Summary
System is more stable
- Fixed bugs that worked as bottleneck to the system such as #608. It is much more stable to be used in the production environment.
- Updated deprecated dependencies that might have potential security issues.
API Update
- Keras deployer is supported now.
- Several API updates including
delete_versioned_modeanddelete_model_links(#603). (API doc will be updated shortly) - query_frontend http configuration is exposed to query_frontend
Kubernetes Support Update
- Multiple clusters can be running side by side on the same K8s cluster (Multi tenancy support)
- Prometheus RBAC is supported as default, so that you don't need to create the one for Kubernetes setup.
- K8s service type is configurable. It supports such as LoadBalancer. (API doc will be updated shortly)
Changelog
New Features:
- Multi tenancy support (#503)
- Change the RPCServicePort constant to the global configuration (#660)
- Implement delete_model_links API (#654)
- Expose http configurations to clipper_admin (#524)
- Fluentd Logging System Part 1 (#652)
- Rename 'delete_application' to 'unregister_application' in clipper_admin (#653)
- Keras deployer (#594)
- Change k8s ServiceTypes to be changeable (#667)
- Implement delete_versioned_model API (#602)
- #692: Added get_metric_addr to the ClipperConnection (#693)
- Adding RBAC support in kubernetes config (V2) (#694)
Fix & Dependency Update:
- Backport 0.3.0 fixes (#511)
- Fix docker version to be greater than 3.0 (#515)
- Fix 100% CPU utilization issue [V2] (#608)
- Upgrade requirements and pyyaml versions for security issues (#647)
- Fix a bug that ManagementFrontend adds a garbage to http response (#582)
- Fix Python anti pattern (#651)
- change set_num_replicas in docker_container_manager (#616)
- Revert ClipperLibBaseDockerfile (#622)
- Specify PyTorch version in use (#664)
- Use CLIPPER_REGISTRY, CLIPPER_TAG environment variable if existed (#666)
- add urllib3 version restriction to requirements.txt (#679)
- Remove metric log from query frontend (#680)
- Enable TCP keep alive for redox (#717)
- Added a return statement (#697)
- Fix a bug that a new model replica couldn't be registerd with the Clipper's RPCService (#689)
- Updagrade the version of mxnet to 1.4.* for mxnetcontainerdockerfile. (#675)
- Upgrade Tensorflow version(1.6 -> 1.13) (#720)
- Handle trailing slash on predict url (#721)
- Update SimpleWebServer to the latest version(29 Apr, 2019 - d6985ed6) (#691)
- Print necessary stackstace information when crashed (#700)
- Clean up the versioned model's batch_size and metrics when deleting it (#711)
- Update the version of base libraries and OS (#714)
- Add check for external redis ip address to prevent KeyError (#571) (#719)
Infra:
- [CI] Several CI/Infrastructure Improvement & Fix (#595)
- [CI] Cleaning up #595 (#609)
- Try to fix CI (#649)
- [CI] Remove the dependency about Debian jessie backport (#658)
- [CI] Update the apt package list before installing openjdk (#659)
- [CI] Add retry-routine to 'docker push' in clipper_docker.cfg.py (#655)
- Decrease test samples for test_stop_models case (#657)
- Increase sleep time from 500us to 1000us in timers_test to avoid unintended errors (#661)
- Retry when docker push fails (#662)
- Fix some bugs of PySpark integration test & Disable Spark UI (#665)
- Catch more errors when pushing a Docker image (#663)
- [CI] Try to stabilize Travis job (#685)
- [CI] Randomize RPCService's port and handle NoSuchProcess error properly (#690)
- [CI] Recover CI job failures (#699)
- [CI] Fix maintenance scripts (#707)
Doc: