cfgen: overrid config for tw hosts

Summary: title says it Differential Revision: D58302088 fbshipit-source-id: 71a1e94335004fae24381598c7c378b83884c141
facebookincubator · Jun 7, 2024 · 57b302d · 57b302d
1 parent 8b50ed0
commit 57b302d
Show file tree

Hide file tree

Showing 9 changed files with 572 additions and 3 deletions.
diff --git a/src/oomd/cfgen/src/cfgen.rs b/src/oomd/cfgen/src/cfgen.rs
@@ -866,7 +866,7 @@ fn should_setup_iocost(node: &Node) -> bool {
 }
 
 fn fbtax2_blacklisted_jobs(node: &Node) -> Vec<&'static str> {
-    if get_host_type(node) == HostType::TwShared {
+    if [HostType::TwShared, HostType::Tw].contains(&get_host_type(node)) {
         return vec![
             // This ML model has extremely high memory usage, they need to fix
             // their stuff at some point.
@@ -897,7 +897,7 @@ fn senpai_targets(node: &Node) -> Option<String> {
     }
 
     match get_host_type(node) {
-        HostType::TwShared => Some(String::from(
+        HostType::TwShared | HostType::Tw => Some(String::from(
             "system.slice,workload.slice/workload-wdb.slice,hostcritical.slice,workload.slice/workload-wdb.slice/*,hostcritical.slice/*",
         )),
         HostType::Synmon => Some(String::from("system.slice")),
@@ -906,7 +906,7 @@ fn senpai_targets(node: &Node) -> Option<String> {
 }
 
 fn senpai_limit_min_bytes(node: &Node) -> Option<String> {
-    if [HostType::TwShared, HostType::Synmon].contains(&get_host_type(node)) {
+    if [HostType::TwShared, HostType::Tw, HostType::Synmon].contains(&get_host_type(node)) {
         let min_bytes = 100 * 1024 * 1024;
         return Some(min_bytes.to_string());
     }
@@ -935,6 +935,10 @@ fn get_host_type(node: &Node) -> HostType {
         return HostType::TwShared;
     }
 
+    if node.hostname_prefix() == TW {
+        return HostType::Tw;
+    }
+
     if node.hostname_prefix() == OD {
         return HostType::OnDemand;
     }

diff --git a/src/oomd/cfgen/src/types.rs b/src/oomd/cfgen/src/types.rs
@@ -11,6 +11,7 @@ pub enum HostType {
     DevServer,
     OnDemand,
     TwShared,
+    Tw,
     Synmon,
     Dns,
 }

diff --git a/src/oomd/cfgen/test/cfgen_test_inputs/twpool.json b/src/oomd/cfgen/test/cfgen_test_inputs/twpool.json
@@ -0,0 +1,84 @@
+@generated SignedSource<<f1ded817beb0796e1f26f7e8cf3b2712>>
+@codegen-command arc cfgen update-inputs fb-oomd
+{
+  "fqdn": "tw066.02.cln2.facebook.com",
+  "region": "ireland",
+  "clusterType": "SERVICE_GENERIC_NON_MEMCACHE",
+  "modelId": 333376,
+  "kernelRelease": "5.19.0-0_fbk12_11583_g0bef9520ca2b",
+  "serverType": "TYPE_I_WEB",
+  "experiments": [],
+  "cpuArchitecture": "skylakede",
+  "metalosRootfs": false,
+  "provisioningConfig": {
+    "ethtoolByInterface": {
+      "eth0": {
+        "maxChannelsCombined": 36
+      }
+    },
+    "cpuCoreCount": 18,
+    "parentModelId": 334303,
+    "recoveryEnvironment": false,
+    "deviceType": "SERVER",
+    "datacenter": "cln2",
+    "cluster": "02",
+    "memTotal": 66999275520,
+    "osVersion": {
+      "distribution_name": "CentOS Stream release",
+      "version": 9,
+      "is_in_ramdisk": false,
+      "is_metalos": false
+    },
+    "pciByAddress": {
+      "0000:02:00.0": {
+        "vendor_id": 5555,
+        "device_id": 4117,
+        "class_code": 131072,
+        "board_part_number": "MCX4431N-GCAN_FB"
+      }
+    },
+    "static_smc_tiers": [],
+    "machine": "x86_64"
+  },
+  "bootConfig": {
+    "ethtoolByInterface": {
+      "eth0": {
+        "driver": "mlx5_core",
+        "driver_version": "5.19.0-0_fbk12_11583_g0bef9520c",
+        "firmware_version": "14.32.1206 (FB_0000000006)",
+        "bus_info": "0000:02:00.0"
+      }
+    }
+  },
+  "runtimeConfig": {
+    "hasHighPrivCert": true,
+    "regionRoutableCluster": "cln1.02",
+    "block_devices": {
+      "block_devices": {
+        "nvme0n1": {
+          "size_bytes": 256060514304,
+          "is_rotational": false,
+          "model": "HFS256GD9TNG-62A0A",
+          "serial": "MJ91N499211501J6C",
+          "physical_block_size": 512,
+          "logical_block_size": 512,
+          "is_root": true
+        }
+      }
+    },
+    "dynamic_smc_tiers": [],
+    "cluster_state": "CLUSTER_IN_USE",
+    "installed_platforms": [
+      "platform010",
+      "platform010-compat"
+    ],
+    "device_nics_enum": [
+      "ETH0",
+      "SVC0"
+    ]
+  },
+  "reservationConfig": {
+    "active_machine_materialization_id": "",
+    "current_reservation_host_profile_id": "NEWLY_PROVISIONED_PROFILE"
+  }
+}
diff --git a/src/oomd/cfgen/test/cfgen_test_inputs/twpool_no_senpai.json b/src/oomd/cfgen/test/cfgen_test_inputs/twpool_no_senpai.json
@@ -0,0 +1,84 @@
+@generated SignedSource<<385d00650b8b80058480fedbea087963>>
+@codegen-command arc cfgen update-inputs fb-oomd
+{
+  "fqdn": "tw023.04.vll2.facebook.com",
+  "region": "newmexico",
+  "clusterType": "HADOOP",
+  "modelId": 331202,
+  "kernelRelease": "5.19.0-0_fbk12_11583_g0bef9520ca2b",
+  "serverType": "TYPE_VI_FEED",
+  "experiments": [],
+  "cpuArchitecture": "skylake",
+  "metalosRootfs": false,
+  "provisioningConfig": {
+    "ethtoolByInterface": {
+      "eth0": {
+        "maxChannelsCombined": 63
+      }
+    },
+    "cpuCoreCount": 40,
+    "parentModelId": 333803,
+    "recoveryEnvironment": false,
+    "deviceType": "SERVER",
+    "datacenter": "vll2",
+    "cluster": "04",
+    "memTotal": 270019551232,
+    "osVersion": {
+      "distribution_name": "CentOS Stream release",
+      "version": 9,
+      "is_in_ramdisk": false,
+      "is_metalos": false
+    },
+    "pciByAddress": {
+      "0000:5e:00.0": {
+        "vendor_id": 5555,
+        "device_id": 4117,
+        "class_code": 131072,
+        "board_part_number": "MCX4411D-ACAN_FB"
+      }
+    },
+    "static_smc_tiers": [],
+    "machine": "x86_64"
+  },
+  "bootConfig": {
+    "ethtoolByInterface": {
+      "eth0": {
+        "driver": "mlx5_core",
+        "driver_version": "5.19.0-0_fbk12_11583_g0bef9520c",
+        "firmware_version": "14.27.2606 (FB_0000000005)",
+        "bus_info": "0000:5e:00.0"
+      }
+    }
+  },
+  "runtimeConfig": {
+    "hasHighPrivCert": true,
+    "regionRoutableCluster": "vll1.02",
+    "block_devices": {
+      "block_devices": {
+        "sda": {
+          "size_bytes": 2000398934016,
+          "is_rotational": true,
+          "model": "HGST HUS722T2TALA604",
+          "serial": "WMC6N0P94N1J",
+          "physical_block_size": 512,
+          "logical_block_size": 512,
+          "is_root": true
+        }
+      }
+    },
+    "dynamic_smc_tiers": [],
+    "cluster_state": "CLUSTER_IN_USE",
+    "installed_platforms": [
+      "platform010",
+      "platform010-compat"
+    ],
+    "device_nics_enum": [
+      "ETH0",
+      "SVC0"
+    ]
+  },
+  "reservationConfig": {
+    "active_machine_materialization_id": "",
+    "current_reservation_host_profile_id": "NEWLY_PROVISIONED_PROFILE"
+  }
+}
diff --git a/src/oomd/cfgen/test/cfgen_test_manifest.yml b/src/oomd/cfgen/test/cfgen_test_manifest.yml
@@ -32,3 +32,8 @@ samples:
   devbig:
     # this host is a sample from the "wdb config monitor" scuba table
     production_host: devbig284.ash8.facebook.com
+  twpool:
+    # this host is a sample from the "wdb config monitor" scuba table
+    production_host: tw066.02.cln2.facebook.com
+  twpool_no_senpai:
+    production_host: tw023.04.vll2.facebook.com
diff --git a/src/oomd/cfgen/test/cfgen_test_outputs/twpool/50-change-propagator.conf b/src/oomd/cfgen/test/cfgen_test_outputs/twpool/50-change-propagator.conf
@@ -0,0 +1,6 @@
+@generated SignedSource<<3dd8c7637bb7afa680fc168e9c49060d>>
+@codegen-command arc cfgen update-outputs fb-oomd
+[Service]
+Environment=OOMD_ARGS='--interval 1 --config /etc/oomd2.json --drop-in-dir /run/oomd/dropin'
+
+[Unit]