♻️🗃️ Is1004/modify resource tracker backend after container label cha…

…nges (#4488)

packages/postgres-database/src/simcore_postgres_database/migration/versions/ef931143b7cd_refactoring_of_resource_tracker.py

@@ -0,0 +1,96 @@
+"""refactoring of resource_tracker_container table
+
+Revision ID: ef931143b7cd
+Revises: a8762d5d43ae
+Create Date: 2023-07-11 14:37:57.455348+00:00
+
+"""
+import sqlalchemy as sa
+from alembic import op
+from sqlalchemy.dialects import postgresql
+
+# revision identifiers, used by Alembic.
+revision = "ef931143b7cd"
+down_revision = "a8762d5d43ae"
+branch_labels = None
+depends_on = None
+
+
+def upgrade():
+    container_classification_enum = postgresql.ENUM(
+        "DYNAMIC_SIDECAR", "USER_SERVICE", name="containerclassification"
+    )
+    container_classification_enum.create(op.get_bind())
+
+    op.execute("DELETE FROM resource_tracker_container;")
+
+    # ### commands auto generated by Alembic - please adjust! ###
+    op.add_column(
+        "resource_tracker_container",
+        sa.Column("cpu_limit", sa.Numeric(precision=3, scale=2), nullable=False),
+    )
+    op.add_column(
+        "resource_tracker_container",
+        sa.Column("memory_limit", sa.BigInteger(), nullable=False),
+    )
+    op.add_column(
+        "resource_tracker_container",
+        sa.Column(
+            "classification",
+            sa.Enum("DYNAMIC_SIDECAR", "USER_SERVICE", name="containerclassification"),
+            nullable=True,
+        ),
+    )
+    op.drop_column(
+        "resource_tracker_container", "service_settings_reservation_nano_cpus"
+    )
+    op.drop_column("resource_tracker_container", "service_settings_limit_nano_cpus")
+    op.drop_column("resource_tracker_container", "service_settings_limit_memory_bytes")
+    op.drop_column(
+        "resource_tracker_container", "service_settings_reservation_memory_bytes"
+    )
+    # ### end Alembic commands ###
+
+
+def downgrade():
+    # ### commands auto generated by Alembic - please adjust! ###
+    op.add_column(
+        "resource_tracker_container",
+        sa.Column(
+            "service_settings_reservation_memory_bytes",
+            sa.BIGINT(),
+            autoincrement=False,
+            nullable=True,
+        ),
+    )
+    op.add_column(
+        "resource_tracker_container",
+        sa.Column(
+            "service_settings_limit_memory_bytes",
+            sa.BIGINT(),
+            autoincrement=False,
+            nullable=True,
+        ),
+    )
+    op.add_column(
+        "resource_tracker_container",
+        sa.Column(
+            "service_settings_limit_nano_cpus",
+            sa.BIGINT(),
+            autoincrement=False,
+            nullable=True,
+        ),
+    )
+    op.add_column(
+        "resource_tracker_container",
+        sa.Column(
+            "service_settings_reservation_nano_cpus",
+            sa.BIGINT(),
+            autoincrement=False,
+            nullable=True,
+        ),
+    )
+    op.drop_column("resource_tracker_container", "classification")
+    op.drop_column("resource_tracker_container", "memory_limit")
+    op.drop_column("resource_tracker_container", "cpu_limit")
+    # ### end Alembic commands ###

packages/postgres-database/src/simcore_postgres_database/models/resource_tracker.py

@@ -3,13 +3,20 @@
     - Table where we store the resource usage of each container that
     we scrape via resource-usage-tracker service
 """
+import enum
 
 import sqlalchemy as sa
 from sqlalchemy.dialects.postgresql import JSONB
 
 from ._common import column_modified_datetime
 from .base import metadata
 
+
+class ContainerClassification(str, enum.Enum):
+    DYNAMIC_SIDECAR = enum.auto()
+    USER_SERVICE = enum.auto()
+
+
 resource_tracker_container = sa.Table(
     "resource_tracker_container",
     metadata,
@@ -39,18 +46,6 @@
         doc="product_name label scraped via Prometheus (taken from container labels)",
         index=True,
     ),
-    sa.Column(
-        "service_settings_reservation_nano_cpus",
-        sa.BigInteger,
-        nullable=True,
-        doc="CPU resource allocated to a container, ex.500000000 means that the container is allocated 0.5 CPU shares",
-    ),
-    sa.Column(
-        "service_settings_reservation_memory_bytes",
-        sa.BigInteger,
-        nullable=True,
-        doc="memory limit in bytes scraped via Prometheus",
-    ),
     sa.Column(
         "service_settings_reservation_additional_info",
         JSONB,
@@ -90,18 +85,6 @@
         nullable=True,
         doc="instance label scraped via Prometheus (taken from container labels, ex.: gpu1)",
     ),
-    sa.Column(
-        "service_settings_limit_nano_cpus",
-        sa.BigInteger,
-        nullable=True,
-        doc="CPU resource limit allocated to a container, ex.500000000 means that the container has limit for 0.5 CPU shares",
-    ),
-    sa.Column(
-        "service_settings_limit_memory_bytes",
-        sa.BigInteger,
-        nullable=True,
-        doc="memory limit in bytes scraped via Prometheus",
-    ),
     sa.Column(
         "project_name",
         sa.String,
@@ -126,6 +109,23 @@
         nullable=False,
         doc="Service Version (parsed from image label scraped via Prometheus)",
     ),
+    sa.Column(
+        "cpu_limit",
+        sa.Numeric(precision=3, scale=2),
+        nullable=False,
+        doc="CPU resource allocated to a container, ex.0.5 CPU shares",
+    ),
+    sa.Column(
+        "memory_limit",
+        sa.BigInteger,
+        nullable=False,
+        doc="memory limit in bytes scraped via Prometheus",
+    ),
+    sa.Column(
+        "classification",
+        sa.Enum(ContainerClassification),
+        doc="Our custom classification of the container type",
+    ),
     # ---------------------------
     sa.PrimaryKeyConstraint("container_id", name="resource_tracker_container_pkey"),
 )

services/resource-usage-tracker/src/simcore_service_resource_usage_tracker/models/resource_tracker_container.py

@@ -8,7 +8,8 @@
 from models_library.projects_nodes_io import NodeID
 from models_library.services import ServiceKey, ServiceVersion
 from models_library.users import UserID
-from pydantic import BaseModel, Field, PositiveInt
+from pydantic import BaseModel, ByteSize, Field, PositiveInt
+from simcore_postgres_database.models.resource_tracker import ContainerClassification
 
 # Scraped from prometheus
 
@@ -26,17 +27,18 @@ class ContainerScrapedResourceUsageMetric(BaseModel):
         None,
         description="Instance label scraped via Prometheus (taken from container labels, ex.: gpu1)",
     )
-    service_settings_reservation_nano_cpus: int | None = Field(
-        None,
-        description="CPU resource limit allocated to a container, ex.500000000 means that the container has limit for 0.5 CPU shares",
-    )
-    service_settings_reservation_memory_bytes: int | None
     service_settings_reservation_additional_info: dict[str, Any] = Field(
         {},
         description="Storing additional information about the reservation settings, such as what type of graphic card is used.",
     )
-    service_settings_limit_nano_cpus: int | None
-    service_settings_limit_memory_bytes: int | None
+    memory_limit: ByteSize = Field(
+        None,
+        description="Memory bytes limit set by the runtime, ex. 17179869184 means that the container has limit for 16GB of memory",
+    )
+    cpu_limit: float = Field(
+        None,
+        description="CPU limit set by the runtime, ex. 3.5 Shares of one CPU cores",
+    )
     service_key: ServiceKey
     service_version: ServiceVersion
 
@@ -50,8 +52,14 @@ class Config:
         arbitrary_types_allowed = True
 
 
+class ContainerScrapedResourceUsageCustom(BaseModel):
+    classification: ContainerClassification
+
+
 class ContainerScrapedResourceUsage(
-    ContainerScrapedResourceUsageMetric, ContainerScrapedResourceUsageValues
+    ContainerScrapedResourceUsageMetric,
+    ContainerScrapedResourceUsageValues,
+    ContainerScrapedResourceUsageCustom,
 ):
     ...
 
@@ -60,8 +68,8 @@ class ContainerScrapedResourceUsage(
 
 
 class ContainerGetDB(BaseModel):
-    service_settings_reservation_nano_cpus: int | None
-    service_settings_reservation_memory_bytes: int | None
+    cpu_limit: float
+    memory_limit: int
     prometheus_created: datetime
     prometheus_last_scraped: datetime
     project_uuid: ProjectID

services/resource-usage-tracker/src/simcore_service_resource_usage_tracker/modules/db/repositories/resource_tracker.py

@@ -39,8 +39,8 @@ async def upsert_resource_tracker_container_data(
                 project_uuid=f"{data.project_uuid}",
                 project_name=data.project_name,
                 product_name=data.product_name,
-                service_settings_reservation_nano_cpus=data.service_settings_reservation_nano_cpus,
-                service_settings_reservation_memory_bytes=data.service_settings_reservation_memory_bytes,
+                cpu_limit=data.cpu_limit,
+                memory_limit=data.memory_limit,
                 service_settings_reservation_additional_info=data.service_settings_reservation_additional_info,
                 container_cpu_usage_seconds_total=data.container_cpu_usage_seconds_total,
                 prometheus_created=data.prometheus_created.datetime,
@@ -49,10 +49,9 @@ async def upsert_resource_tracker_container_data(
                 node_uuid=f"{data.node_uuid}",
                 node_label=data.node_label,
                 instance=data.instance,
-                service_settings_limit_nano_cpus=data.service_settings_limit_nano_cpus,
-                service_settings_limit_memory_bytes=data.service_settings_limit_memory_bytes,
                 service_key=data.service_key,
                 service_version=data.service_version,
+                classification=data.classification,
             )
 
             on_update_stmt = insert_stmt.on_conflict_do_update(
@@ -84,8 +83,8 @@ async def list_containers_by_user_and_product(
         async with self.db_engine.begin() as conn:
             query = (
                 sa.select(
-                    resource_tracker_container.c.service_settings_reservation_nano_cpus,
-                    resource_tracker_container.c.service_settings_reservation_memory_bytes,
+                    resource_tracker_container.c.cpu_limit,
+                    resource_tracker_container.c.memory_limit,
                     resource_tracker_container.c.prometheus_created,
                     resource_tracker_container.c.prometheus_last_scraped,
                     resource_tracker_container.c.project_uuid,
@@ -106,7 +105,7 @@ async def list_containers_by_user_and_product(
 
             result = await conn.execute(query)
             containers_list = [
-                ContainerGetDB.construct(**row)  # type: ignore[arg-type]
+                ContainerGetDB(**row)  # type: ignore[arg-type]
                 for row in result.fetchall()
             ]