Allow engine resource requests for engine to be configurable in proto definition for CRD #70
Assignees
Comments
ukclivecox
changed the title
|
Available now. |
agrski
pushed a commit
that referenced
this issue
Dec 2, 2022
* Change model version to unit32 * Initial Agent version changes. Code refactor. * lint and ensure model name is changed * Add unload model functionality to agent to handle versions * Add version cleanup to scheduler * Updates for k8s * lint * small updates * change download to use hash for rclone folder to eventually allow caching of copy sync * Modify protos to remove version and restructure * k8s testing via grpc * Update operator * ensure k8s generation works * Allow multiple vesions in ModelStatus call * Handle terminate and k8s updates * lint * add retry for status update and fix event creation * lint * fix retries for client -> scheduler * Add triton server handler to agent * grpc test * Add Triton * fix server http port typos * Fix Envoy resource bug and add server memory notebook tests * Add triton xgboost example and always copy triton config * lint * Initial Manager reconcile for Server * Add servers to kustomize * Updates for agent start from env and updates to serverconfigs * Add Status Handling for Servers * handle deleted servers and consistent ordering of replicas * Add Server Notification and events * review comments changes * lint * change minReplicas to 1 if not specified * remove unnecesary get model * fix test * fix model reschedule to server on server delete * updated to fix delete server tests in k8s notebook * review comments for client state * fix test after merge * Move channel comms to central hub * improve agent server tests * lint * review comment updates * Fix hub channel close safety * review comment updates * Make addListener signal type of channel * Set channel as send only for config updates * update Makefile to allow servers to be deployed * Update scheduler server deployment (as opposed to operator server deployment) and add debug for filters when they fail in scheduling * lint * Updates from review * fix typos * Update mlserver tests * review comments * updated envoy resource for REST regex and tests in notebook * review comments * update from review * Review updates * Update log-level in Makefile * review comments * fix loglevel debug on agent startup * Make rclone folder hash dependent on model name and srcUri not just srcUri to stop multiple simultaneous syncs for models sharing srcUri * review comments * review comment fixes * Change serverType to non pointer and add tests for empty and unknown server type * Update serverconfigs from review * Fix server svc reconcilliation * Fix typo * review fixes * Fix bug that server is not set when models loaded when an agent connects. Also model status was not updated * Add server status for schedule failed * Change envoy port name to http2 to allow grpc traffic from istio * change REST/gRPCheader matching and REST substitution for envoy requests * notebook updates * review updates * Readded top level SVC for servers as will need it for each metric craping by prometheus * Add missing file * Prometheus monitoring updates * review fixes * review fixes * Add Name() to sorters in scheduler * lint * Update server manifests and generated code * Add model:version to scheduler debug logs * review fixes - pass version to updateLoadedModels * Return error when no model to unload * fix statefulset test * istio and treafik examples * Ambassador examples * update scheduler local notebook * review comments * small todo comment * lint
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
No description provided.
The text was updated successfully, but these errors were encountered: