Proto commits in eth-easl/dirigent

These 73 commits are when the Protocol Buffers files have changed:

2025-02-14

Added RequestedGpu and GPUs fields to proto.

2024-12-18

Abolishing NAT. Routing directly to workload Signed-off-by: Lazar Cvetković <l.cvetkovic.997@gmail.com>

2024-12-16

Route manager with update handler and CNI deletion on WN startup Signed-off-by: Lazar Cvetković <l.cvetkovic.997@gmail.com>

Dynamic CIDR allocation Signed-off-by: Lazar Cvetković <l.cvetkovic.997@gmail.com>

Routing distribution on (de)registration and unschedulability Signed-off-by: Lazar Cvetković <l.cvetkovic.997@gmail.com>

2024-10-30

Worker nodes send images on registration.

Use milli-CPU and MiB as units of CPU and memory. This makes the units of CPU and memory resources used throughout the codebase consistent. It also fixes a bug in the Kubernetes placement policy where the requested resources were 100x larger than available, due to a value being expressed in percent but treated as 0-1.

Allow image prepulling during registration. Prepulling makes function registration asynchronous, so this pull request also introduces another endpoint to the registration server that allows us to check whether the registration has finished.

2024-10-29

Program args working - workaround Signed-off-by: Lazar Cvetković <l.cvetkovic.997@gmail.com>

gRPC workload with health and gRPC MUX Signed-off-by: Lazar Cvetković <l.cvetkovic.997@gmail.com>

2024-10-28

Implemented data parallelism support for the control plane.

2024-10-18

Workflow invocation (#238) * Workflow parsing + basic scheduling + rough testing implementation * Added function I/O management to workflows * Fixed test cases + small changes * Added general workflow datastructure + dandelion workflow parsing, processing & exporting to control plane * Added workflow registration protocol implementation * Initial workflow invocation implementation. * Added check that workflow functions are already registered * Added worker support to register a function composition * Added autoscaling and placing for tasks with function compositions + some refactoring of function states to more general service states. * Updated test cases to changes. * Partitioning bugfix, addressed some TODOs for ServiceState * Added function registration request parameter to set the number arguments and returns, partitioning bugfixes, addressed some TODOs for ServiceState * ConcurrentFifo scheduler fix, made scheduler configurable. * Some bugfixes * Partition method now set via config or request. * Added worker runtime to log. * Moved PartitionMethod/SchedulerType parsing into separate functions. * Adjusted fake_snapshot task creation time.

2024-10-10

Added workflow registration protocol implementation

2024-06-27

Propagate container concurrency patch to data plane

2024-05-30

Replace Knative abstractions by Dirigent abstraction in multiscaler / autoscaler

Integrate RPS Metrics in Dirigent

Initial version of busy loop on sandbox startup Signed-off-by: Lazar Cvetković <l.cvetkovic.997@gmail.com>

Use autoscaling configuration in multiscaler.go and improve code quality

Refactor gRPC communication between control plane and data plane

Remove many code smells and improve code quality

2024-04-23

Simply code structure & clean-up old files

2024-03-12

HAProxy redundancy on all nodes with registrations server replicas Signed-off-by: Lazar Cvetković <l.cvetkovic.997@gmail.com>

The documentation is generated from this commit.

2024-02-21

Ported leader election to gRPC. Some tests still not fixed. Signed-off-by: Lazar Cvetković <l.cvetkovic.997@gmail.com>

Integration of leader election into the CP (still buggy) Signed-off-by: Lazar Cvetković <l.cvetkovic.997@gmail.com>

2024-02-06

Implement Dynamic kubernetes policy

2024-01-29

Remove old logic for gRPC requests

2023-11-21

Further instrumenting worker nodes Signed-off-by: Lazar Cvetković <l.cvetkovic.997@gmail.com>

2023-11-16

Draining on the control plane side Signed-off-by: Lazar Cvetković <l.cvetkovic.997@gmail.com>

2023-10-31

Add support for multiple data planes

Service deregistration

Add timestamp check in DB driver

2023-10-16

Readiness probing on the worker node Signed-off-by: Lazar Cvetković <l.cvetkovic.997@gmail.com>

2023-10-12

Instrumention Firecracker VM creation Signed-off-by: Lazar Cvetković <l.cvetkovic.997@gmail.com>

Replacing peer.Context with hardcoded IPs Addressing Francois review partially Signed-off-by: Lazar Cvetković <l.cvetkovic.997@gmail.com>

2023-09-30

Replace int by uint64 in resource_map.go

2023-09-14

Moved scaling method to autoscaling configuration

2023-09-08

Catch container failure Signed-off-by: Lazar Cvetković <l.cvetkovic.997@gmail.com>

Control plane failure handler for container crash Signed-off-by: Lazar Cvetković <l.cvetkovic.997@gmail.com>

Worker node and container failure handlers Create failure events Atomic map cloning bugfixes -- copy by reference Control plane proto file update Downscale data structures consistency bugfix Signed-off-by: Lazar Cvetković <l.cvetkovic.997@gmail.com>

2023-09-04

Endpoints reconstruction - first part

2023-08-29

Remove endpoints from the persistence layer

Add support for integrations tests

Update control plane breakdown with dataplane propagation

Extracted worker_node interface in its own proto file & fixed a few bugs with lock in control_plane.go

2023-08-11

Support for automatic measurement

Control plane overhead breakdown - DB Signed-off-by: Lazar Cvetković <l.cvetkovic.997@gmail.com>

2023-07-27

Add worker node deregistration

Add data plane deregistration

Replace key value by proto serialization

2023-07-11

Remove old cpp code

2023-06-28

Refactoring tests & fix in hardware.go

Send heartbeat to the master node

Add kubernetes placement policy

2023-06-21

Fixing scripts for folder restructuring Signed-off-by: Lazar Cvetković <l.cvetkovic.997@gmail.com>

Folder restructuring Signed-off-by: Lazar Cvetković <l.cvetkovic.997@gmail.com>

2023-06-20

Data plane request tracing Signed-off-by: Lazar Cvetković <l.cvetkovic.997@gmail.com>

Cold start tracing Signed-off-by: Lazar Cvetković <l.cvetkovic.997@gmail.com>

2023-06-14

Cold start delay Logs for debugging Signed-off-by: Lazar Cvetković <l.cvetkovic.997@gmail.com>

Multifunction setup Signed-off-by: Lazar Cvetković <l.cvetkovic.997@gmail.com>

2023-06-07

No hardcoded data plane address in the control plane Signed-off-by: Lazar Cvetković <l.cvetkovic.997@gmail.com>

2023-06-06

Avoiding port reuse Signed-off-by: Lazar Cvetković <l.cvetkovic.997@gmail.com>

MaxScaleDown rate added Panic mode exit bugfix Signed-off-by: Lazar Cvetković <l.cvetkovic.997@gmail.com>

Knative autoscaling integration with bugfixing Signed-off-by: Lazar Cvetković <l.cvetkovic.997@gmail.com>

2023-06-02

Primitive downscaling Signed-off-by: Lazar Cvetković <l.cvetkovic.997@gmail.com>

Downscaling mechanism with random eviction policy Single port per container Signed-off-by: Lazar Cvetković <l.cvetkovic.997@gmail.com>

2023-06-01

Deployment gRPC control plane call Signed-off-by: Lazar Cvetković <l.cvetkovic.997@gmail.com>

Delete sandbox call Signed-off-by: Lazar Cvetković <l.cvetkovic.997@gmail.com>

2023-05-31

Primitive cold start mechanism Signed-off-by: Lazar Cvetković <l.cvetkovic.997@gmail.com>

Node registration and heartbeat Cold start bugfix Signed-off-by: Lazar Cvetković <l.cvetkovic.997@gmail.com>

Protobuf/gRPC handlers for CP/DP calls Control <-> data plane connectivity Scale from zero on request path gRPC connection helper code Signed-off-by: Lazar Cvetković <l.cvetkovic.997@gmail.com>

2023-05-23

Moved CPP code to control plane folder Signed-off-by: Lazar Cvetković <l.cvetkovic.997@gmail.com>

gRPC proxy and cold start handlers Signed-off-by: Lazar Cvetković <l.cvetkovic.997@gmail.com>

2023-05-15

Initial commit Signed-off-by: Lazar Cvetković <l.cvetkovic.997@gmail.com>