Update capacity planning doc for blocks storage and add capacity test…

development/capacity-test/.env

@@ -0,0 +1,4 @@
+CORTEX_VERSION=latest
+GRAFANA_VERSION=10.4.2
+PROMETHEUS_VERSION=v3.2.1
+SEAWEEDFS_VERSION=3.67

development/capacity-test/cortex-config.yaml

@@ -0,0 +1,68 @@
+# Cortex configuration for capacity testing in microservices mode.
+# Runs distributor, ingester, querier, and store-gateway as separate
+# containers using memberlist for ring discovery (no Consul needed).
+
+auth_enabled: true
+
+server:
+  grpc_server_max_recv_msg_size: 104857600
+  grpc_server_max_send_msg_size: 104857600
+  grpc_server_max_concurrent_streams: 1000
+
+distributor:
+  shard_by_all_labels: true
+  pool:
+    health_check_ingesters: true
+
+ingester_client:
+  grpc_client_config:
+    max_recv_msg_size: 104857600
+    max_send_msg_size: 104857600
+    grpc_compression: gzip
+
+ingester:
+  lifecycler:
+    min_ready_duration: 0s
+    final_sleep: 0s
+    num_tokens: 512
+    ring:
+      kvstore:
+        store: memberlist
+      replication_factor: 1
+
+blocks_storage:
+  backend: s3
+  s3:
+    endpoint: seaweedfs:8333
+    region: local
+    bucket_name: cortex-blocks
+    access_key_id: any
+    secret_access_key: any
+    insecure: true
+  tsdb:
+    dir: /data/tsdb
+  bucket_store:
+    sync_dir: /data/tsdb-sync
+    bucket_index:
+      enabled: true
+
+compactor:
+  data_dir: /tmp/cortex/compactor
+  sharding_ring:
+    kvstore:
+      store: memberlist
+
+store_gateway:
+  sharding_enabled: true
+  sharding_ring:
+    replication_factor: 1
+    wait_stability_min_duration: 0s
+    kvstore:
+      store: memberlist
+
+frontend_worker:
+  match_max_concurrent: true
+
+memberlist:
+  join_members:
+    - ingester:7946

development/capacity-test/docker-compose.yaml

@@ -0,0 +1,118 @@
+services:
+  distributor:
+    image: quay.io/cortexproject/cortex:${CORTEX_VERSION}
+    command:
+      - -config.file=/config/cortex-config.yaml
+      - -target=distributor
+      - -server.http-listen-port=8001
+      - -server.grpc-listen-port=9001
+    volumes:
+      - ./cortex-config.yaml:/config/cortex-config.yaml:ro
+    depends_on:
+      ingester:
+        condition: service_started
+    ports:
+      - "8001:8001"
+    healthcheck:
+      test: wget -qO- http://127.0.0.1:8001/ready
+      interval: 10s
+      timeout: 10s
+      retries: 3
+    restart: on-failure
+  ingester:
+    image: quay.io/cortexproject/cortex:${CORTEX_VERSION}
+    command:
+      - -config.file=/config/cortex-config.yaml
+      - -target=ingester
+      - -server.http-listen-port=8002
+      - -server.grpc-listen-port=9002
+    volumes:
+      - ./cortex-config.yaml:/config/cortex-config.yaml:ro
+    ports:
+      - "8002:8002"
+    healthcheck:
+      test: wget -qO- http://127.0.0.1:8002/ready
+      interval: 10s
+      timeout: 10s
+      retries: 3
+    restart: on-failure
+  querier:
+    image: quay.io/cortexproject/cortex:${CORTEX_VERSION}
+    command:
+      - -config.file=/config/cortex-config.yaml
+      - -target=querier
+      - -server.http-listen-port=8003
+      - -server.grpc-listen-port=9003
+    volumes:
+      - ./cortex-config.yaml:/config/cortex-config.yaml:ro
+    depends_on:
+      ingester:
+        condition: service_started
+    ports:
+      - "8003:8003"
+    healthcheck:
+      test: wget -qO- http://127.0.0.1:8003/ready
+      interval: 10s
+      timeout: 10s
+      retries: 3
+    restart: on-failure
+  store-gateway:
+    image: quay.io/cortexproject/cortex:${CORTEX_VERSION}
+    command:
+      - -config.file=/config/cortex-config.yaml
+      - -target=store-gateway
+      - -server.http-listen-port=8004
+      - -server.grpc-listen-port=9004
+    volumes:
+      - ./cortex-config.yaml:/config/cortex-config.yaml:ro
+    depends_on:
+      ingester:
+        condition: service_started
+    ports:
+      - "8004:8004"
+    healthcheck:
+      test: wget -qO- http://127.0.0.1:8004/ready
+      interval: 10s
+      timeout: 10s
+      retries: 3
+    restart: on-failure
+  grafana:
+    image: grafana/grafana:${GRAFANA_VERSION}
+    environment:
+      - GF_AUTH_ANONYMOUS_ENABLED=true
+      - GF_AUTH_ANONYMOUS_ORG_ROLE=Admin
+      - GF_USERS_DEFAULT_THEME=light
+      - GF_LOG_MODE=console
+      - GF_LOG_LEVEL=critical
+    volumes:
+      - ./grafana-datasource.yaml:/etc/grafana/provisioning/datasources/datasource.yaml:ro
+      - ../../docs/getting-started/grafana-dashboard.yaml:/etc/grafana/provisioning/dashboards/dashboards.yaml:ro
+      - ../../docs/getting-started/dashboards/:/var/lib/grafana/dashboards/:ro
+    ports:
+      - "3000:3000"
+  prometheus:
+    image: prom/prometheus:${PROMETHEUS_VERSION}
+    command:
+      - --config.file=/config/prometheus-config.yaml
+    volumes:
+      - ./prometheus-config.yaml:/config/prometheus-config.yaml:ro
+    ports:
+      - "9090:9090"
+  seaweedfs:
+    image: chrislusf/seaweedfs:${SEAWEEDFS_VERSION}
+    command:
+      - server
+      - -s3
+      - -s3.config=/workspace/seaweedfs-config.json
+    ports:
+      - "8333:8333"
+    post_start:
+      - command: /seaweedfs-init.sh
+    volumes:
+      - ../../docs/getting-started/seaweedfs-config.json:/workspace/seaweedfs-config.json:ro
+      - ../../docs/getting-started/seaweedfs-init.sh:/seaweedfs-init.sh:ro
+    healthcheck:
+      test: wget -qO- http://127.0.0.1:8333/status
+      interval: 10s
+      timeout: 10s
+      retries: 3

development/capacity-test/grafana-datasource.yaml

@@ -0,0 +1,32 @@
+# Grafana datasource configuration for capacity test (microservices mode).
+# Queries go through the querier service.
+apiVersion: 1
+
+datasources:
+  - name: Cortex (tenant-01)
+    type: prometheus
+    access: proxy
+    orgId: 1
+    url: http://querier:8003/api/prom
+    jsonData: &jsonData
+      cacheLevel: None
+      httpHeaderName1: X-Scope-OrgID
+      httpMethod: POST
+      prometheusType: Cortex
+      prometheusVersion: 1.14.0
+      timeInterval: 15s
+    secureJsonData:
+      httpHeaderValue1: tenant-01
+    version: 1
+    editable: true
+    isDefault: true
+  - name: Tenant 02
+    type: prometheus
+    access: proxy
+    orgId: 1
+    url: http://querier:8003/api/prom
+    jsonData: *jsonData
+    secureJsonData:
+      httpHeaderValue1: tenant-02
+    version: 1
+    editable: true

development/capacity-test/prometheus-config.yaml

@@ -0,0 +1,99 @@
+# Prometheus configuration for capacity testing.
+# Scrapes distributor and ingester separately so we can isolate
+# ingester memory/CPU metrics from other components.
+# Uses 25 tenants to increase series count and reduce the impact
+# of fixed overhead on per-series estimates.
+global:
+  scrape_interval: 15s
+  evaluation_interval: 15s
+
+scrape_configs:
+  - job_name: "prometheus"
+    static_configs:
+      - targets: ["localhost:9090"]
+  - job_name: "distributor"
+    static_configs:
+      - targets: ["distributor:8001"]
+  - job_name: "ingester"
+    static_configs:
+      - targets: ["ingester:8002"]
+  - job_name: "querier"
+    static_configs:
+      - targets: ["querier:8003"]
+
+remote_write:
+  - url: http://distributor:8001/api/v1/push
+    headers:
+      X-Scope-OrgID: "tenant-01"
+  - url: http://distributor:8001/api/v1/push
+    headers:
+      X-Scope-OrgID: "tenant-02"
+  - url: http://distributor:8001/api/v1/push
+    headers:
+      X-Scope-OrgID: "tenant-03"
+  - url: http://distributor:8001/api/v1/push
+    headers:
+      X-Scope-OrgID: "tenant-04"
+  - url: http://distributor:8001/api/v1/push
+    headers:
+      X-Scope-OrgID: "tenant-05"
+  - url: http://distributor:8001/api/v1/push
+    headers:
+      X-Scope-OrgID: "tenant-06"
+  - url: http://distributor:8001/api/v1/push
+    headers:
+      X-Scope-OrgID: "tenant-07"
+  - url: http://distributor:8001/api/v1/push
+    headers:
+      X-Scope-OrgID: "tenant-08"
+  - url: http://distributor:8001/api/v1/push
+    headers:
+      X-Scope-OrgID: "tenant-09"
+  - url: http://distributor:8001/api/v1/push
+    headers:
+      X-Scope-OrgID: "tenant-10"
+  - url: http://distributor:8001/api/v1/push
+    headers:
+      X-Scope-OrgID: "tenant-11"
+  - url: http://distributor:8001/api/v1/push
+    headers:
+      X-Scope-OrgID: "tenant-12"
+  - url: http://distributor:8001/api/v1/push
+    headers:
+      X-Scope-OrgID: "tenant-13"
+  - url: http://distributor:8001/api/v1/push
+    headers:
+      X-Scope-OrgID: "tenant-14"
+  - url: http://distributor:8001/api/v1/push
+    headers:
+      X-Scope-OrgID: "tenant-15"
+  - url: http://distributor:8001/api/v1/push
+    headers:
+      X-Scope-OrgID: "tenant-16"
+  - url: http://distributor:8001/api/v1/push
+    headers:
+      X-Scope-OrgID: "tenant-17"
+  - url: http://distributor:8001/api/v1/push
+    headers:
+      X-Scope-OrgID: "tenant-18"
+  - url: http://distributor:8001/api/v1/push
+    headers:
+      X-Scope-OrgID: "tenant-19"
+  - url: http://distributor:8001/api/v1/push
+    headers:
+      X-Scope-OrgID: "tenant-20"
+  - url: http://distributor:8001/api/v1/push
+    headers:
+      X-Scope-OrgID: "tenant-21"
+  - url: http://distributor:8001/api/v1/push
+    headers:
+      X-Scope-OrgID: "tenant-22"
+  - url: http://distributor:8001/api/v1/push
+    headers:
+      X-Scope-OrgID: "tenant-23"
+  - url: http://distributor:8001/api/v1/push
+    headers:
+      X-Scope-OrgID: "tenant-24"
+  - url: http://distributor:8001/api/v1/push
+    headers:
+      X-Scope-OrgID: "tenant-25"