Merge remote-tracking branch 'upstream/main'

mrnicegyu11 · mrnicegyu11 · commit 77ee45e4112b · 2024-11-25T08:55:40.000Z
diff --git a/.github/PULL_REQUEST_TEMPLATE.md b/.github/PULL_REQUEST_TEMPLATE.md
@@ -17,7 +17,7 @@
 - [ ] Service has resource limits and reservations
 - [ ] Service has placement constraints or is global
 - [ ] Service is restartable
-  - [ ] The restart is zero-downtime
+- [ ] Service restart is zero-downtime
 - [ ] Service is not bound to one specific node (e.g. via files or volumes)
-- [ ] Relevant OPS E2E Test for this stack were added
--->
+- [ ] Relevant OPS E2E Test are added
+- [ ] Service's Public URL is included in maintenance mode -->
diff --git a/scripts/common.Makefile b/scripts/common.Makefile
@@ -245,10 +245,22 @@ clean-default: .check_clean ## Cleans all outputs
 	@.venv/bin/pip3 install --upgrade pip wheel setuptools
 	@.venv/bin/pip3 install jinja2 j2cli[yaml]
 
+
+# https://github.com/kolypto/j2cli?tab=readme-ov-file#customization
+ifeq ($(shell test -f j2cli_customization.py && echo -n yes),yes)
+
+define jinja
+	.venv/bin/j2 --format=env $(1) .env -o $(2) --customize j2cli_customization.py
+endef
+
+else
+
 define jinja
 	.venv/bin/j2 --format=env $(1) .env -o $(2)
 endef
 
+endif
+
 # Gracefully use defaults and potentially overwrite them, via https://stackoverflow.com/a/49804748
 %:  %-default
 	@ true
diff --git a/services/jaeger/docker-compose.yml b/services/jaeger/docker-compose.yml
@@ -51,6 +51,7 @@ services:
     networks:
       - public
       - monitored
+      - interactive_services_subnet
     environment:
       TRACING_OPENTELEMETRY_COLLECTOR_BATCH_SIZE: ${TRACING_OPENTELEMETRY_COLLECTOR_BATCH_SIZE}
       TRACING_OPENTELEMETRY_COLLECTOR_SAMPLING_PERCENTAGE: ${TRACING_OPENTELEMETRY_COLLECTOR_SAMPLING_PERCENTAGE}
@@ -63,3 +64,6 @@ networks:
   monitored:
     name: ${MONITORED_NETWORK}
     external: true
+  interactive_services_subnet:
+    name: ${SWARM_STACK_NAME}_interactive_services_subnet
+    external: true
diff --git a/services/jaeger/opentelemetry-collector-config.yaml b/services/jaeger/opentelemetry-collector-config.yaml
@@ -13,7 +13,7 @@ service:
     traces:
       receivers: [otlp]
       exporters: [otlphttp]
-      processors: [batch,probabilistic_sampler]
+      processors: [batch,probabilistic_sampler,filter/drop_healthcheck]
   telemetry:
     logs:
       level: ${TRACING_OPENTELEMETRY_COLLECTOR_SERVICE_TELEMETRY_LOG_LEVEL}
@@ -23,3 +23,11 @@ processors:
     send_batch_size: ${TRACING_OPENTELEMETRY_COLLECTOR_BATCH_SIZE}
   probabilistic_sampler:
     sampling_percentage: ${TRACING_OPENTELEMETRY_COLLECTOR_SAMPLING_PERCENTAGE}
+  filter/drop_healthcheck:
+    error_mode: ignore
+    traces:
+      span:
+        - attributes["http.route"] == "healthcheck_readiness_probe"
+        - attributes["http.route"] == "healthcheck_liveness_probe"
+        - attributes["http.target"] == "/metrics" and IsMatch(attributes["http.user_agent"], ".*Prometheus.*") == true
+        - attributes["db.statement"] == "PING" and attributes["db.system"] == "redis"
diff --git a/services/jaeger/template.env b/services/jaeger/template.env
@@ -5,3 +5,4 @@ TRACING_OPENTELEMETRY_COLLECTOR_BATCH_SIZE=${TRACING_OPENTELEMETRY_COLLECTOR_BAT
 TRACING_OPENTELEMETRY_COLLECTOR_SAMPLING_PERCENTAGE=${TRACING_OPENTELEMETRY_COLLECTOR_SAMPLING_PERCENTAGE}
 TRACING_OPENTELEMETRY_COLLECTOR_EXPORTER_ENDPOINT=${TRACING_OPENTELEMETRY_COLLECTOR_EXPORTER_ENDPOINT}
 TRACING_OPENTELEMETRY_COLLECTOR_SERVICE_TELEMETRY_LOG_LEVEL=${TRACING_OPENTELEMETRY_COLLECTOR_SERVICE_TELEMETRY_LOG_LEVEL}
+SWARM_STACK_NAME=${SWARM_STACK_NAME}
diff --git a/services/maintenance-page/docker-compose.yml.j2 b/services/maintenance-page/docker-compose.yml.j2
@@ -28,8 +28,7 @@ services:
         - traefik.enable=true
         - traefik.docker.network=${PUBLIC_NETWORK}
         - traefik.http.routers.{{"maintenance_" + j2item.replace('@','').replace(' ','').replace('.','').replace('-','').replace('\'','') + "_html"}}.priority={{MAINTENANCE_PAGES_TRAEFIK_PRIORITY}}
-        - traefik.http.routers.{{"maintenance_" + j2item.replace('@','').replace(' ','').replace('.','').replace('-','').replace('\'','') + "_html"}}.rule={{  "Host(`" + j2item + "`)" }}
-        - traefik.http.routers.{{"maintenance_" + j2item.replace('@','').replace(' ','').replace('.','').replace('-','').replace('\'','') + "_html"}}.rule=(Host(`{{j2item}}`) && PathPrefix(`/`)) || (HostRegexp(`services.{{j2item}}`,`{subhost:[a-zA-Z0-9-]+}.services.{{j2item}}`) && PathPrefix(`/`))
+        - traefik.http.routers.{{"maintenance_" + j2item.replace('@','').replace(' ','').replace('.','').replace('-','').replace('\'','') + "_html"}}.rule=Host(`{{VENDOR_MANUAL_SUBDOMAIN_PREFIX}}.{{j2item}}`) || (Host(`{{j2item}}`) && PathPrefix(`/`)) || (HostRegexp(`services.{{j2item}}`,`{subhost:[a-zA-Z0-9-]+}.services.{{j2item}}`) && PathPrefix(`/`))
         - traefik.http.routers.{{"maintenance_" + j2item.replace('@','').replace(' ','').replace('.','').replace('-','').replace('\'','') + "_html"}}.tls=true
         - traefik.http.services.{{"maintenance_" + j2item.replace('@','').replace(' ','').replace('.','').replace('-','').replace('\'','') + "_html"}}.loadbalancer.server.port=80
         - traefik.http.routers.{{"maintenance_" + j2item.replace('@','').replace(' ','').replace('.','').replace('-','').replace('\'','') + "_html"}}.entrypoints=https
diff --git a/services/maintenance-page/template.env b/services/maintenance-page/template.env
@@ -8,3 +8,4 @@ PUBLIC_NETWORK=${PUBLIC_NETWORK}
 MONITORED_NETWORK=${MONITORED_NETWORK}
 REPO_CONFIG_LOCATION=${REPO_CONFIG_LOCATION}
 MAINTENANCE_PAGES_TRAEFIK_PRIORITY=${MAINTENANCE_PAGES_TRAEFIK_PRIORITY}
+VENDOR_MANUAL_SUBDOMAIN_PREFIX=${VENDOR_MANUAL_SUBDOMAIN_PREFIX}
diff --git a/services/registry/template.env b/services/registry/template.env
@@ -13,7 +13,7 @@ S3_SECRET_KEY=${REGISTRY_S3_SECRET_KEY}
 S3_ENDPOINT=${S3_ENDPOINT}
 S3_BUCKET=${REGISTRY_DOMAIN}
 REGISTRY_S3_BUCKET=${REGISTRY_S3_BUCKET}
-REGISTRY_S3_ENDPOINT=${S3_ENDPOINT}
+REGISTRY_S3_ENDPOINT=${REGISTRY_S3_ENDPOINT}
 REGISTRY_S3_SECURE=${REGISTRY_S3_SECURE}
 REGISTRY_AWS_REGION=${REGISTRY_AWS_REGION}
 
diff --git a/services/simcore/docker-compose.deploy.aws.yml b/services/simcore/docker-compose.deploy.aws.yml
@@ -17,9 +17,7 @@ services:
   clusters-keeper:
     deploy:
       replicas: 1
-  resource-usage-tracker:
-    deploy:
-      replicas: 3
+
   static-webserver:
     hostname: "{{.Node.Hostname}}-{{.Service.Name}}"
 
diff --git a/services/simcore/docker-compose.deploy.dalco.yml b/services/simcore/docker-compose.deploy.dalco.yml
@@ -17,9 +17,7 @@ services:
       placement:
         constraints:
           - node.role == worker
-  resource-usage-tracker:
-    deploy:
-      replicas: 3
+
   traefik:
     command:
       - "--api=true"
diff --git a/services/simcore/docker-compose.deploy.public.yml b/services/simcore/docker-compose.deploy.public.yml
@@ -10,9 +10,6 @@ services:
         constraints:
           - node.labels.postgres==true
 
-  resource-usage-tracker:
-    deploy:
-      replicas: 3
   static-webserver:
     hostname: "{{.Node.Hostname}}-{{.Service.Name}}"
   traefik:
diff --git a/services/simcore/docker-compose.yml b/services/simcore/docker-compose.yml
@@ -649,7 +649,7 @@ services:
       # NOTE: https://github.com/ITISFoundation/osparc-simcore/pull/4286
       # NOTE: this MUSTN'T change, or weird things might happen
       # this will stay until all legacy dynamic services are gone.
-      replicas: 1
+      replicas: ${RESOURCE_USAGE_TRACKER_REPLICAS}
       update_config:
         parallelism: 2
         order: start-first
diff --git a/services/vendors/.gitignore b/services/vendors/.gitignore
@@ -0,0 +1 @@
+docker-compose.yml
diff --git a/services/vendors/Makefile b/services/vendors/Makefile
@@ -30,6 +30,12 @@ up-local: up
 
 # Helpers -------------------------------------------------
 
+.PHONY: docker-compose.yml
+docker-compose.yml: .venv .env
+	@$(call jinja, docker-compose.yml.j2, docker-compose.yml.unlinted) && \
+	$(_yq) docker-compose.yml.unlinted > docker-compose.yml; \
+	rm docker-compose.yml.unlinted >/dev/null 2>&1;
+
 .PHONY: ${TEMP_COMPOSE}
 ${TEMP_COMPOSE}: docker-compose.yml .env
 	${REPO_BASE_DIR}/scripts/docker-stack-config.bash -e .env $< > $@
diff --git a/services/vendors/docker-compose.yml.j2 b/services/vendors/docker-compose.yml.j2
@@ -4,7 +4,9 @@ services:
   manual:
     image: ${VENDOR_MANUAL_IMAGE}
     init: true
+{%- raw %}
     hostname: "{{.Node.Hostname}}-{{.Task.Slot}}"
+{%- endraw %}
     deploy:
       replicas: ${VENDOR_MANUAL_REPLICAS}
       placement:
@@ -28,7 +30,7 @@ services:
         - traefik.http.services.vendor_manual.loadbalancer.server.port=${VENDOR_MANUAL_PORT}
         - traefik.http.routers.vendor_manual.entrypoints=https
         - traefik.http.routers.vendor_manual.tls=true
-        - traefik.http.routers.vendor_manual.rule=Host(`${VENDOR_MANUAL_DOMAIN}`)
+        - traefik.http.routers.vendor_manual.rule={{ generate_vendors_manual_traefik_rule(VENDOR_MANUAL_PRODUCTS, VENDOR_MANUAL_SUBDOMAIN_PREFIX) }}
         - traefik.http.routers.vendor_manual.middlewares=ops_gzip@swarm, authenticated_platform_user@swarm
     networks:
       - public
diff --git a/services/vendors/j2cli_customization.py b/services/vendors/j2cli_customization.py
@@ -0,0 +1,11 @@
+def _generate_vendors_manual_traefik_rule(domains: str, subdomain_prefix: str) -> str:
+    domain_list = domains.strip().strip(",").split(",")
+    domains = [f"{subdomain_prefix}.{domain}" for domain in domain_list]
+    return " || ".join(f"Host(`{d}`)" for d in domains)
+
+
+def j2_environment(env):
+    env.globals.update(
+        generate_vendors_manual_traefik_rule=_generate_vendors_manual_traefik_rule
+    )
+    return env
diff --git a/services/vendors/template.env b/services/vendors/template.env
@@ -1,5 +1,6 @@
 VENDOR_MANUAL_IMAGE=${VENDOR_MANUAL_IMAGE}
 VENDOR_MANUAL_REPLICAS=${VENDOR_MANUAL_REPLICAS}
-VENDOR_MANUAL_DOMAIN=${VENDOR_MANUAL_DOMAIN}
+VENDOR_MANUAL_SUBDOMAIN_PREFIX=${VENDOR_MANUAL_SUBDOMAIN_PREFIX}
+VENDOR_MANUAL_PRODUCTS=${VENDOR_MANUAL_PRODUCTS}
 VENDOR_MANUAL_PORT=${VENDOR_MANUAL_PORT}
 PUBLIC_NETWORK=${PUBLIC_NETWORK}
