Runtime: add protocol information for service discovery (#1767)

python/cloudtik/core/_private/service_discovery/utils.py

@@ -7,7 +7,8 @@
 # The standard keys and values used for service discovery
 
 SERVICE_DISCOVERY_PROTOCOL = "protocol"
-SERVICE_DISCOVERY_PROTOCOL_TCP = "TCP"
+SERVICE_DISCOVERY_PROTOCOL_TCP = "tcp"
+SERVICE_DISCOVERY_PROTOCOL_HTTP = "http"
 
 SERVICE_DISCOVERY_PORT = "port"
 
@@ -83,9 +84,12 @@ def define_runtime_service(
         service_discovery_config: Optional[Dict[str, Any]],
         service_port,
         node_kind=SERVICE_DISCOVERY_NODE_KIND_NODE,
+        protocol: str = None,
         metrics: bool = False):
+    if not protocol:
+        protocol = SERVICE_DISCOVERY_PROTOCOL_TCP
     service_def = {
-        SERVICE_DISCOVERY_PROTOCOL: SERVICE_DISCOVERY_PROTOCOL_TCP,
+        SERVICE_DISCOVERY_PROTOCOL: protocol,
         SERVICE_DISCOVERY_PORT: service_port,
     }
 
@@ -107,35 +111,41 @@ def define_runtime_service(
 def define_runtime_service_on_worker(
         service_discovery_config: Optional[Dict[str, Any]],
         service_port,
+        protocol: str = None,
         metrics: bool = False):
     return define_runtime_service(
         service_discovery_config,
         service_port,
         node_kind=SERVICE_DISCOVERY_NODE_KIND_WORKER,
+        protocol=protocol,
         metrics=metrics)
 
 
 def define_runtime_service_on_head(
         service_discovery_config,
         service_port,
+        protocol: str = None,
         metrics: bool = False):
     return define_runtime_service(
         service_discovery_config,
         service_port,
         node_kind=SERVICE_DISCOVERY_NODE_KIND_HEAD,
+        protocol=protocol,
         metrics=metrics)
 
 
 def define_runtime_service_on_head_or_all(
         service_discovery_config,
         service_port, head_or_all,
+        protocol: str = None,
         metrics: bool = False):
     node_kind = SERVICE_DISCOVERY_NODE_KIND_NODE \
         if head_or_all else SERVICE_DISCOVERY_NODE_KIND_HEAD
     return define_runtime_service(
         service_discovery_config,
         service_port,
         node_kind=node_kind,
+        protocol=protocol,
         metrics=metrics)
 
 

python/cloudtik/runtime/ai/utils.py

@@ -4,7 +4,7 @@
 from cloudtik.core._private.providers import _get_node_provider, _get_workspace_provider
 from cloudtik.core._private.runtime_factory import BUILT_IN_RUNTIME_AI
 from cloudtik.core._private.service_discovery.utils import get_canonical_service_name, define_runtime_service_on_head, \
-    get_service_discovery_config
+    get_service_discovery_config, SERVICE_DISCOVERY_PROTOCOL_HTTP
 from cloudtik.core._private.utils import export_runtime_flags
 from cloudtik.runtime.common.utils import get_runtime_endpoints_of
 
@@ -93,6 +93,7 @@ def _get_runtime_services(
         service_discovery_config, cluster_name, MLFLOW_SERVICE_NAME)
     services = {
         service_name: define_runtime_service_on_head(
-            service_discovery_config, MLFLOW_SERVICE_PORT),
+            service_discovery_config, MLFLOW_SERVICE_PORT,
+            protocol=SERVICE_DISCOVERY_PROTOCOL_HTTP),
     }
     return services

python/cloudtik/runtime/consul/utils.py

@@ -43,6 +43,7 @@
 SERVICE_CHECK_TIMEOUT_DEFAULT = 5
 
 CONSUL_TAG_CLUSTER_NAME_FORMAT = "cloudtik-c-{}"
+CONSUL_TAG_PROTOCOL_FORMAT = "cloudtik-p-{}"
 CONSUL_TAG_METRICS = "cloudtik-m-metrics"
 
 
@@ -152,6 +153,8 @@ def _generate_service_config(cluster_name, runtime_type, runtime_service):
     if metrics:
         tags.append(CONSUL_TAG_METRICS)
 
+    # TODO: protocol as tag
+
     labels = get_config_for_update(
         service_config, SERVICE_DISCOVERY_LABELS)
 

python/cloudtik/runtime/grafana/utils.py

@@ -11,7 +11,7 @@
     get_services_of_runtime
 from cloudtik.core._private.service_discovery.utils import \
     get_canonical_service_name, define_runtime_service_on_head_or_all, get_service_discovery_config, \
-    SERVICE_DISCOVERY_PORT, serialize_service_selector
+    SERVICE_DISCOVERY_PORT, serialize_service_selector, SERVICE_DISCOVERY_PROTOCOL_HTTP
 from cloudtik.core._private.utils import RUNTIME_CONFIG_KEY
 
 RUNTIME_PROCESSES = [
@@ -161,6 +161,7 @@ def _get_runtime_services(
         service_name: define_runtime_service_on_head_or_all(
             service_discovery_config, service_port,
             _is_high_availability(grafana_config),
+            protocol=SERVICE_DISCOVERY_PROTOCOL_HTTP,
             metrics=True),
     }
     return services

python/cloudtik/runtime/nginx/conf/nginx-web-base.conf

@@ -1,5 +1,3 @@
-load_module /etc/nginx/modules/ngx_http_js_module.so;
-
 http {
     include       /etc/nginx/mime.types;
     default_type  application/octet-stream;

python/cloudtik/runtime/nginx/utils.py

@@ -9,7 +9,7 @@
 from cloudtik.core._private.service_discovery.runtime_services import get_service_discovery_runtime
 from cloudtik.core._private.service_discovery.utils import get_canonical_service_name, \
     get_service_discovery_config, define_runtime_service_on_head_or_all, exclude_runtime_of_cluster, \
-    serialize_service_selector
+    serialize_service_selector, SERVICE_DISCOVERY_PROTOCOL_HTTP
 from cloudtik.core._private.utils import RUNTIME_CONFIG_KEY
 from cloudtik.runtime.common.service_discovery.consul import get_service_dns_name, select_dns_service_tag
 
@@ -125,7 +125,9 @@ def _get_runtime_services(
     services = {
         service_name: define_runtime_service_on_head_or_all(
             service_discovery_config, service_port,
-            _is_high_availability(nginx_config))
+            _is_high_availability(nginx_config),
+            protocol=SERVICE_DISCOVERY_PROTOCOL_HTTP,
+        )
     }
     return services
 

python/cloudtik/runtime/node_exporter/utils.py

@@ -3,7 +3,7 @@
 
 from cloudtik.core._private.service_discovery.utils import \
     get_canonical_service_name, define_runtime_service, \
-    get_service_discovery_config
+    get_service_discovery_config, SERVICE_DISCOVERY_PROTOCOL_HTTP
 
 RUNTIME_PROCESSES = [
         # The first element is the substring to filter.
@@ -67,6 +67,7 @@ def _get_runtime_services(
     services = {
         service_name: define_runtime_service(
             service_discovery_config, service_port,
+            protocol=SERVICE_DISCOVERY_PROTOCOL_HTTP,
             metrics=True),
     }
     return services

python/cloudtik/runtime/prometheus/utils.py

@@ -12,7 +12,7 @@
     define_runtime_service_on_head_or_all, get_service_discovery_config, is_service_for_metrics, SERVICE_DISCOVERY_PORT, \
     SERVICE_SELECTOR_SERVICES, SERVICE_SELECTOR_TAGS, SERVICE_SELECTOR_LABELS, SERVICE_SELECTOR_EXCLUDE_LABELS, \
     SERVICE_SELECTOR_RUNTIMES, SERVICE_SELECTOR_CLUSTERS, SERVICE_DISCOVERY_LABEL_RUNTIME, \
-    SERVICE_DISCOVERY_LABEL_CLUSTER, SERVICE_SELECTOR_EXCLUDE_JOINED_LABELS
+    SERVICE_DISCOVERY_LABEL_CLUSTER, SERVICE_SELECTOR_EXCLUDE_JOINED_LABELS, SERVICE_DISCOVERY_PROTOCOL_HTTP
 from cloudtik.core._private.utils import RUNTIME_CONFIG_KEY
 
 RUNTIME_PROCESSES = [
@@ -224,6 +224,7 @@ def _get_runtime_services(
         service_name: define_runtime_service_on_head_or_all(
             service_discovery_config, service_port,
             _is_high_availability(prometheus_config),
+            protocol=SERVICE_DISCOVERY_PROTOCOL_HTTP,
             metrics=True)
     }
     return services