Some upstream Kling deployments still expect the legacy `model` key
instead of `model_name`.
This change adds the `model` field to `requestPayload` and populates it
with the same value as `model_name`, ensuring the generated JSON works
with both old and new versions.
Changes:
• Added `Model string "json:\"model,omitempty\""` to `requestPayload`
• Set `Model` alongside `ModelName` in `convertToRequestPayload`
• Updated comments to clarify compatibility purpose
Result:
Kling task requests now contain both `model_name` and `model`, removing
integration issues with upstreams that only recognize one of the keys.
Previously, the KlingRequestConvert middleware only extracted model name from
the 'model_name' field, which caused 503 errors when requests used the 'model'
field instead. This enhancement improves API compatibility by supporting both
field names.
Changes:
- Modified KlingRequestConvert() to check for 'model' field if 'model_name' is empty
- Maintains backward compatibility with existing 'model_name' usage
- Fixes "no available channels for model" error when model field was not recognized
This resolves issues where valid Kling API requests were failing due to field
name mismatches, improving the overall user experience for video generation APIs.
- Updated TokenAuth middleware to handle requests for both `/v1beta/models/` and `/v1/models/`.
- Adjusted distributor middleware to recognize the new model path.
- Enhanced relay mode determination to include the new model path.
- Added route for handling POST requests to `/models/*path`.
These changes ensure compatibility with the new model API structure, improving the overall routing and authentication flow.
Summary
• Migrated all ratio-related sources into `setting/ratio_setting/`
– `model_ratio.go` (renamed from model-ratio.go)
– `cache_ratio.go`
– `group_ratio.go`
• Changed package name to `ratio_setting` and relocated initialization (`ratio_setting.InitRatioSettings()` in main).
• Updated every import & call site:
– Model / cache / completion / image ratio helpers
– Group ratio helpers (`GetGroupRatio*`, `ContainsGroupRatio`, `CheckGroupRatio`, etc.)
– JSON-serialization & update helpers (`*Ratio2JSONString`, `Update*RatioByJSONString`)
• Adjusted controllers, middleware, relay helpers, services and models to reference the new package.
• Removed obsolete `setting` / `operation_setting` imports; added missing `ratio_setting` imports.
• Adopted idiomatic map iteration (`for key := range m`) where value is unused.
• Ran static checks to ensure clean build.
This commit centralises all ratio configuration (model, cache and group) in one cohesive module, simplifying future maintenance and improving code clarity.
- Introduced a new constant `AzureNoRemoveDotTime` in `constant/azure.go` to manage model name formatting for channels created after May 10, 2025.
- Updated `distributor.go` to set `channel_create_time` in the context.
- Modified `adaptor.go` to conditionally remove dots from model names based on the channel creation time.
- Enhanced `relay_info.go` to include `ChannelCreateTime` in the `RelayInfo` struct.
- Updated English localization files to reflect changes in model name handling for new channels.
Reason: The original steps 1 and 3 in the redisRateLimitHandler method were not atomic, leading to poor precision under high concurrent requests. For example, with a rate limit set to 60, sending 200 concurrent requests would result in none being blocked, whereas theoretically around 140 should be intercepted.
Solution: I chose not to merge steps 1 and 3 into a single Lua script because a single atomic operation involving read, write, and delete operations could suffer from performance issues under high concurrency. Instead, I implemented a token bucket algorithm to optimize this, reducing the atomic operation to just read and write steps while significantly decreasing the memory footprint.
- Add new context keys for user-related information
- Modify user cache and authentication middleware to populate context
- Refactor quota and notification services to use context-based user data
- Remove redundant database queries by leveraging context information
- Update various components to use new context-based user retrieval methods
