Skip to content

@simon-mo simon-mo released this Jun 7, 2019 · 9 commits to develop since this release

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_mode and delete_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:

  • Update Basic example README.md (#600)
  • Fix typo (#586)
  • Example (#644)
Assets 2
Jun 7, 2019
Fix package_data in setup.py

@dcrankshaw dcrankshaw released this Jun 6, 2018 · 61 commits to develop since this release

  • commit • Only link models to apps in k8s test the first time model is registered (#290)
  • commit • update benchmark readme (#293)
  • commit • Batch size estimation fix (#294)
  • commit • R Model Deployment (#261)
  • commit • Minor cleanup to rpc.py (#295)
  • commit • Update docker publish script to publish minor version tags (#298)
  • commit • Fix typos in k8s method calls. (#299)
  • commit • Fix broken links (#303)
  • commit • Add CLIPPER_INPUT_TYPE to environment when launching k8s model container. (#304)
  • commit • Fix Issue with ClipperException and Docker not running case (#308)
  • commit • Support InternalIP address for K8S (#316)
  • commit • Support for Generic Tensorflow Models (#319)
  • commit • Converted path to posixpath in python deployer (#318)
  • commit • Fix a broken link in the README.md (#323)
  • commit • Fixed conda object manipulation problem (#297)
  • commit • Support Deployment of SparkML Models (#324)
  • commit • Handle exceptions associated with loading nonexistent libraries (#344)
  • commit • Support deploying Saved Tensorflow Models (#328)
  • commit • Batch predict (#321)
  • commit • Support for PyTorch model (#346)
  • commit • Adding function: Allow users to designate the number of batches (#320)
  • commit • Metrics Monitoring: A minimal viable product (#348)
  • commit • Fix dict unpacking lint error (#356)
  • commit • Support Boost 1.66.0 (#358)
  • commit • Turn off ssl verification to fix conda http error (#369)
  • commit • Fix error (#373)
  • commit • Fix port already bound error in integration tests (#368)
  • commit • Metrics and Monitoring V0.0.2: Model Container, Latencies, and Example (#357)
  • commit • Integration test for fixed-batch-size model deployment (#370)
  • commit • fixed decoding error when when parsing graph_def from string. (#377)
  • commit • Metric Monitoring: Set Prometheus Version Tag (#384)
  • commit • replaced queues in RPC layer with MoodyCamel's lock-free queues (#380)
  • commit • Handle ChunkedEncodingError on check_solvability_write_deps() (#398)
  • commit • Support latest conda version (#385)
  • commit • Metric Monitoring: Add to/Delete from the metric config of Prometheus (#402)
  • commit • Fix flakiness in fixed batch size integration test (#387)
  • commit • StateDB Cache (#350)
  • commit • Explicitly install redis-server when building test container (#410)
  • commit • Added MXNet Deployment Capability (replaces #390) (#405)
  • commit • [Metric] Refactor Frontend-exporter (#413)
  • commit • Improved README with direct links to github profiles (#414)
  • commit • Fix a bug that QueryFrontend adds a garbage to http response (#415)
  • commit • PyTorch + ONNX + Caffe2 Model deployer (#362)
  • commit • results of profiling JSON parsing latency (#418)
  • commit • Fix installation order errors in deployer containers (#426)
  • commit • Container Manager Wait Until Ready (#417)
  • commit • fix libsodium issue (#432)
  • commit • [Metrics] Frontend-exporter fault tolerance quick fix (#429)
  • commit • Model container testing function to clipper admin (Vanilla python) (#394)
  • commit • Upgrade SimpleWebServer to version 2.1.1 (#438)
  • commit • [Metrics] Kubernetes Support & Maintenance (#427)
  • commit • Add ability to install packages + xgboost model deployer + example (#431)
  • commit • Add missing integration tests, clean up dockerfiles, and remove Anaconda dependencies (#435)
  • commit • Implement Metric Server (#440)
  • commit • Fix QueryFrontend bug (#458)
  • commit • Ensure container/model is ready to be queried (#447)
  • commit • Fix hard coded checkpoint prefix for tensorflow deployer. (#464)
  • commit • Add functionality to delete applications (#457)
  • commit • R model serialization fixes (#444)
  • commit • Address k8s del_svc api change (#470)
  • commit • Container heartbeat (#378)
  • commit • Add Grafana image tag to avoid pulling all versions (#471)
  • commit • Shorten input for kube integration test (#480)
  • commit • Temporarily Disable Caffe2 Build & Test (#479)
  • commit • update java version in docker files (#485)
  • commit • Adds support for kubectl proxy in kubernetes container manager (#455)
  • commit • Add Python 3 support (#416)
  • commit • Don't quit on warning when installing R dependencies (#487)
  • commit • Query Frontend Scale-out via Kubernetes Service (#462)
  • commit • Added functionality to query previous version models (#454)
  • commit • Update FindBoost.cmake to support Boost 1.67.0 (#489)
  • commit • Fix an infinite loop of GarbageCollectionThread (#481)
  • commit • Fix issues with model version querying (#497)
  • commit • Issue with Number of Frontend Replicas (#496)
  • commit • fixed bug in querying for specific versions (#493)
  • commit • Performance improvements: RPC, Data management, hashing (#393)
  • commit • Fix the following build error of rpc_service.cpp (#502)
  • commit • Move reading of Clipper env variables out of user code into RPC system code (#505)
  • commit • Add Python 3 support for the remaining model deployers (#488)
  • commit • Example of Querying with An Image (#504)
  • commit • Adaptive Batching v2 (#451)
  • commit • Support k8s namespaces (#500)
  • commit • Add API documentation for model deployers and R models (#498)
  • commit • added readthedocs requirements file (#507)
  • commit • Update readme (#508)
  • commit • Fix Documentation (#510)
  • Assets 2

    @dcrankshaw dcrankshaw released this Oct 3, 2017 · 149 commits to develop since this release

    Release Notes

    Major Changes:

    • Major refactor of the Clipper admin Python library to make it more robust and modular
    • Support for running Clipper on Kubernetes through the KubernetesContainerManager

    Changelog

  • commit • minor fix to unittest script for jenkins (#202)
  • commit • Fix example code in README (#205)
  • commit • removed pip install findspark from run_unittests.sh (#211)
  • commit • fixed backslash escape issue for removing remote containers (#210)
  • commit • Restarting on containers is no longer the default behavior (#213)
  • commit • Base64 decoding for JSON byte data (#214)
  • commit • Allow model versions to be strings (#197)
  • commit • R model deployment (#195)
  • commit • One-line app registration and model deployment (#223)
  • commit • [CLIPPER-227] Fix EWMA behavior for meters (#228)
  • commit • RPC container content fix (#232)
  • commit • minor fix to unittest script for jenkins (#202)
  • commit • Fix example code in README (#205)
  • commit • removed pip install findspark from run_unittests.sh (#211)
  • commit • fixed backslash escape issue for removing remote containers (#210)
  • commit • Restarting on containers is no longer the default behavior (#213)
  • commit • Base64 decoding for JSON byte data (#214)
  • commit • Allow model versions to be strings (#197)
  • commit • R model deployment (#195)
  • commit • One-line app registration and model deployment (#223)
  • commit • [CLIPPER-227] Fix EWMA behavior for meters (#228)
  • commit • RPC container content fix (#232)
  • commit • Wording fix (#234)
  • commit • Fixed broken R container (#247)
  • commit • Handling for internal server errors (#246)
  • commit • Update Python container to correctly run "pip install cryptography" (#243)
  • commit • Updated libmzmq3 to libzmq5 (#248)
  • commit • Benchmarking utilities (#225)
  • commit • App/model links (add only) (#226)
  • commit • Make mgmt frontend check for input type object equality (#250)
  • commit • Save module dependencies not in conda/pip (#237)
  • commit • Pass by value where appropriate (#252)
  • commit • C++ Model Container (#249)
  • commit • Updated README to clarify quickstart (#257)
  • commit • Removed redis cli in favor of clipper.get_model_info (#260)
  • commit • Fix docker image versioning (#264)
  • commit • [CLIPPER-224] Replace boost futures with folly futures (#266)
  • commit • Management frontend: Improve clarity of response messages (#271)
  • commit • Fix histogram overflow errors (#272)
  • commit • [CLIPPER-212] Add CLOCK cache eviction policy (#268)
  • commit • change CODE_VERSION build arg to develop (#273)
  • commit • Make unit tests run in a Docker container (#274)
  • commit • Refactor and clean up Clipper admin python package (#251)
  • commit • remove default label (#279)
  • commit • Installs latest version of pyspark via pip install (#282)
  • commit • Fixes readthedocs documentation (#278)
  • commit • Build new docker images in build_model with in-memory Dockerfile (#280)
  • commit • Update docker build script to optionally publish images to DockerHub (#283)
  • commit • Pick up kube credentials from env variables for k8s integration test (#284)
  • commit • Include k8s yaml files in pip package (#286)
  • Assets 2

    @dcrankshaw dcrankshaw released this Jun 27, 2017 · 188 commits to develop since this release

    This release added three new features as well as number of bugfixes:

    • Deploy R models directly from a Python/R training script
    • Deploy a model and register the corresponding application in a single method
    • Allow model versions to be arbitrary strings. Model versions can now be git hashes, semantic versions, etc. along with integers.
    Assets 2

    @dcrankshaw dcrankshaw released this Jun 5, 2017 · 188 commits to develop since this release

    • Fixes several minor bugs and usability issues.
    • It also adds support for automatic deployment of PySpark models through clipper_admin.Clipper.deploy_pyspark_model.
    Assets 2
    May 23, 2017
    Clipper version 0.1.2
    
    May 22, 2017
    Clipper version 0.1.1
    
    May 22, 2017
    Clipper version 0.1.0
    
    You can’t perform that action at this time.