feat: gpu support

Signed-off-by: Artur Troian <[email protected]>

.env

@@ -1,4 +1,4 @@
-GOLANG_VERSION=1.19.2
+GOLANG_VERSION=1.20.2
 KINDEST_VERSION=v1.24.7
 GORELEASER_VERSION=v1.11.4
 GO111MODULE=on

.gitignore

@@ -26,3 +26,5 @@ coverage.txt
 
 go.work
 go.work.sum
+
+.devenv

.goreleaser-docker-amd64.yaml → .goreleaser-docker.yaml

@@ -8,6 +8,7 @@ builds:
     main: ./cmd/provider-services
     goarch:
       - amd64
+      - arm64
     goos:
       - linux
     flags:
@@ -17,6 +18,8 @@ builds:
     ldflags:
       - "{{ .Env.BUILD_VARS }}"
       - "{{ .Env.STRIP_FLAGS }}"
+archives:
+ - format: binary
 dockers:
   - dockerfile: Dockerfile
     use: buildx
@@ -33,3 +36,18 @@ dockers:
       - --label=org.opencontainers.image.revision={{ .FullCommit }}
     image_templates:
       - '{{ .Env.DOCKER_IMAGE }}:latest-amd64'
+  - dockerfile: Dockerfile
+    use: buildx
+    goos: linux
+    goarch: arm64
+    build_flag_templates:
+      - --platform=linux/arm64
+      - --label=org.opencontainers.image.title={{ .ProjectName }}
+      - --label=org.opencontainers.image.description={{ .ProjectName }}
+      - --label=org.opencontainers.image.url={{.GitURL}}
+      - --label=org.opencontainers.image.source={{.GitURL}}
+      - --label=org.opencontainers.image.version={{ .Version }}
+      - --label=org.opencontainers.image.created={{ time "2006-01-02T15:04:05Z07:00" }}
+      - --label=org.opencontainers.image.revision={{ .FullCommit }}
+    image_templates:
+      - '{{ .Env.DOCKER_IMAGE }}:latest-arm64'

.goreleaser.yaml

@@ -164,18 +164,18 @@ docker_manifests:
 
 nfpms:
   - id: w/version
-    vendor: "Overclock Labs Inc."
-    homepage: "https://ovrclk.com"
-    maintainer: "Overclock Labs Inc. <[email protected]>"
+    vendor: "Akash Network"
+    homepage: "https://akash.network"
+    maintainer: "Akash Network <[email protected]>"
     file_name_template: "provider-services_{{ .Version }}_{{ .Os }}_{{ .Arch }}"
     formats:
       - deb
       - rpm
     license: Apache 2.0
   - id: wo/version
-    vendor: "Overclock Labs Inc."
-    homepage: "https://ovrclk.com"
-    maintainer: "Overclock Labs Inc. <[email protected]>"
+    vendor: "Akash Network"
+    homepage: "https://akash.network"
+    maintainer: "Akash Network <[email protected]>"
     file_name_template: "provider-services_{{ .Os }}_{{ .Arch }}"
     formats:
       - deb

Makefile

@@ -13,7 +13,7 @@ DOCKER_RUN             := docker run --rm -v $(shell pwd):/workspace -w /workspa
 GOLANGCI_LINT_RUN      := $(GOLANGCI_LINT) run
 LINT                    = $(GOLANGCI_LINT_RUN) ./... --disable-all --deadline=5m --enable
 
-GORELEASER_CONFIG       = .goreleaser.yaml
+GORELEASER_CONFIG      ?= .goreleaser.yaml
 
 GIT_HEAD_COMMIT_LONG   := $(shell git log -1 --format='%H')
 GIT_HEAD_COMMIT_SHORT  := $(shell git rev-parse --short HEAD)

_run/common-commands.mk

@@ -1,4 +1,4 @@
-KEY_NAME   ?= main
+KEY_NAME ?= main
 
 KEY_ADDRESS          ?= $(shell $(AKASH) $(KEY_OPTS) keys show "$(KEY_NAME)" -a)
 PROVIDER_KEY_NAME    ?= provider

_run/common-kind.mk

@@ -4,15 +4,15 @@
 # image's name in GH Actions.
 export KIND_NAME ?= $(shell basename $$PWD)
 
-KIND_CREATE       := $(AP_RUN_DIR)/.kind-create
-
 ifeq (, $(KINDEST_VERSION))
 $(error "KINDEST_VERSION is not set")
 endif
 
 KIND_IMG         ?= kindest/node:$(KINDEST_VERSION)
-
 K8S_CONTEXT      ?= $(shell kubectl config current-context)
+SETUP_KIND       := "$(AP_ROOT)/script/setup-kind.sh"
+KIND_CONFIG      ?= default
+KIND_CONFIG_FILE := $(shell "$(SETUP_KIND)" config-file $$PWD $(KIND_CONFIG))
 
 KIND_HTTP_PORT = $(shell docker inspect \
     --type container "$(KIND_NAME)-control-plane" \
@@ -33,38 +33,6 @@ KIND_K8S_IP = $(shell docker inspect \
 KIND_PORT_BINDINGS = $(shell docker inspect "$(KIND_NAME)-control-plane" \
     --format '{{index .NetworkSettings.Ports "80/tcp" 0 "HostPort"}}' 2> /dev/null)
 
-SETUP_KIND                := "$(AP_ROOT)/script/setup-kind.sh"
-KIND_CONFIG               ?= default
-KIND_CONFIG_FILE          := $(shell "$(SETUP_KIND)" config-file $$PWD $(KIND_CONFIG))
-
-include ../common-kustomize.mk
-
-# certain targets need to use bash
-# detect where bash is installed
-# use akash-node-ready target as example
-BASH_PATH                 := $(shell which bash)
-
-INGRESS_CONFIG_PATH       ?= ../ingress-nginx.yaml
-CALICO_MANIFEST           ?= https://docs.projectcalico.org/v3.8/manifests/calico.yaml
-
-ifeq ($(AKASH_SRC_IS_LOCAL), true)
-AKASH_DOCKER_IMAGE        ?= ghcr.io/akash-network/node:latest-$(UNAME_ARCH)
-else
-AKASH_DOCKER_IMAGE        ?= ghcr.io/akash-network/node:$(AKASH_VERSION)-$(UNAME_ARCH)
-endif
-
-DOCKER_IMAGE              ?= ghcr.io/akash-network/provider:latest-$(UNAME_ARCH)
-
-PROVIDER_HOSTNAME         ?= localhost
-PROVIDER_HOST              = $(PROVIDER_HOSTNAME):$(KIND_HTTP_PORT)
-PROVIDER_ENDPOINT          = http://$(PROVIDER_HOST)
-
-METALLB_CONFIG_PATH       ?= ../metallb.yaml
-METALLB_IP_CONFIG_PATH    ?= ../kind-config-metal-lb-ip.yaml
-METALLB_SERVICE_PATH      ?= ../../_docs/provider/kube/metallb-service.yaml
-
-KIND_ROLLOUT_TIMEOUT      ?= 90
-
 .PHONY: app-http-port
 app-http-port:
 	@echo $(KIND_HTTP_PORT)
@@ -73,112 +41,14 @@ app-http-port:
 kind-k8s-ip:
 	@echo $(KIND_K8S_IP)
 
-.PHONY: kind-prepare-images
-kind-prepare-images:
-ifneq ($(SKIP_BUILD), true)
-	make -C $(AP_ROOT) docker-image
-ifeq ($(AKASH_SRC_IS_LOCAL), true)
-	make -C $(AKASH_LOCAL_PATH) docker-image
-else
-	docker pull $(AKASH_DOCKER_IMAGE)
-endif
-endif
-
-.PHONY: kind-upload-images
-kind-upload-images: $(KIND)
-ifeq ($(KIND_NAME), single)
-	$(KIND) --name "$(KIND_NAME)" load docker-image "$(AKASH_DOCKER_IMAGE)"
-endif
-	$(KIND) --name "$(KIND_NAME)" load docker-image "$(DOCKER_IMAGE)"
-
 .PHONY: kind-port-bindings
 kind-port-bindings: $(KIND)
 	@echo $(KIND_PORT_BINDINGS)
 
-.PHONY: kind-cluster-setup
-kind-cluster-setup: init \
-	kind-prepare-images \
-	kind-cluster-create \
-	kind-setup-ingress \
-	kind-upload-images \
-	kustomize-init \
-	kustomize-deploy-services \
-	kind-deployments-rollout \
-	kind-setup-$(KIND_NAME)
-
-.PHONY: kind-cluster-setup-e2e
-kind-cluster-setup-e2e: kind-cluster-create kind-cluster-setup-e2e-ci
-
-.PHONY: kind-cluster-setup-e2e-ci
-kind-cluster-setup-e2eci:
-kind-cluster-setup-e2e-ci: \
-	kind-setup-ingress \
-	kind-upload-images \
-	kustomize-init \
-	kustomize-deploy-services \
-	kind-deployments-rollout
-
-$(KIND_CREATE): $(KIND) $(AP_RUN_DIR)
-	$(KIND) create cluster --config "$(KIND_CONFIG_FILE)" --name "$(KIND_NAME)" --image "$(KIND_IMG)"
-	touch $@
-
-.INTERMEDIATE: kind-cluster-create
-kind-cluster-create: $(KIND_CREATE)
+.INTERMEDIATE: kube-cluster-create-kind
+kube-cluster-create-kind: $(KUBE_CREATE)
 
-.PHONY: kind-setup-ingress
-kind-setup-ingress: kind-setup-ingress-$(KIND_CONFIG)
-
-.PHONY: kind-setup-ingress-calico
-kind-setup-ingress-calico:
-	kubectl apply -f "$(CALICO_MANIFEST)"
-	# Calico needs to be managing networking before finishing setup
-	kubectl apply -f "$(INGRESS_CONFIG_PATH)"
-	kubectl rollout status deployment -n ingress-nginx ingress-nginx-controller --timeout=$(KIND_ROLLOUT_TIMEOUT)s
-	kubectl apply -f "$(METALLB_CONFIG_PATH)"
-	kubectl apply -f "$(METALLB_IP_CONFIG_PATH)"
-	kubectl apply -f "$(METALLB_SERVICE_PATH)"
-	"$(SETUP_KIND)" calico-metrics
-
-.PHONY: kind-setup-ingress-default
-kind-setup-ingress-default:
-	kubectl label nodes $(KIND_NAME)-control-plane akash.network/role=ingress
-	kubectl apply -f "$(INGRESS_CONFIG_PATH)"
-	kubectl rollout status deployment -n ingress-nginx ingress-nginx-controller --timeout=$(KIND_ROLLOUT_TIMEOUT)s
-	kubectl apply -f "$(METALLB_CONFIG_PATH)"
-	kubectl apply -f "$(METALLB_IP_CONFIG_PATH)"
-	kubectl apply -f "$(METALLB_SERVICE_PATH)"
-	"$(SETUP_KIND)"
-
-.PHONY: kind-cluster-delete
-kind-cluster-delete: $(KIND)
+.PHONY: kube-cluster-delete-kind
+kube-cluster-delete-kind: $(KIND)
 	$(KIND) delete cluster --name "$(KIND_NAME)"
-	rm -rf $(KIND_CREATE)
-
-.PHONY: kind-status-ingress-%
-kind-status-ingress-%:
-	kubectl rollout status -n akash-services ingress $* --timeout=$(KIND_ROLLOUT_TIMEOUT)s
-
-.PHONY: kind-deployment-rollout-%
-kind-deployment-rollout-%:
-	kubectl -n akash-services rollout status deployment $* --timeout=$(KIND_ROLLOUT_TIMEOUT)s
-	kubectl -n akash-services wait pods -l akash.network/component=$* --for condition=Ready --timeout=$(KIND_ROLLOUT_TIMEOUT)s
-
-.PHONY: akash-node-ready
-akash-node-ready: SHELL=$(BASH_PATH)
-akash-node-ready:
-	@( \
-		max_retry=15; \
-		counter=0; \
-		while [[ counter -lt max_retry ]]; do \
-			read block < <(curl -s $(AKASH_NODE)/status | jq -r '.result.sync_info.latest_block_height' 2> /dev/null); \
-			if [[ $$? -ne 0 || $$block -lt 1 ]]; then \
-				echo "unable to get node status. sleep for 1s"; \
-				((counter++)); \
-				sleep 1; \
-			else \
-				echo "latest block height: $${block}"; \
-				exit 0; \
-			fi \
-		done; \
-		exit 1 \
-	)
+	rm -rf $(KUBE_CREATE)