Releases: opensearch-project/ml-commons
Releases · opensearch-project/ml-commons
2.9.0.0
Version 2.9.0.0 Release Notes
Compatible with OpenSearch 2.9.0
Features
- remote inference: add connector; fine tune ML model and tensor class (#1051)
- remote inference: add connector executor (#1052)
- connector transport actions, requests and responses (#1053)
- refactor predictable: add method to check if model is ready (#1057)
- Add basic connector access control classes (#1055)
- connector transport actions and disable native memory CB (#1056)
- restful connector actions and UT (#1065)
- Change connector access control creation allow empty list (#1069)
Enhancements
- create model group automatically with first model version (#1063)
- init master key automatically (#1075))
Infrastructure
- Adding an integration test for redeploying a model (#1016)
- add unit test for connector class in commons (#1058)
- remote inference: add unit test for model and register model input (#1059)
- remote inference: add unit test for StringUtils and remote inference input (#1061)
- more UT for rest and trasport actions (#1066)
- remote inference: add unit test for create connector request/response (#1067)
- Add more UT for remote inference classes (#1077)
- IT Security Tests for model access control (#1095)
- Add integration tests for the remote inference feature (#1073)
- Temporarily ignore OpenAI Tests (#1118)
- Increase sleep time on remote inference integration tests (#1124)
- fix remote inference it: update job interval (#1129)
- Add null key check for remote inference integration tests (#1113)
- Reenable OpenAI Remote Inference Tests (#1127)
Bug Fixes
- Add missing codes from pen test fix (#1060)
- fix cannot specify model access control parameters error (#1068)
- fix memory circuit breaker (#1072)
- PenTest fixes: error codes and update model group fix (#1074)
- Fix rare private ip address bypass SSRF issue (#1070)
- leftover in the 404 Not Found return error (#1079)
- modify error message when model group not unique is provided (#1078)
- stash context before accessing ml config index (#1092)
- fix init master key bug (#1094)
- delete connector successfully if model index is missing (#1106)
- fix encoding US-ASCII error (#1126)
Documentation
Maintenance
- Increment version to 2.9.0-SNAPSHOT (#955)
- Manual CVE backport (#1008)
- Fix build. (#1018)
- Fix the refactor change brought by core backport (#1047)
- change to compileOnly to avoid jarhell (#1062)
- Update ml CI workflow checkout ref to be PR commit (#1114)
- Add shell bash settings for windows github actions to export credentials (#1116)
2.8.0.0
Version 2.8.0.0 Release Notes
Compatible with OpenSearch 2.8.0
Experimental Features
- Model access control. (#928)
Enhancements
- Add a setting to enable/disable model url in register API (#871)
- Add a setting to enable/disable local upload while registering model (#873)
- Check hash value for the pretrained models (#878)
- Add pre-trained model list (#883)
- Add content hash value for the correlation model. (#885)
- Set default access_control_enabled setting to false (#935)
- Enable model access control in secure reset IT (#940)
- Add model group rest ITs (#942)
Bug Fixes
- Fix class not found exception when deserialize model (#899)
- Fix publish shadow publication dependency issue (#919)
- Fix model group index not existing model version query issue and SecureMLRestIT failure ITs (#933)
- Fix model access mode upper case bug (#937)
Documentation
Maintenance
- Increment version to 2.8.0-SNAPSHOT (#896)
Refactoring
- Change mem_size_estimation to memory_size_estimation (#868)
2.7.0.0
Version 2.7.0.0 Release Notes
Compatible with OpenSearch 2.7.0
Experimental Features
- Add metrics correlation algorithm. (#845)
Enhancements
- Add model auto deploy feature (#852)
- Add memory consumption estimation for models in profile API (#853)
- Add text docs ML input (#830)
- Add allow custom deployment plan setting; add deploy to all nodes field in model index (#818)
- Add exclude nodes setting. (#813)
- set model state as partially loaded if unload model from partial nodes (#806)
Bug Fixes
- change to old method to fix missing method createParentDirectories (#759)
- fix delete model API (#861)
- fix breaking changes of Xcontent namespace change (#838)
- Change the ziputil dependency to fix a potential security concern (#824)
- fix checkstyle version (#792)
- Typo fix and minor improvement in maven-publish GHA workflow (#757)
Documentation
- add docker-compose file for starting cluster with dedicated ML node (#799)
Maintenance
Refactoring
2.6.0.0
Compatible with OpenSearch 2.6.0.
Experimental Features
- enable prebuilt model. (#729)
Enhancements
- tune error message for no ML node case (#699)
- [experimental] Enhance profile API to add model centric result controlled by view parameter (#714)
- [experimental] tune model config: change pooling mode to optional (#724)
- check state before deleting model or task (#725)
Bug Fixes
Documentation
- Update gpu doc with docker test (#702)
- add text embedding API example doc (#710)
- fix profile API in example doc (#712)
- change model url to public repo in text embedding model example doc (#713)
- json listing of all the pretrianed models (#730)
Maintenance
- Increment version to 2.6.0-SNAPSHOT (#671)
Refactoring
- add DL model class (#722)
2.5.0.0
Compatible with OpenSearch 2.5.0
Features
- Add native memory circuit breaker. (#689)
Enhancements
- add more parameters for text embedding model (#640)
- add more pooling method and refactor (#672)
- add ML task timeout setting and clean up expired tasks from cache (#662)
- change only run on ml node setting default value to true (#686)
Infrastructure
- unit tests coverage for load/unload/syncup (#592)
- Add .whitesource configuration file (#626)
- bump djl to 0.20 and add onnxruntime-gpu dependency (#644)
- Remove jackson-databind and jackson-annotations dependencies now coming from core (#652)
- Revert "Remove jackson-databind and jackson-annotations dependencies now coming from core" (#687)
- Adding backwards compatibility test for ml-commons plugin (#681)
- Change the inheritance of the BWC test file (#692)
Documentation
- Updated MAINTAINERS.md format (#668)
- Updating maintainers list (#663)
- add doc about how to setup GPU ML node (#677)
Maintenance
- Increment version to 2.5.0-SNAPSHOT (#513)
Refactoring
- change task worker node to list; add target worker node to cache (#656)
2.0.0.0
Compatible with OpenSearch 2.0.0
Enhancements
- Add circuit breaker trigger count stat.(#274)
Bug Fixes
- support dispatching execute task; don't dispatch ML task again (#279)
- Fix NPE in anomaly localization (#280)
- create model/task index with correct mapping (#284)
Infrastructure
- drop support for JDK 14 (#267)
- Add UT/IT Coverage for action/models and action/tasks. (#268)
- Default qualifier to alpha1 and fix workflows (#269)
- Remove additional vars in build.gradle that are not used (#271)
- Add UT for Search transport action. (#272)
- updated issue templates for bugs and features. (#273)
- Add more test to improve coverage of abstract search action(#275)
- Add UT for RestMLExecuteAction, and remove it out from the jacoco exclusive list. (#278)
- add coverage badges (#281)
- Re-enable docker image tests for 2.0. (#288)
Maintenance
- Change 2.0-alpha1 to 2.0-rc1. (#282)
- bump RCF version to 3.0-rc2.1 (#289)
- bump tribuo version to 4.2.1 (#312)
Refactoring
- Removed RCF jars and updated to fetch RCF 3.0-rc2 from maven (#277)
1.3.0.0
Compatible with OpenSearch 1.3.0
Features
- add anomaly localization implementation (#103)
- refactor ML task data model; add create ML task index method (#116)
- integration step 1 and 2 for anomaly localization (#113)
- anomaly localization integration step 3 (#114)
- support train ML model in either sync or async way (#124)
- anomaly localization integration step 4 and 5 (#125)
- add train and predict API (#126)
- add ML Model get API (#117)
- integrate tribuo anomaly detection based on libSVM (#96)
- add ML Delete model API (#136)
- add fixed in time rcf (#138)
- Add ML Model Search API (#140)
- add circuit breaker (#142)
- add batch RCF for non-time-series data (#145)
- Add ML Task GET/Delete API (#146)
- Add Search Task API and Refactor search actions and handlers (#149)
- add minimum top contributor candidate queue size (#151)
- add more stats: request/failure/model count on algo/action level (#159)
- add tasks API in Client (#200)
Enhancements
- support float type in data frame (#129)
- support short and long type in data frame (#131)
- use threadpool in execute task runner (#156)
- do not return model and task id in response (#171)
- more strict check on input parameters by applying non-coerce mode (#173)
- move anomaly localization to the last position to avoid BWC issue (#189)
Bug Fixes
- use latest version of tribuo to fix modify thread group permission issue (#112)
- fix jarhell error from SQL build (#137)
- fix EpochMilli parse error in MLTask (#147)
- fix permission when accessing ML system indices (#148)
- fix system index permission issue in train/predict runner (#150)
- cleanup task cache once task done (#152)
- fix update task semaphore; don't return task id for sync request (#153)
- restore context after accessing system index to check user permission on non-system index (#154)
- fix verbose error message thrown by invalid enum (#167)
- fix no permission to create model/task index bug;add security IT for train/predict API (#177)
Infrastructure
- add git ignore file (#92)
- change common utils to 1.2 snapshot;add more test (#94)
- Remove jcenter dependency (#121)
- add integration test for train and predict API (#157)
- fix build/CI and add backport workflow (#161)
- publish ml client to maven (#165)
- Add integ tests for model APIs (#166)
- add security IT (#168)
- fix maven group (#170)
- add more UT for ml-algorithms (#182)
- add java 8 to CI workflow (#194)
- add more UT and IT for rest actions (#192)
- add more UT to client module (#203)
- add more UT for task manager/runner (#206)
- create config and workflow files for release note (#209)
- use 1.3.0 docker to run CI (#212)
Documentation
- Add support for codeowners to repo (#91)
- add how to develop new function doc to readme (#95)
- update license header (#134)
Maintenance
Refactoring
- Merge develop branch into main branch (#87)
- refactor API input/output/URL; add execute API for non-model based algorithm (#93)
- cleanup code and refactor (#106)
- support registering ML objects; refactor ML engine interface (#108)
- refactor persisting ML model (#109)
- refactor transport APIs;fix class cast exception (#127)
- add ML custom exceptions (#133)
- rename tribuo AD algorithm name (#144)