chore: prime-node / vector-node / readiness probe (#54)

demeter-run · Jul 8, 2024 · 5ac3578 · 5ac3578
1 parent 60e5b41
commit 5ac3578
Show file tree

Hide file tree

Showing 5 changed files with 215 additions and 5 deletions.
diff --git a/bootstrap/configs/prime-testnet/config.json b/bootstrap/configs/prime-testnet/config.json
@@ -0,0 +1,109 @@
+{
+  "ByronGenesisFile": "/genesis/byron/genesis.json",
+  "ShelleyGenesisFile": "/genesis/shelley/genesis.json",
+  "AlonzoGenesisFile": "/genesis/shelley/genesis.alonzo.json",
+  "ConwayGenesisFile": "/genesis/shelley/genesis.conway.json",
+  "ByronGenesisHash": "acd3e7c4cc071ae176fa253e654f0dd4e759892b16e958d8bd3c393f5c6c66d2",
+  "ShelleyGenesisHash": "0b67b0cb16e973478888efb1c7f69fca8e80489369afed94624a588e94a94f2b",
+  "AlonzoGenesisHash": "0f715ff78fcab739c6cb2cc1073eb87d0ffe3514218bfa0337d416f6fbaf7886",
+  "ConwayGenesisHash": "d43d3f38e01848bddf67e66fe29fbd236aca252de58c7f2c1904c67107c17842",
+  "SocketPath": "db/node.socket",
+  "PBftSignatureThreshold": 0.6,
+  "MaxConcurrencyBulkSync": 1,
+  "MaxConcurrencyDeadline": 2,
+  "Protocol": "Cardano",
+  "RequiresNetworkMagic": "RequiresMagic",
+  "LastKnownBlockVersion-Major": 6,
+  "LastKnownBlockVersion-Minor": 0,
+  "LastKnownBlockVersion-Alt": 0,
+  "TurnOnLogging": true,
+  "TurnOnLogMetrics": true,
+  "minSeverity": "Debug",
+  "TracingVerbosity": "NormalVerbosity",
+  "setupBackends": ["KatipBK"],
+  "defaultBackends": ["KatipBK"],
+  "hasEKG": 12788,
+  "hasPrometheus": [
+    "0.0.0.0",
+    12798
+  ],
+  "setupScribes": [
+    {
+      "scKind": "StdoutSK",
+      "scName": "stdout",
+      "scFormat": "ScText"
+    }
+  ],
+  "defaultScribes": [
+    ["StdoutSK", "stdout"]
+  ],
+  "rotation": {
+    "rpLogLimitBytes": 5000000,
+    "rpKeepFilesNum": 3,
+    "rpMaxAgeHours": 24
+  },
+  "TraceBlockFetchClient": false,
+  "TraceBlockFetchDecisions": false,
+  "TraceBlockFetchProtocol": false,
+  "TraceBlockFetchProtocolSerialised": false,
+  "TraceBlockFetchServer": false,
+  "TraceBlockchainTime": false,
+  "TraceChainDb": true,
+  "TraceChainSyncClient": false,
+  "TraceChainSyncBlockServer": false,
+  "TraceChainSyncHeaderServer": false,
+  "TraceChainSyncProtocol": false,
+  "TraceDNSResolver": true,
+  "TraceDNSSubscription": true,
+  "TraceErrorPolicy": true,
+  "TraceLocalErrorPolicy": true,
+  "TraceForge": true,
+  "TraceHandshake": false,
+  "TraceIpSubscription": true,
+  "TraceLocalRootPeers": true,
+  "TracePublicRootPeers": true,
+  "TracePeerSelection": true,
+  "TraceDebugPeerSelection": false,
+  "TracePeerSelectionActions": true,
+  "TraceConnectionManager": true,
+  "TraceServer": true,
+  "TraceLocalConnectionManager": false,
+  "TraceLocalServer": false,
+  "TraceLocalChainSyncProtocol": false,
+  "TraceLocalHandshake": false,
+  "TraceLocalTxSubmissionProtocol": false,
+  "TraceLocalTxSubmissionServer": false,
+  "TraceMempool": true,
+  "TraceMux": false,
+  "TraceTxInbound": false,
+  "TraceTxOutbound": false,
+  "TraceTxSubmissionProtocol": false,
+  "options": {
+    "mapBackends": {
+      "cardano.node.metrics": ["EKGViewBK"]
+    },
+    "mapScribes": {
+      "cardano.node.metrics": ["FileSK::logs/mainnet.log"]
+    },
+    "mapSeverity": {
+      "cardano.node.ChainDB": "Notice",
+      "cardano.node.DnsSubscription": "Debug"
+    }
+  },
+  "TestShelleyHardForkAtEpoch": 0,
+  "TestAllegraHardForkAtEpoch": 0,
+  "TestMaryHardForkAtEpoch": 0,
+  "TestAlonzoHardForkAtEpoch": 0,
+  "TestBabbageHardForkAtEpoch": 0,
+  "ExperimentalProtocolsEnabled": true,
+  "EnableP2P": true,
+  "TargetNumberOfRootPeers": 20,
+  "TargetNumberOfKnownPeers": 20,
+  "TargetNumberOfEstablishedPeers": 10,
+  "TargetNumberOfActivePeers": 10,
+  "TraceInboundGovernorCounters": true,
+  "TraceConnectionManagerTransitions": true,
+  "TraceLedgerPeers": true,
+  "TracePeerSelectionCounters": true,
+  "TracePeerStateActions": true
+}
diff --git a/bootstrap/configs/prime-testnet/topology.json b/bootstrap/configs/prime-testnet/topology.json
@@ -0,0 +1,14 @@
+{
+  "localRoots": [
+  ],
+  "publicRoots" : [
+    { "accessPoints": [
+        { "address": "relay-0.prime.testnet.apexfusion.org", "port": 5521 },
+        { "address": "relay-1.prime.testnet.apexfusion.org", "port": 5521 }
+      ],
+      "advertise": true,
+      "valency": 1
+    }
+  ],
+  "useLedgerAfterSlot": 0
+}
diff --git a/bootstrap/feature/config.tf b/bootstrap/feature/config.tf
@@ -0,0 +1,10 @@
+resource "kubernetes_config_map" "node-readiness" {
+  metadata {
+    namespace = var.namespace
+    name      = "node-readiness"
+  }
+
+  data = {
+    "readiness.sh"   = "${file("${path.module}/readiness.sh")}"
+  }
+}
diff --git a/bootstrap/feature/readiness.sh b/bootstrap/feature/readiness.sh
@@ -0,0 +1,33 @@
+#!/bin/bash
+
+if ! command -v jq &> /dev/null; then
+    echo "Error: jq is not installed. Please install jq to use this script."
+    exit 1
+fi
+
+if ! command -v cardano-cli &> /dev/null; then
+    echo "Error: cardano-cli is not installed. Please install cardano-cli to use this script."
+    exit 1
+fi
+
+if ! command -v bc &> /dev/null; then
+    echo "Error: bc is not installed. Please install bc to use this script."
+    exit 1
+fi
+
+if [ "$CARDANO_NODE_NETWORK_ID" == "764824073" ]; then
+    JSON_DATA=$(cardano-cli query tip --mainnet)
+else
+    JSON_DATA=$(cardano-cli query tip --testnet-magic "$CARDANO_NODE_NETWORK_ID")
+fi
+
+SYNC_PROGRESS=$(echo "$JSON_DATA" | jq -r '.syncProgress')
+MIN_EXPECTED_SYNC_PROGRESS="99.00"
+MAX_EXPECTED_SYNC_PROGRESS="100.00"
+
+if (( $(echo "$SYNC_PROGRESS >= $MIN_EXPECTED_SYNC_PROGRESS" | bc -l) )) && (( $(echo "$SYNC_PROGRESS <= $MAX_EXPECTED_SYNC_PROGRESS" | bc -l) )); then
+    echo "syncProgress is within the acceptable range of 99 to 100"
+else
+    echo "Error: syncProgress is not within the acceptable range of 99 to 100"
+    exit 1
+fi
diff --git a/bootstrap/instance/node.tf b/bootstrap/instance/node.tf
@@ -21,7 +21,7 @@ locals {
     "--port",
     "3000"
   ]
-  arguments = var.is_custom == true ? local.custom_arguments : local.default_arguments
+  arguments = var.network == "vector-testnet" ? [] : var.is_custom == true ? local.custom_arguments : local.default_arguments
 
   n2n_port_name = contains(["mainnet", "preview", "preprod"], var.network) && var.release == "stable" ? "n2n-${var.network}" : "n2n"
 }
@@ -157,6 +157,14 @@ resource "kubernetes_stateful_set_v1" "node" {
           }
         }
 
+        volume {
+          name = "node-readiness"
+          config_map {
+            name         = "node-readiness"
+            default_mode = "0500"
+          }
+        }
+
         container {
           image = "${var.node_image}:${var.node_image_tag}"
           name  = "main"
@@ -173,6 +181,33 @@ resource "kubernetes_stateful_set_v1" "node" {
             value = var.restore
           }
 
+          env {
+            name  = "CARDANO_NODE_SOCKET_PATH"
+            value = "/ipc/node.socket"
+          }
+
+          env {
+            name  = "CARDANO_NODE_NETWORK_ID"
+            value = var.magic
+          }
+
+          dynamic "env" {
+            for_each = var.network == "vector-testnet" ? toset([1]) : toset([])
+
+            content {
+              name  = "PORT"
+              value = "3000"
+            }
+          }
+
+          dynamic "env" {
+            for_each = var.network == "vector-testnet" ? toset([1]) : toset([])
+            content {
+              name  = "NETWORK"
+              value = "testnet"
+            }
+          }
+
           resources {
             limits   = var.node_resources.limits
             requests = var.node_resources.requests
@@ -198,6 +233,11 @@ resource "kubernetes_stateful_set_v1" "node" {
             name       = "ipc"
           }
 
+          volume_mount {
+            mount_path = "/probes"
+            name       = "node-readiness"
+          }
+
           dynamic "volume_mount" {
             for_each = var.is_custom == true ? toset([1]) : toset([])
 
@@ -207,10 +247,14 @@ resource "kubernetes_stateful_set_v1" "node" {
             }
           }
 
-          readiness_probe {
-            initial_delay_seconds = 20
-            exec {
-              command = ["test", "-S", "/ipc/node.socket"]
+          dynamic "readiness_probe" {
+            for_each = var.network != "vector-testnet" ? toset([1]) : toset([])
+
+            content {
+              initial_delay_seconds = 20
+              exec {
+                command = ["/probes/readiness.sh"]
+              }
             }
           }
         }