Skip to content

Commit f1768e0

Browse files
RobertLucianRobertLucian
authored
Add GCP support to Cortex (#1655)
1 parent fe19887 commit f1768e0

File tree

116 files changed

+6449
-2137
lines changed

Some content is hidden

Large Commits have some content hidden by default. Use the searchbox below for content that may be hidden.

116 files changed

+6449
-2137
lines changed

‎Makefile‎

Lines changed: 143 additions & 130 deletions
Original file line numberDiff line numberDiff line change
@@ -15,8 +15,10 @@
1515
# limitations under the License.
1616

1717
SHELL := /bin/bash
18-
-include ./dev/config/env.sh
19-
export $(shell sed 's/=.*//' ./dev/config/env.sh 2>/dev/null)
18+
export BASH_ENV=./dev/config/env.sh
19+
20+
# declare all targets as phony to avoid collisions with local files or folders
21+
.PHONY: $(MAKECMDGOALS)
2022

2123
#######
2224
# Dev #
@@ -25,208 +27,219 @@ export $(shell sed 's/=.*//' ./dev/config/env.sh 2>/dev/null)
2527
# Cortex
2628

2729
# build cli, start local operator, and watch for changes
28-
.PHONY: devstart
29-
devstart:
30-
@$(MAKE) operator-stop || true
31-
@./dev/operator_local.sh || true
30+
devstart-aws:
31+
@$(MAKE) operator-stop-aws || true
32+
@./dev/operator_local.sh --aws || true
33+
devstart-gcp:
34+
@$(MAKE) operator-stop-gcp || true
35+
@./dev/operator_local.sh --gcp || true
3236

33-
.PHONY: cli
3437
cli:
3538
@mkdir -p ./bin
3639
@go build -o ./bin/cortex ./cli
3740

3841
# build cli and watch for changes
39-
.PHONY: cli-watch
4042
cli-watch:
4143
@rerun -watch ./pkg ./cli -run sh -c "clear && echo 'building cli...' && go build -o ./bin/cortex ./cli && clear && echo '\033[1;32mCLI built\033[0m'" || true
4244

4345
# start local operator and watch for changes
44-
.PHONY: operator-local
45-
operator-local:
46-
@$(MAKE) operator-stop || true
47-
@./dev/operator_local.sh --operator-only || true
48-
49-
# configure kubectl to point to the cluster specified in dev/config/cluster.yaml
50-
.PHONY: kubectl
51-
kubectl:
52-
@eval $$(python3 ./manager/cluster_config_env.py ./dev/config/cluster.yaml) && eksctl utils write-kubeconfig --cluster="$$CORTEX_CLUSTER_NAME" --region="$$CORTEX_REGION" | grep -v "saved kubeconfig as" | grep -v "using region" | grep -v "eksctl version" || true
53-
54-
.PHONY: cluster-up
55-
cluster-up:
56-
@$(MAKE) registry-all
46+
operator-local-aws:
47+
@$(MAKE) operator-stop-aws || true
48+
@./dev/operator_local.sh --operator-only --aws || true
49+
operator-local-gcp:
50+
@$(MAKE) operator-stop-gcp || true
51+
@./dev/operator_local.sh --operator-only --gcp || true
52+
53+
# configure kubectl to point to the cluster specified in dev/config/cluster-[aws|gcp].yaml
54+
kubectl-aws:
55+
@eval $$(python3 ./manager/cluster_config_env.py ./dev/config/cluster-aws.yaml) && eksctl utils write-kubeconfig --cluster="$$CORTEX_CLUSTER_NAME" --region="$$CORTEX_REGION" | grep -v "saved kubeconfig as" | grep -v "using region" | grep -v "eksctl version" || true
56+
kubectl-gcp:
57+
@eval $$(python3 ./manager/cluster_config_env.py ./dev/config/cluster-gcp.yaml) && gcloud container clusters get-credentials "$$CORTEX_CLUSTER_NAME" --zone "$$CORTEX_ZONE" --project "$$CORTEX_PROJECT" 2>&1 | grep -v "Fetching cluster" | grep -v "kubeconfig entry generated" || true
58+
59+
cluster-up-aws:
60+
@$(MAKE) images-all-aws
61+
@$(MAKE) cli
62+
@kill $(shell pgrep -f rerun) >/dev/null 2>&1 || true
63+
@eval $$(python3 ./manager/cluster_config_env.py ./dev/config/cluster-aws.yaml) && ./bin/cortex cluster up --config=./dev/config/cluster-aws.yaml --configure-env="$$CORTEX_CLUSTER_NAME-aws" --aws-key=$$AWS_ACCESS_KEY_ID --aws-secret=$$AWS_SECRET_ACCESS_KEY --cluster-aws-key=$$CLUSTER_AWS_ACCESS_KEY_ID --cluster-aws-secret=$$CLUSTER_AWS_SECRET_ACCESS_KEY && ./bin/cortex env default "$$CORTEX_CLUSTER_NAME-aws"
64+
@$(MAKE) kubectl-aws
65+
cluster-up-gcp:
66+
@$(MAKE) images-all-gcp
5767
@$(MAKE) cli
5868
@kill $(shell pgrep -f rerun) >/dev/null 2>&1 || true
59-
@eval $$(python3 ./manager/cluster_config_env.py ./dev/config/cluster.yaml) && ./bin/cortex cluster up --config=./dev/config/cluster.yaml --configure-env="$$CORTEX_CLUSTER_NAME" --aws-key=$$AWS_ACCESS_KEY_ID --aws-secret=$$AWS_SECRET_ACCESS_KEY --cluster-aws-key=$$CLUSTER_AWS_ACCESS_KEY_ID --cluster-aws-secret=$$CLUSTER_AWS_SECRET_ACCESS_KEY && ./bin/cortex env default "$$CORTEX_CLUSTER_NAME"
60-
@$(MAKE) kubectl
69+
@eval $$(python3 ./manager/cluster_config_env.py ./dev/config/cluster-gcp.yaml) && ./bin/cortex cluster-gcp up --config=./dev/config/cluster-gcp.yaml --configure-env="$$CORTEX_CLUSTER_NAME-gcp" && ./bin/cortex env default "$$CORTEX_CLUSTER_NAME-gcp"
70+
@$(MAKE) kubectl-gcp
6171

62-
.PHONY: cluster-up-y
63-
cluster-up-y:
64-
@$(MAKE) registry-all
72+
cluster-up-aws-y:
73+
@$(MAKE) images-all-aws
6574
@$(MAKE) cli
6675
@kill $(shell pgrep -f rerun) >/dev/null 2>&1 || true
67-
@eval $$(python3 ./manager/cluster_config_env.py ./dev/config/cluster.yaml) && ./bin/cortex cluster up --config=./dev/config/cluster.yaml --configure-env="$$CORTEX_CLUSTER_NAME" --aws-key=$$AWS_ACCESS_KEY_ID --aws-secret=$$AWS_SECRET_ACCESS_KEY --cluster-aws-key=$$CLUSTER_AWS_ACCESS_KEY_ID --cluster-aws-secret=$$CLUSTER_AWS_SECRET_ACCESS_KEY --yes && ./bin/cortex env default "$$CORTEX_CLUSTER_NAME"
68-
@$(MAKE) kubectl
76+
@eval $$(python3 ./manager/cluster_config_env.py ./dev/config/cluster-aws.yaml) && ./bin/cortex cluster up --config=./dev/config/cluster-aws.yaml --configure-env="$$CORTEX_CLUSTER_NAME-aws" --aws-key=$$AWS_ACCESS_KEY_ID --aws-secret=$$AWS_SECRET_ACCESS_KEY --cluster-aws-key=$$CLUSTER_AWS_ACCESS_KEY_ID --cluster-aws-secret=$$CLUSTER_AWS_SECRET_ACCESS_KEY --yes && ./bin/cortex env default "$$CORTEX_CLUSTER_NAME-aws"
77+
@$(MAKE) kubectl-aws
78+
cluster-up-gcp-y:
79+
@$(MAKE) images-all-gcp
80+
@$(MAKE) cli
81+
@kill $(shell pgrep -f rerun) >/dev/null 2>&1 || true
82+
@eval $$(python3 ./manager/cluster_config_env.py ./dev/config/cluster-gcp.yaml) && ./bin/cortex cluster-gcp up --config=./dev/config/cluster-gcp.yaml --configure-env="$$CORTEX_CLUSTER_NAME-gcp" --yes && ./bin/cortex env default "$$CORTEX_CLUSTER_NAME-gcp"
83+
@$(MAKE) kubectl-gcp
6984

70-
.PHONY: cluster-down
71-
cluster-down:
72-
@$(MAKE) manager-local
85+
cluster-down-aws:
86+
@$(MAKE) images-manager-local
87+
@$(MAKE) cli
88+
@kill $(shell pgrep -f rerun) >/dev/null 2>&1 || true
89+
@./bin/cortex cluster down --config=./dev/config/cluster-aws.yaml --aws-key=$$AWS_ACCESS_KEY_ID --aws-secret=$$AWS_SECRET_ACCESS_KEY
90+
cluster-down-gcp:
91+
@$(MAKE) images-manager-local
7392
@$(MAKE) cli
7493
@kill $(shell pgrep -f rerun) >/dev/null 2>&1 || true
75-
@./bin/cortex cluster down --config=./dev/config/cluster.yaml --aws-key=$$AWS_ACCESS_KEY_ID --aws-secret=$$AWS_SECRET_ACCESS_KEY
94+
@./bin/cortex cluster-gcp down --config=./dev/config/cluster-gcp.yaml
7695

77-
.PHONY: cluster-down-y
78-
cluster-down-y:
79-
@$(MAKE) manager-local
96+
cluster-down-aws-y:
97+
@$(MAKE) images-manager-local
8098
@$(MAKE) cli
8199
@kill $(shell pgrep -f rerun) >/dev/null 2>&1 || true
82-
@./bin/cortex cluster down --config=./dev/config/cluster.yaml --aws-key=$$AWS_ACCESS_KEY_ID --aws-secret=$$AWS_SECRET_ACCESS_KEY --yes
100+
@./bin/cortex cluster down --config=./dev/config/cluster-aws.yaml --aws-key=$$AWS_ACCESS_KEY_ID --aws-secret=$$AWS_SECRET_ACCESS_KEY --yes
101+
cluster-down-gcp-y:
102+
@$(MAKE) images-manager-local
103+
@$(MAKE) cli
104+
@kill $(shell pgrep -f rerun) >/dev/null 2>&1 || true
105+
@./bin/cortex cluster-gcp down --config=./dev/config/cluster-gcp.yaml --yes
83106

84-
.PHONY: cluster-info
85-
cluster-info:
86-
@$(MAKE) manager-local
107+
cluster-info-aws:
108+
@$(MAKE) images-manager-local
87109
@$(MAKE) cli
88-
@eval $$(python3 ./manager/cluster_config_env.py ./dev/config/cluster.yaml) && ./bin/cortex cluster info --config=./dev/config/cluster.yaml --configure-env="$$CORTEX_CLUSTER_NAME" --aws-key=$$AWS_ACCESS_KEY_ID --aws-secret=$$AWS_SECRET_ACCESS_KEY && ./bin/cortex env default "$$CORTEX_CLUSTER_NAME"
110+
@eval $$(python3 ./manager/cluster_config_env.py ./dev/config/cluster-aws.yaml) && ./bin/cortex cluster info --config=./dev/config/cluster-aws.yaml --configure-env="$$CORTEX_CLUSTER_NAME-aws" --aws-key=$$AWS_ACCESS_KEY_ID --aws-secret=$$AWS_SECRET_ACCESS_KEY && ./bin/cortex env default "$$CORTEX_CLUSTER_NAME-aws"
111+
cluster-info-gcp:
112+
@$(MAKE) images-manager-local
113+
@$(MAKE) cli
114+
@eval $$(python3 ./manager/cluster_config_env.py ./dev/config/cluster-gcp.yaml) && ./bin/cortex cluster-gcp info --config=./dev/config/cluster-gcp.yaml --configure-env="$$CORTEX_CLUSTER_NAME-gcp" && ./bin/cortex env default "$$CORTEX_CLUSTER_NAME-gcp"
89115

90-
.PHONY: cluster-configure
91-
cluster-configure:
92-
@$(MAKE) registry-all
116+
cluster-configure-aws:
117+
@$(MAKE) images-all-aws
93118
@$(MAKE) cli
94119
@kill $(shell pgrep -f rerun) >/dev/null 2>&1 || true
95-
@eval $$(python3 ./manager/cluster_config_env.py ./dev/config/cluster.yaml) && ./bin/cortex cluster configure --config=./dev/config/cluster.yaml --configure-env="$$CORTEX_CLUSTER_NAME" --aws-key=$$AWS_ACCESS_KEY_ID --aws-secret=$$AWS_SECRET_ACCESS_KEY --cluster-aws-key=$$CLUSTER_AWS_ACCESS_KEY_ID --cluster-aws-secret=$$CLUSTER_AWS_SECRET_ACCESS_KEY && ./bin/cortex env default "$$CORTEX_CLUSTER_NAME"
96-
97-
.PHONY: cluster-configure-y
98-
cluster-configure-y:
99-
@$(MAKE) registry-all
120+
@eval $$(python3 ./manager/cluster_config_env.py ./dev/config/cluster-aws.yaml) && ./bin/cortex cluster configure --config=./dev/config/cluster-aws.yaml --configure-env="$$CORTEX_CLUSTER_NAME-aws" --aws-key=$$AWS_ACCESS_KEY_ID --aws-secret=$$AWS_SECRET_ACCESS_KEY --cluster-aws-key=$$CLUSTER_AWS_ACCESS_KEY_ID --cluster-aws-secret=$$CLUSTER_AWS_SECRET_ACCESS_KEY && ./bin/cortex env default "$$CORTEX_CLUSTER_NAME-aws"
121+
# cluster-configure-gcp:
122+
# @$(MAKE) images-all-gcp
123+
# @$(MAKE) cli
124+
# @kill $(shell pgrep -f rerun) >/dev/null 2>&1 || true
125+
# @eval $$(python3 ./manager/cluster_config_env.py ./dev/config/cluster-gcp.yaml) && ./bin/cortex cluster-gcp configure --config=./dev/config/cluster-gcp.yaml --configure-env="$$CORTEX_CLUSTER_NAME-gcp" && ./bin/cortex env default "$$CORTEX_CLUSTER_NAME-gcp"
126+
127+
cluster-configure-aws-y:
128+
@$(MAKE) images-all-aws
100129
@$(MAKE) cli
101130
@kill $(shell pgrep -f rerun) >/dev/null 2>&1 || true
102-
@eval $$(python3 ./manager/cluster_config_env.py ./dev/config/cluster.yaml) && ./bin/cortex cluster configure --config=./dev/config/cluster.yaml --configure-env="$$CORTEX_CLUSTER_NAME" --aws-key=$$AWS_ACCESS_KEY_ID --aws-secret=$$AWS_SECRET_ACCESS_KEY --cluster-aws-key=$$CLUSTER_AWS_ACCESS_KEY_ID --cluster-aws-secret=$$CLUSTER_AWS_SECRET_ACCESS_KEY --yes && ./bin/cortex env default "$$CORTEX_CLUSTER_NAME"
131+
@eval $$(python3 ./manager/cluster_config_env.py ./dev/config/cluster-aws.yaml) && ./bin/cortex cluster configure --config=./dev/config/cluster-aws.yaml --configure-env="$$CORTEX_CLUSTER_NAME-aws" --aws-key=$$AWS_ACCESS_KEY_ID --aws-secret=$$AWS_SECRET_ACCESS_KEY --cluster-aws-key=$$CLUSTER_AWS_ACCESS_KEY_ID --cluster-aws-secret=$$CLUSTER_AWS_SECRET_ACCESS_KEY --yes && ./bin/cortex env default "$$CORTEX_CLUSTER_NAME-aws"
132+
# cluster-configure-gcp-y:
133+
# @$(MAKE) images-all-gcp
134+
# @$(MAKE) cli
135+
# @kill $(shell pgrep -f rerun) >/dev/null 2>&1 || true
136+
# @eval $$(python3 ./manager/cluster_config_env.py ./dev/config/cluster-gcp.yaml) && ./bin/cortex cluster-gcp configure --config=./dev/config/cluster-gcp.yaml --configure-env="$$CORTEX_CLUSTER_NAME-gcp" --yes && ./bin/cortex env default "$$CORTEX_CLUSTER_NAME-gcp"
103137

104138
# stop the in-cluster operator
105-
.PHONY: operator-stop
106-
operator-stop:
107-
@$(MAKE) kubectl
139+
operator-stop-aws:
140+
@$(MAKE) kubectl-aws
141+
@kubectl delete --namespace=default --ignore-not-found=true deployment operator
142+
operator-stop-gcp:
143+
@$(MAKE) kubectl-gcp
108144
@kubectl delete --namespace=default --ignore-not-found=true deployment operator
109145

110146
# Docker images
111147

112-
.PHONY: registry-all
113-
registry-all:
114-
@./dev/registry.sh update all
115-
.PHONY: registry-all-local
116-
registry-all-local:
117-
@./dev/registry.sh update all --skip-push
118-
.PHONY: registry-all-slim
119-
registry-all-slim:
120-
@./dev/registry.sh update all --include-slim
121-
.PHONY: registry-all-slim-local
122-
registry-all-slim-local:
123-
@./dev/registry.sh update all --include-slim --skip-push
124-
.PHONY: registry-all-local-slim
125-
registry-all-local-slim:
126-
@./dev/registry.sh update all --include-slim --skip-push
127-
128-
.PHONY: registry-dev
129-
registry-dev:
130-
@./dev/registry.sh update dev
131-
.PHONY: registry-dev-local
132-
registry-dev-local:
133-
@./dev/registry.sh update dev --skip-push
134-
.PHONY: registry-dev-slim
135-
registry-dev-slim:
136-
@./dev/registry.sh update dev --include-slim
137-
.PHONY: registry-dev-slim-local
138-
registry-dev-slim-local:
139-
@./dev/registry.sh update dev --include-slim --skip-push
140-
.PHONY: registry-dev-local-slim
141-
registry-dev-local-slim:
142-
@./dev/registry.sh update dev --include-slim --skip-push
143-
144-
.PHONY: registry-api
145-
registry-api:
146-
@./dev/registry.sh update api
147-
.PHONY: registry-api-local
148-
registry-api-local:
149-
@./dev/registry.sh update api --skip-push
150-
.PHONY: registry-api-slim
151-
registry-api-slim:
152-
@./dev/registry.sh update api --include-slim
153-
.PHONY: registry-api-slim-local
154-
registry-api-slim-local:
155-
@./dev/registry.sh update api --include-slim --skip-push
156-
.PHONY: registry-api-local-slim
157-
registry-api-local-slim:
158-
@./dev/registry.sh update api --include-slim --skip-push
159-
160-
.PHONY: registry-create
161-
registry-create:
162-
@./dev/registry.sh create
163-
164-
.PHONY: registry-clean
165-
registry-clean:
166-
@./dev/registry.sh clean
167-
168-
.PHONY: manager-local
169-
manager-local:
170-
@./dev/registry.sh update-manager-local
148+
images-all-aws:
149+
@./dev/registry.sh update all -p aws
150+
images-all-gcp:
151+
@./dev/registry.sh update all -p gcp
152+
images-all-local:
153+
@./dev/registry.sh update all -p local
154+
images-all-slim-aws:
155+
@./dev/registry.sh update all -p aws --include-slim
156+
images-all-slim-gcp:
157+
@./dev/registry.sh update all -p gcp --include-slim
158+
images-all-slim-local:
159+
@./dev/registry.sh update all -p local --include-slim
160+
161+
images-dev-aws:
162+
@./dev/registry.sh update dev -p aws
163+
images-dev-gcp:
164+
@./dev/registry.sh update dev -p gcp
165+
images-dev-local:
166+
@./dev/registry.sh update dev -p local
167+
images-dev-slim-aws:
168+
@./dev/registry.sh update dev -p aws --include-slim
169+
images-dev-slim-gcp:
170+
@./dev/registry.sh update dev -p gcp --include-slim
171+
images-dev-slim-local:
172+
@./dev/registry.sh update dev -p local --include-slim
173+
174+
images-api-aws:
175+
@./dev/registry.sh update api -p aws
176+
images-api-gcp:
177+
@./dev/registry.sh update api -p gcp
178+
images-api-local:
179+
@./dev/registry.sh update api -p local
180+
images-api-slim-aws:
181+
@./dev/registry.sh update api -p aws --include-slim
182+
images-api-slim-gcp:
183+
@./dev/registry.sh update api -p gcp --include-slim
184+
images-api-slim-local:
185+
@./dev/registry.sh update api -p local --include-slim
186+
187+
images-manager-local:
188+
@./dev/registry.sh update-single manager -p local
189+
images-iris-local:
190+
@./dev/registry.sh update-single python-predictor-cpu -p local
191+
images-iris-aws:
192+
@./dev/registry.sh update-single python-predictor-cpu -p aws
193+
images-iris-gcp:
194+
@./dev/registry.sh update-single python-predictor-cpu -p gcp
195+
196+
registry-create-aws:
197+
@./dev/registry.sh create -p aws
198+
199+
registry-clean-aws:
200+
@./dev/registry.sh clean -p aws
201+
registry-clean-local:
202+
@./dev/registry.sh clean -p local
171203

172204
# Misc
173205

174-
.PHONY: aws-clear-bucket
175-
aws-clear-bucket:
176-
@./dev/aws.sh clear-bucket
177-
178-
.PHONY: tools
179206
tools:
180207
@go get -u -v golang.org/x/lint/golint
181208
@go get -u -v github.com/VojtechVitek/rerun/cmd/rerun
182209
@python3 -m pip install black 'pydoc-markdown>=3.0.0,<4.0.0'
183210
@if [[ "$$OSTYPE" == "darwin"* ]]; then brew install parallel; elif [[ "$$OSTYPE" == "linux"* ]]; then sudo apt-get install -y parallel; else echo "your operating system is not supported"; fi
184211

185-
.PHONY: format
186212
format:
187213
@./dev/format.sh
188214

189215
#########
190216
# Tests #
191217
#########
192218

193-
.PHONY: test
194219
test:
195220
@./build/test.sh
196221

197-
.PHONY: test-go
198222
test-go:
199223
@./build/test.sh go
200224

201-
.PHONY: test-python
202225
test-python:
203226
@./build/test.sh python
204227

205-
.PHONY: lint
206228
lint:
207229
@./build/lint.sh
208230

209-
.PHONY: test-examples
210-
test-examples:
211-
@$(MAKE) registry-all
212-
@./build/test-examples.sh
213-
214231
###############
215232
# CI Commands #
216233
###############
217234

218-
.PHONY: ci-build-images
219235
ci-build-images:
220236
@./build/build-images.sh
221237

222-
.PHONY: ci-push-images
223238
ci-push-images:
224239
@./build/push-images.sh
225240

226-
.PHONY: ci-build-cli
227241
ci-build-cli:
228242
@./build/cli.sh
229243

230-
.PHONY: ci-build-and-upload-cli
231244
ci-build-and-upload-cli:
232245
@./build/cli.sh upload

‎build/build-image.sh‎

Lines changed: 2 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -20,7 +20,6 @@ set -euo pipefail
2020
ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")"/.. >/dev/null && pwd)"
2121

2222
CORTEX_VERSION=master
23-
REGISTRY_URL=quay.io
2423

2524
image=$1
2625
dir="${ROOT}/images/${image/-slim}"
@@ -39,8 +38,8 @@ if [ "${image}" == "python-predictor-gpu-slim" ]; then
3938
cudnn=("7" "7" "8" "7" "8" "8" "8")
4039
for i in ${!cudnn[@]}; do
4140
build_args="${build_args} --build-arg CUDA_VERSION=${cuda[$i]} --build-arg CUDNN=${cudnn[$i]}"
42-
docker build "$ROOT" -f $dir/Dockerfile $build_args -t $REGISTRY_URL/cortexlabs/${image}:${CORTEX_VERSION}-cuda${cuda[$i]}-cudnn${cudnn[$i]}
41+
docker build "$ROOT" -f $dir/Dockerfile $build_args -t quay.io/cortexlabs/${image}:${CORTEX_VERSION}-cuda${cuda[$i]}-cudnn${cudnn[$i]}
4342
done
4443
else
45-
docker build "$ROOT" -f $dir/Dockerfile $build_args -t $REGISTRY_URL/cortexlabs/${image}:${CORTEX_VERSION}
44+
docker build "$ROOT" -f $dir/Dockerfile $build_args -t quay.io/cortexlabs/${image}:${CORTEX_VERSION}
4645
fi

0 commit comments

Comments
 (0)