Optimized memory usage with access tracking
Added immediate cleanup of expired items during access
Connection Management:
Enabled connection reuse with keepalives
Reduced connection pool size to conserve memory
Implemented more aggressive connection timeout
Added connection limits per host to prevent resource exhaustion
Removed global metadata cache and sync.Once
Each middleware instance now handles its own metadata initialization
Added tests to verify multiple instances work correctly
The changes ensure that:
Each route gets its own properly initialized middleware instance
Metadata is fetched and set correctly for each instance
No shared state between instances that could cause conflicts
Each instance can handle requests independently
The added test verifies this by creating multiple middleware instances with different routes and confirming they all initialize and function correctly. The test specifically checks that:
Each instance initializes successfully
Each instance gets its own metadata configuration
Each instance can handle requests independently
Callback URLs are correctly set per route
Previous implementations blocked the traefik startup until OIDC plugin was loaded.
This caused chicken-or-egg issue when called OIDC endpoint was hosted by the same traefik as well,
generating rather ridiculous situation when traefik couldn't come up because plugin tried to call the
discovery endpoint which was hosted by the same traefik.
This version resolves the issue allowing for quickstart and lazy loading of the provider metadata.
Disadvantage is - until discovery is done, the plugin will not provide any access to the client.
lukaszraczylo
