|
| 1 | +#!/usr/bin/env bash |
| 2 | +# Deploy strata-qa as a container-image Lambda with an IAM-authed Function URL. |
| 3 | +# Prereqs: aws CLI v2, docker, and CURSOR_API_KEY exported (personal or |
| 4 | +# service-account key). Run from the REPO ROOT — the image build context is the |
| 5 | +# repo root, because the image needs both docs/ and strata-qa/. |
| 6 | +set -euo pipefail |
| 7 | + |
| 8 | +AWS_REGION="${AWS_REGION:-us-east-1}" |
| 9 | +FUNCTION_NAME="${FUNCTION_NAME:-strata-qa}" |
| 10 | +ECR_REPO="${ECR_REPO:-strata-qa-lambda}" |
| 11 | +IMAGE_TAG="${IMAGE_TAG:-latest}" |
| 12 | +ARCH="${ARCH:-arm64}" # arm64 | x86_64 |
| 13 | +DOCKER_PLATFORM="linux/${ARCH/x86_64/amd64}" |
| 14 | +MEMORY_MB="${MEMORY_MB:-2048}" |
| 15 | +TIMEOUT_S="${TIMEOUT_S:-120}" |
| 16 | +AGENT_TIMEOUT_MS="${AGENT_TIMEOUT_MS:-90000}" |
| 17 | +# At ~190k tokens per question (NOTES.md), an endpoint with unreserved |
| 18 | +# concurrency is an unbounded spend on the Cursor API -- and a 504 at 90s is |
| 19 | +# exactly the response a client library retries. Cap it. |
| 20 | +RESERVED_CONCURRENCY="${RESERVED_CONCURRENCY:-3}" |
| 21 | +SECRET_NAME="${SECRET_NAME:-strata-qa/cursor-api-key}" |
| 22 | +ROLE_NAME="${ROLE_NAME:-strata-qa-lambda-role}" |
| 23 | + |
| 24 | +: "${CURSOR_API_KEY:?export CURSOR_API_KEY (personal or service-account key) before deploying}" |
| 25 | + |
| 26 | +# The agent's own timeout must fire first, so the handler returns a clean 504 |
| 27 | +# rather than Lambda hard-killing the invocation mid-flight. |
| 28 | +if (( AGENT_TIMEOUT_MS >= TIMEOUT_S * 1000 )); then |
| 29 | + echo "AGENT_TIMEOUT_MS (${AGENT_TIMEOUT_MS}) must be less than TIMEOUT_S (${TIMEOUT_S}s)" >&2 |
| 30 | + exit 1 |
| 31 | +fi |
| 32 | + |
| 33 | +ACCOUNT_ID="$(aws sts get-caller-identity --query Account --output text)" |
| 34 | +ECR_URI="${ACCOUNT_ID}.dkr.ecr.${AWS_REGION}.amazonaws.com/${ECR_REPO}" |
| 35 | +IMAGE_URI="${ECR_URI}:${IMAGE_TAG}" |
| 36 | +GIT_SHA="$(git rev-parse HEAD)" |
| 37 | + |
| 38 | +echo "==> ECR repo" |
| 39 | +aws ecr describe-repositories --repository-names "$ECR_REPO" --region "$AWS_REGION" >/dev/null 2>&1 \ |
| 40 | + || aws ecr create-repository --repository-name "$ECR_REPO" --region "$AWS_REGION" >/dev/null |
| 41 | + |
| 42 | +echo "==> Secret" |
| 43 | +if aws secretsmanager describe-secret --secret-id "$SECRET_NAME" --region "$AWS_REGION" >/dev/null 2>&1; then |
| 44 | + aws secretsmanager put-secret-value --secret-id "$SECRET_NAME" \ |
| 45 | + --secret-string "$CURSOR_API_KEY" --region "$AWS_REGION" >/dev/null |
| 46 | +else |
| 47 | + aws secretsmanager create-secret --name "$SECRET_NAME" \ |
| 48 | + --secret-string "$CURSOR_API_KEY" --region "$AWS_REGION" >/dev/null |
| 49 | +fi |
| 50 | +SECRET_ARN="$(aws secretsmanager describe-secret --secret-id "$SECRET_NAME" \ |
| 51 | + --region "$AWS_REGION" --query ARN --output text)" |
| 52 | + |
| 53 | +echo "==> IAM execution role" |
| 54 | +if ! aws iam get-role --role-name "$ROLE_NAME" >/dev/null 2>&1; then |
| 55 | + aws iam create-role --role-name "$ROLE_NAME" \ |
| 56 | + --assume-role-policy-document '{ |
| 57 | + "Version":"2012-10-17", |
| 58 | + "Statement":[{"Effect":"Allow","Principal":{"Service":"lambda.amazonaws.com"},"Action":"sts:AssumeRole"}] |
| 59 | + }' >/dev/null |
| 60 | + aws iam attach-role-policy --role-name "$ROLE_NAME" \ |
| 61 | + --policy-arn arn:aws:iam::aws:policy/service-role/AWSLambdaBasicExecutionRole >/dev/null |
| 62 | +fi |
| 63 | +# Read just this one secret, nothing else. |
| 64 | +aws iam put-role-policy --role-name "$ROLE_NAME" --policy-name read-cursor-secret \ |
| 65 | + --policy-document "{ |
| 66 | + \"Version\":\"2012-10-17\", |
| 67 | + \"Statement\":[{\"Effect\":\"Allow\",\"Action\":\"secretsmanager:GetSecretValue\",\"Resource\":\"${SECRET_ARN}\"}] |
| 68 | + }" >/dev/null |
| 69 | +ROLE_ARN="$(aws iam get-role --role-name "$ROLE_NAME" --query Role.Arn --output text)" |
| 70 | + |
| 71 | +echo "==> Build & push image (${DOCKER_PLATFORM}, context = repo root)" |
| 72 | +aws ecr get-login-password --region "$AWS_REGION" \ |
| 73 | + | docker login --username AWS --password-stdin "${ACCOUNT_ID}.dkr.ecr.${AWS_REGION}.amazonaws.com" |
| 74 | +docker build -f strata-qa/Dockerfile --platform "$DOCKER_PLATFORM" \ |
| 75 | + --build-arg GIT_SHA="$GIT_SHA" -t "$IMAGE_URI" . |
| 76 | +docker push "$IMAGE_URI" |
| 77 | + |
| 78 | +ENV_VARS="Variables={HOME=/tmp,DOCS_ROOT=/var/task,QA_LOG_DIR=/tmp/qa,AGENT_TIMEOUT_MS=${AGENT_TIMEOUT_MS},CURSOR_API_KEY_SECRET_ID=${SECRET_ARN}}" |
| 79 | + |
| 80 | +echo "==> Lambda function" |
| 81 | +if aws lambda get-function --function-name "$FUNCTION_NAME" --region "$AWS_REGION" >/dev/null 2>&1; then |
| 82 | + aws lambda update-function-code --function-name "$FUNCTION_NAME" \ |
| 83 | + --image-uri "$IMAGE_URI" --region "$AWS_REGION" >/dev/null |
| 84 | + aws lambda wait function-updated --function-name "$FUNCTION_NAME" --region "$AWS_REGION" |
| 85 | + aws lambda update-function-configuration --function-name "$FUNCTION_NAME" \ |
| 86 | + --timeout "$TIMEOUT_S" --memory-size "$MEMORY_MB" \ |
| 87 | + --environment "$ENV_VARS" --region "$AWS_REGION" >/dev/null |
| 88 | + aws lambda wait function-updated --function-name "$FUNCTION_NAME" --region "$AWS_REGION" |
| 89 | +else |
| 90 | + # IAM role propagation can lag; retry create briefly. |
| 91 | + for i in 1 2 3 4 5; do |
| 92 | + aws lambda create-function --function-name "$FUNCTION_NAME" \ |
| 93 | + --package-type Image --code "ImageUri=${IMAGE_URI}" \ |
| 94 | + --role "$ROLE_ARN" --architectures "$ARCH" \ |
| 95 | + --timeout "$TIMEOUT_S" --memory-size "$MEMORY_MB" \ |
| 96 | + --environment "$ENV_VARS" --region "$AWS_REGION" >/dev/null && break |
| 97 | + echo " create failed (role may still be propagating); retry $i..." && sleep 10 |
| 98 | + done |
| 99 | +fi |
| 100 | +aws lambda wait function-active-v2 --function-name "$FUNCTION_NAME" --region "$AWS_REGION" |
| 101 | + |
| 102 | +echo "==> Reserved concurrency (${RESERVED_CONCURRENCY}) — cost ceiling" |
| 103 | +aws lambda put-function-concurrency --function-name "$FUNCTION_NAME" \ |
| 104 | + --reserved-concurrent-executions "$RESERVED_CONCURRENCY" --region "$AWS_REGION" >/dev/null |
| 105 | + |
| 106 | +echo "==> Function URL (AWS_IAM auth)" |
| 107 | +aws lambda get-function-url-config --function-name "$FUNCTION_NAME" --region "$AWS_REGION" >/dev/null 2>&1 \ |
| 108 | + || aws lambda create-function-url-config --function-name "$FUNCTION_NAME" \ |
| 109 | + --auth-type AWS_IAM --region "$AWS_REGION" >/dev/null |
| 110 | +FUNCTION_URL="$(aws lambda get-function-url-config --function-name "$FUNCTION_NAME" \ |
| 111 | + --region "$AWS_REGION" --query FunctionUrl --output text)" |
| 112 | + |
| 113 | +echo "==> Deployed ${GIT_SHA:0:12}. Function URL: ${FUNCTION_URL}" |
| 114 | +echo " Invoke with SigV4, e.g.:" |
| 115 | +echo " awscurl --service lambda --region ${AWS_REGION} -X POST \\" |
| 116 | +echo " -d '{\"question\":\"What does the nava-platform CLI wrap to install templates?\"}' \\" |
| 117 | +echo " ${FUNCTION_URL}" |
0 commit comments