Collect baseline job costs

gantry/__main__.py

@@ -28,6 +28,7 @@ async def apply_migrations(db: aiosqlite.Connection):
         # and not inadvertently added to the migrations folder
         ("001_initial.sql", 1),
         ("002_spec_index.sql", 2),
+        ("003_job_cost.sql", 3),
     ]
 
     # apply migrations that have not been applied
@@ -45,6 +46,8 @@ async def apply_migrations(db: aiosqlite.Connection):
 async def init_db(app: web.Application):
     db = await aiosqlite.connect(os.environ["DB_FILE"])
     await apply_migrations(db)
+    # ensure foreign key constraints are enabled
+    await db.execute("PRAGMA foreign_keys = ON")
     app["db"] = db
     yield
     await db.close()

gantry/clients/prometheus/job.py

@@ -1,5 +1,7 @@
 import json
 
+import aiosqlite
+
 from gantry.clients.prometheus import util
 from gantry.util.spec import spec_variants
 
@@ -152,3 +154,68 @@ async def get_usage(self, pod: str, start: float, end: float) -> dict:
             "mem_min": mem_usage["min"],
             "mem_stddev": mem_usage["stddev"],
         }
+
+    async def get_costs(
+        self,
+        db: aiosqlite.Connection,
+        start: float,
+        end: float,
+        node_id: int,
+    ) -> dict:
+        """
+        Calculates the costs associated with a job.
+
+        args:
+            db: a database connection
+            resources: job requests and limits
+            usage: job memory and cpu usage
+            start: job start time
+            end: job end time
+            node_id: the node that the job ran on
+
+        returns:
+            dict of: job_cost_instance (cost of the instance over the job's lifetime)
+        """
+        costs = {}
+        async with db.execute(
+            """
+                select capacity_type, instance_type, zone
+                from nodes where id = ?
+            """,
+            (node_id,),
+        ) as cursor:
+            node = await cursor.fetchone()
+
+        if not node:
+            # this is a temporary condition that will happen during the transition
+            # to collecting
+            raise util.IncompleteData(
+                f"node instance metadata is missing from db. node={node_id}"
+            )
+
+        capacity_type, instance_type, zone = node
+
+        # spot instance prices can change, so we avg the cost over the job's runtime
+        instance_costs = await self.client.query_range(
+            query={
+                "metric": "karpenter_cloudprovider_instance_type_offering_price_estimate",  # noqa: E501
+                "filters": {
+                    "capacity_type": capacity_type,
+                    "instance_type": instance_type,
+                    "zone": zone,
+                },
+            },
+            start=start,
+            end=end,
+        )
+
+        if not instance_costs:
+            raise util.IncompleteData(f"node cost is missing. node={node_id}")
+
+        instance_costs = [float(value) for _, value in instance_costs[0]["values"]]
+        # average hourly cost of the instance over the job's lifetime
+        instance_cost = sum(instance_costs) / len(instance_costs)
+        # compute cost: hourly instance cost * job runtime in hours
+        node_cost = instance_cost * ((end - start) / 3600)  # seconds to hours
+        costs["job_cost_instance"] = node_cost
+        return costs

gantry/clients/prometheus/node.py

@@ -53,4 +53,6 @@ async def get_labels(self, hostname: str, time: float) -> dict:
             "arch": labels["label_kubernetes_io_arch"],
             "os": labels["label_kubernetes_io_os"],
             "instance_type": labels["label_node_kubernetes_io_instance_type"],
+            "capacity_type": labels["label_karpenter_sh_capacity_type"],
+            "zone": labels["label_topology_kubernetes_io_zone"],
         }

gantry/routes/collection.py

@@ -73,6 +73,8 @@ async def fetch_job(
         )
         usage = await prometheus.job.get_usage(annotations["pod"], job.start, job.end)
         node_id = await fetch_node(db_conn, prometheus, node_hostname, job.midpoint)
+        costs = await prometheus.job.get_costs(db_conn, job.start, job.end, node_id)
+
     except aiohttp.ClientError as e:
         logger.error(f"Request failed: {e}")
         return
@@ -93,6 +95,7 @@ async def fetch_job(
             **annotations,
             **resources,
             **usage,
+            **costs,
         },
     )
 
@@ -139,5 +142,7 @@ async def fetch_node(
             "arch": node_labels["arch"],
             "os": node_labels["os"],
             "instance_type": node_labels["instance_type"],
+            "capacity_type": node_labels["capacity_type"],
+            "zone": node_labels["zone"],
         },
     )

gantry/tests/defs/collection.py

@@ -20,8 +20,8 @@
 
 # used to compare successful insertions
 # run SELECT * FROM table_name WHERE id = 1; from python sqlite api and grab fetchone() result
-INSERTED_JOB = (1, 'runner-hwwb-i3u-project-2-concurrent-1-s10tq41z', 1, 1706117046, 1706118420, 9892514, 'success', 'pr42264_bugfix/mathomp4/hdf5-appleclang15', 'gmsh', '4.8.4', '{"alglib": true, "cairo": false, "cgns": true, "compression": true, "eigen": false, "external": false, "fltk": true, "gmp": true, "hdf5": false, "ipo": false, "med": true, "metis": true, "mmg": true, "mpi": true, "netgen": true, "oce": true, "opencascade": false, "openmp": false, "petsc": false, "privateapi": false, "shared": true, "slepc": false, "tetgen": true, "voropp": true, "build_system": "cmake", "build_type": "Release", "generator": "make"}', 'gcc', '11.4.0', 'linux-ubuntu20.04-x86_64_v3', 'e4s', 16, 0.75, None, 1.899768349523097, 0.2971597591741076, 4.128116379389054, 0.2483743618267752, 1.7602635378120381, 2000000000.0, 48000000000.0, 143698407.6190476, 2785280.0, 594620416.0, 2785280.0, 252073065.82263485)
-INSERTED_NODE = (1, 'ec253b04-b1dc-f08b-acac-e23df83b3602', 'ip-192-168-86-107.ec2.internal', 24.0, 196608000000.0, 'amd64', 'linux', 'i3en.6xlarge')
+INSERTED_JOB = (1, 'runner-hwwb-i3u-project-2-concurrent-1-s10tq41z', 1, 1706117046, 1706118420, 9892514, 'success', 'pr42264_bugfix/mathomp4/hdf5-appleclang15', 'gmsh', '4.8.4', '{"alglib": true, "cairo": false, "cgns": true, "compression": true, "eigen": false, "external": false, "fltk": true, "gmp": true, "hdf5": false, "ipo": false, "med": true, "metis": true, "mmg": true, "mpi": true, "netgen": true, "oce": true, "opencascade": false, "openmp": false, "petsc": false, "privateapi": false, "shared": true, "slepc": false, "tetgen": true, "voropp": true, "build_system": "cmake", "build_type": "Release", "generator": "make"}', 'gcc', '11.4.0', 'linux-ubuntu20.04-x86_64_v3', 'e4s', 16, 0.75, None, 1.899768349523097, 0.2971597591741076, 4.128116379389054, 0.2483743618267752, 1.7602635378120381, 2000000000.0, 48000000000.0, 143698407.6190476, 2785280.0, 594620416.0, 2785280.0, 252073065.82263485, 0.19083333333333333)
+INSERTED_NODE = (1, 'ec253b04-b1dc-f08b-acac-e23df83b3602', 'ip-192-168-86-107.ec2.internal', 24.0, 196608000000.0, 'amd64', 'linux', 'i3en.6xlarge', 'us-east-1c', 'spot')
 
 # these were obtained by executing the respective queries to Prometheus and capturing the JSON output
 # or the raw output of PrometheusClient._query
@@ -32,6 +32,7 @@
 VALID_CPU_USAGE = {'status': 'success', 'data': {'resultType': 'matrix', 'result': [{'metric': {'container': 'build', 'cpu': 'total', 'endpoint': 'https-metrics', 'id': '/kubepods.slice/kubepods-burstable.slice/kubepods-burstable-podd7aa13e0_998c_4f21_b1d6_62781f4980b0.slice/cri-containerd-48a5e9e7d46655e73ba119fa16b65fa94ceed23c55157db8269b0b12f18f55d1.scope', 'image': 'ghcr.io/spack/ubuntu20.04-runner-amd64-gcc-11.4:2023.08.01', 'instance': '192.168.86.107:10250', 'job': 'kubelet', 'metrics_path': '/metrics/cadvisor', 'name': '48a5e9e7d46655e73ba119fa16b65fa94ceed23c55157db8269b0b12f18f55d1', 'namespace': 'pipeline', 'node': 'ip-192-168-86-107.ec2.internal', 'pod': 'runner-hwwb-i3u-project-2-concurrent-1-s10tq41z', 'service': 'kube-prometheus-stack-kubelet'}, 'values': [[1706117145, '0.2483743618267752'], [1706117146, '0.25650526138466395'], [1706117147, '0.26463616094255266'], [1706117148, '0.2727670605004414'], [1706117149, '0.28089796005833007'], [1706117150, '0.2890288596162188'], [1706117151, '0.2971597591741076'], [1706117357, '3.7319005481816236'], [1706117358, '3.7319005481816236'], [1706117359, '3.7319005481816236'], [1706117360, '3.7319005481816245'], [1706117361, '3.7319005481816245'], [1706118420, '4.128116379389054']]}]}}
 VALID_NODE_INFO = {'status': 'success', 'data': {'resultType': 'vector', 'result': [{'metric': {'__name__': 'kube_node_info', 'container': 'kube-state-metrics', 'container_runtime_version': 'containerd://1.7.2', 'endpoint': 'http', 'instance': '192.168.164.84:8080', 'internal_ip': '192.168.86.107', 'job': 'kube-state-metrics', 'kernel_version': '5.10.205-195.804.amzn2.x86_64', 'kubelet_version': 'v1.27.9-eks-5e0fdde', 'kubeproxy_version': 'v1.27.9-eks-5e0fdde', 'namespace': 'monitoring', 'node': 'ip-192-168-86-107.ec2.internal', 'os_image': 'Amazon Linux 2', 'pod': 'kube-prometheus-stack-kube-state-metrics-dbd66d8c7-6ftw8', 'provider_id': 'aws:///us-east-1c/i-0fe9d9c99fdb3631d', 'service': 'kube-prometheus-stack-kube-state-metrics', 'system_uuid': 'ec253b04-b1dc-f08b-acac-e23df83b3602'}, 'value': [1706117733, '1']}]}}
 VALID_NODE_LABELS = {'status': 'success', 'data': {'resultType': 'vector', 'result': [{'metric': {'__name__': 'kube_node_labels', 'container': 'kube-state-metrics', 'endpoint': 'http', 'instance': '192.168.164.84:8080', 'job': 'kube-state-metrics', 'label_beta_kubernetes_io_arch': 'amd64', 'label_beta_kubernetes_io_instance_type': 'i3en.6xlarge', 'label_beta_kubernetes_io_os': 'linux', 'label_failure_domain_beta_kubernetes_io_region': 'us-east-1', 'label_failure_domain_beta_kubernetes_io_zone': 'us-east-1c', 'label_k8s_io_cloud_provider_aws': 'ceb9f9cc8e47252a6f7fe7d6bded2655', 'label_karpenter_k8s_aws_instance_category': 'i', 'label_karpenter_k8s_aws_instance_cpu': '24', 'label_karpenter_k8s_aws_instance_encryption_in_transit_supported': 'true', 'label_karpenter_k8s_aws_instance_family': 'i3en', 'label_karpenter_k8s_aws_instance_generation': '3', 'label_karpenter_k8s_aws_instance_hypervisor': 'nitro', 'label_karpenter_k8s_aws_instance_local_nvme': '15000', 'label_karpenter_k8s_aws_instance_memory': '196608', 'label_karpenter_k8s_aws_instance_network_bandwidth': '25000', 'label_karpenter_k8s_aws_instance_pods': '234', 'label_karpenter_k8s_aws_instance_size': '6xlarge', 'label_karpenter_sh_capacity_type': 'spot', 'label_karpenter_sh_initialized': 'true', 'label_karpenter_sh_provisioner_name': 'glr-x86-64-v4', 'label_kubernetes_io_arch': 'amd64', 'label_kubernetes_io_hostname': 'ip-192-168-86-107.ec2.internal', 'label_kubernetes_io_os': 'linux', 'label_node_kubernetes_io_instance_type': 'i3en.6xlarge', 'label_spack_io_pipeline': 'true', 'label_spack_io_x86_64': 'v4', 'label_topology_ebs_csi_aws_com_zone': 'us-east-1c', 'label_topology_kubernetes_io_region': 'us-east-1', 'label_topology_kubernetes_io_zone': 'us-east-1c', 'namespace': 'monitoring', 'node': 'ip-192-168-86-107.ec2.internal', 'pod': 'kube-prometheus-stack-kube-state-metrics-dbd66d8c7-6ftw8', 'service': 'kube-prometheus-stack-kube-state-metrics'}, 'value': [1706117733, '1']}]}}
+VALID_NODE_COST = {'status': 'success', 'data': {'resultType': 'matrix', 'result': [{'metric': {'__name__': 'karpenter_cloudprovider_instance_type_offering_price_estimate', 'capacity_type': 'spot', 'container': 'controller', 'endpoint': 'http-metrics', 'instance': '192.168.240.113:8000', 'instance_type': 'i3en.6xlarge', 'job': 'karpenter', 'namespace': 'karpenter', 'pod': 'karpenter-8488f7f6dc-ml7q8', 'region': 'us-east-1', 'service': 'karpenter', 'zone': 'us-east-1c'}, 'values': [[1723838829, '0.5']]}]}}
 
 # modified version of VALID_MEMORY_USAGE to make the mean/stddev 0
 INVALID_MEMORY_USAGE = {'status': 'success', 'data': {'resultType': 'matrix', 'result': [{'metric': {'__name__': 'container_memory_working_set_bytes', 'container': 'build', 'endpoint': 'https-metrics', 'id': '/kubepods.slice/kubepods-burstable.slice/kubepods-burstable-podd7aa13e0_998c_4f21_b1d6_62781f4980b0.slice/cri-containerd-48a5e9e7d46655e73ba119fa16b65fa94ceed23c55157db8269b0b12f18f55d1.scope', 'image': 'ghcr.io/spack/ubuntu20.04-runner-amd64-gcc-11.4:2023.08.01', 'instance': '192.168.86.107:10250', 'job': 'kubelet', 'metrics_path': '/metrics/cadvisor', 'name': '48a5e9e7d46655e73ba119fa16b65fa94ceed23c55157db8269b0b12f18f55d1', 'namespace': 'pipeline', 'node': 'ip-192-168-86-107.ec2.internal', 'pod': 'runner-hwwb-i3u-project-2-concurrent-1-s10tq41z', 'service': 'kube-prometheus-stack-kubelet'}, 'values': [[1706117115, '0']]}]}}

gantry/tests/defs/db.py

@@ -2,4 +2,4 @@
 # fmt: off
 
 # valid input into insert_node
-NODE_INSERT_DICT = {"uuid": "ec253b04-b1dc-f08b-acac-e23df83b3602", "hostname": "ip-192-168-86-107.ec2.internal", "cores": 24.0, "mem": 196608000000.0, "arch": "amd64", "os": "linux", "instance_type": "i3en.6xlarge"}
+NODE_INSERT_DICT = {"uuid": "ec253b04-b1dc-f08b-acac-e23df83b3602", "hostname": "ip-192-168-86-107.ec2.internal", "cores": 24.0, "mem": 196608000000.0, "arch": "amd64", "os": "linux", "instance_type": "i3en.6xlarge", "zone": "us-east-1c", "capacity_type": "spot"}

gantry/tests/sql/insert_job.sql

@@ -1 +1 @@
-INSERT INTO jobs VALUES(1,'runner-hwwb-i3u-project-2-concurrent-1-s10tq41z',2,1706117046,1706118420,9892514,'success','pr42264_bugfix/mathomp4/hdf5-appleclang15','gmsh','4.8.4','{"alglib": true, "cairo": false, "cgns": true, "compression": true, "eigen": false, "external": false, "fltk": true, "gmp": true, "hdf5": false, "ipo": false, "med": true, "metis": true, "mmg": true, "mpi": true, "netgen": true, "oce": true, "opencascade": false, "openmp": false, "petsc": false, "privateapi": false, "shared": true, "slepc": false, "tetgen": true, "voropp": true, "build_system": "cmake", "build_type": "Release", "generator": "make"}','gcc','11.4.0','linux-ubuntu20.04-x86_64_v3','e4s',16,0.75,NULL,4.12532286694540495,3.15805864677520409,11.6038107294648877,0.248374361826775191,3.34888880339475214,2000000000.0,48000000000.0,1649868862.72588062,999763968.0,5679742976.0,2785280.0,1378705563.21018671);
+INSERT INTO jobs VALUES(1,'runner-hwwb-i3u-project-2-concurrent-1-s10tq41z',2,1706117046,1706118420,9892514,'success','pr42264_bugfix/mathomp4/hdf5-appleclang15','gmsh','4.8.4','{"alglib": true, "cairo": false, "cgns": true, "compression": true, "eigen": false, "external": false, "fltk": true, "gmp": true, "hdf5": false, "ipo": false, "med": true, "metis": true, "mmg": true, "mpi": true, "netgen": true, "oce": true, "opencascade": false, "openmp": false, "petsc": false, "privateapi": false, "shared": true, "slepc": false, "tetgen": true, "voropp": true, "build_system": "cmake", "build_type": "Release", "generator": "make"}','gcc','11.4.0','linux-ubuntu20.04-x86_64_v3','e4s',16,0.75,NULL,4.12532286694540495,3.15805864677520409,11.6038107294648877,0.248374361826775191,3.34888880339475214,2000000000.0,48000000000.0,1649868862.72588062,999763968.0,5679742976.0,2785280.0,1378705563.21018671,0.19083333333333333);

gantry/tests/sql/insert_node.sql

@@ -1,2 +1,2 @@
 --- primary key is set to 2 to set up the test that checks for race conditions
-INSERT INTO nodes VALUES(2,'ec253b04-b1dc-f08b-acac-e23df83b3602','ip-192-168-86-107.ec2.internal',24.0,196608000000.0,'amd64','linux','i3en.6xlarge');
+INSERT INTO nodes VALUES(2,'ec253b04-b1dc-f08b-acac-e23df83b3602','ip-192-168-86-107.ec2.internal',24.0,196608000000.0,'amd64','linux','i3en.6xlarge','us-east-1c','spot');