@aztec/end-to-end 0.0.1-commit.bf2612ae → 0.0.1-commit.c2595eba

package/src/spartan/utils/nodes.ts

@@ -1,8 +1,9 @@
 import { createLogger } from '@aztec/aztec.js/log';
+import { createAztecNodeClient } from '@aztec/aztec.js/node';
 import type { RollupCheatCodes } from '@aztec/aztec/testing';
 import type { CheckpointNumber } from '@aztec/foundation/branded-types';
 import type { Logger } from '@aztec/foundation/log';
-import { makeBackoff, retry } from '@aztec/foundation/retry';
+import { makeBackoff, retry, retryUntil } from '@aztec/foundation/retry';
 import { sleep } from '@aztec/foundation/sleep';
 import {
   type AztecNodeAdmin,
@@ -15,7 +16,13 @@ import { promisify } from 'util';
 
 import type { TestConfig } from './config.js';
 import { execHelmCommand } from './helm.js';
-import { deleteResourceByLabel, getChartDir, startPortForward, waitForResourceByLabel } from './k8s.js';
+import {
+  deleteResourceByLabel,
+  getChartDir,
+  startPortForward,
+  waitForResourceByLabel,
+  waitForStatefulSetsReady,
+} from './k8s.js';
 
 const execAsync = promisify(exec);
 
@@ -42,6 +49,63 @@ export async function awaitCheckpointNumber(
   }
 }
 
+/**
+ * Waits until the proven block number increases.
+ *
+ * @param rpcUrl - URL of an Aztec RPC node to query
+ * @param log - Logger instance
+ * @param timeoutSeconds - Maximum time to wait
+ * @param pollIntervalSeconds - How often to check
+ */
+export async function waitForProvenToAdvance(
+  rpcUrl: string,
+  log: Logger,
+  timeoutSeconds: number = 300,
+  pollIntervalSeconds: number = 12, // slot duration
+): Promise<void> {
+  const node = createAztecNodeClient(rpcUrl);
+
+  log.info('Waiting for proven block to advance (indicating epoch proof just submitted)...');
+
+  // Get current proven block number
+  let initialProvenBlock: number;
+  try {
+    const tips = await node.getL2Tips();
+    initialProvenBlock = Number(tips.proven.block.number);
+    log.info(`Current proven block: ${initialProvenBlock}. Waiting for it to increase...`);
+  } catch (err) {
+    log.warn(`Error getting initial tips: ${err}. Will poll until successful.`);
+    initialProvenBlock = 0;
+  }
+
+  await retryUntil(
+    async () => {
+      try {
+        const tips = await node.getL2Tips();
+        const currentProvenBlock = Number(tips.proven.block.number);
+        const proposedBlock = Number(tips.proposed.number);
+
+        log.verbose(
+          `Chain state: proposed=${proposedBlock}, proven=${currentProvenBlock} (waiting for > ${initialProvenBlock})`,
+        );
+
+        if (currentProvenBlock > initialProvenBlock) {
+          log.info(`Proven block advanced from ${initialProvenBlock} to ${currentProvenBlock}.`);
+          return true;
+        }
+
+        return false;
+      } catch (err) {
+        log.verbose(`Error checking tips: ${err}`);
+        return false;
+      }
+    },
+    'proven block to advance',
+    timeoutSeconds,
+    pollIntervalSeconds,
+  );
+}
+
 export async function getSequencers(namespace: string) {
   const selectors = [
     'app.kubernetes.io/name=validator',
@@ -127,6 +191,66 @@ export async function withSequencersAdmin<T>(env: TestConfig, fn: (node: AztecNo
   return results;
 }
 
+async function getAztecImageForMigrations(namespace: string): Promise<string> {
+  const aztecDockerImage = process.env.AZTEC_DOCKER_IMAGE;
+  if (aztecDockerImage) {
+    return aztecDockerImage;
+  }
+
+  const { stdout } = await execAsync(
+    `kubectl get pods -l app.kubernetes.io/name=validator -n ${namespace} -o jsonpath='{.items[0].spec.containers[?(@.name=="aztec")].image}' | cat`,
+  );
+  const image = stdout.trim().replace(/^'|'$/g, '');
+  if (!image) {
+    throw new Error(`Could not detect aztec image from validator pod in namespace ${namespace}`);
+  }
+  return image;
+}
+
+async function getHaDbConnectionUrl(namespace: string): Promise<string> {
+  const secretName = `${namespace}-validator-ha-db-postgres`;
+  const { stdout } = await execAsync(`kubectl get secret ${secretName} -n ${namespace} -o json`);
+  const secret = JSON.parse(stdout);
+  const data = secret?.data ?? {};
+  const decode = (value?: string) => (value ? Buffer.from(value, 'base64').toString('utf8') : '');
+  const user = decode(data.POSTGRES_USER);
+  const password = decode(data.POSTGRES_PASSWORD);
+  const database = decode(data.POSTGRES_DB);
+  if (!user || !password || !database) {
+    throw new Error(`Missing HA DB credentials in secret ${secretName}`);
+  }
+  const host = `${namespace}-validator-ha-db-postgres.${namespace}.svc.cluster.local`;
+  return `postgresql://${encodeURIComponent(user)}:${encodeURIComponent(password)}@${host}:5432/${database}`;
+}
+
+export async function initHADb(namespace: string): Promise<void> {
+  const databaseUrl = await getHaDbConnectionUrl(namespace);
+  const image = await getAztecImageForMigrations(namespace);
+  const jobName = `${namespace}-validator-ha-db-migrate`;
+  await execAsync(`kubectl delete pod ${jobName} -n ${namespace} --ignore-not-found=true`).catch(() => undefined);
+
+  const migrateCmd = [
+    `kubectl run ${jobName} -n ${namespace}`,
+    '--rm -i',
+    '--restart=Never',
+    `--image=${image}`,
+    `--env=DATABASE_URL=${databaseUrl}`,
+    '--command -- node --no-warnings /usr/src/yarn-project/aztec/dest/bin/index.js migrate-ha-db up',
+  ].join(' ');
+  const migrateCmdForLog = migrateCmd.replace(/--env=DATABASE_URL=\S+/, '--env=DATABASE_URL=<redacted>');
+
+  await retry(
+    async () => {
+      logger.info(`command: ${migrateCmdForLog}`);
+      await execAsync(migrateCmd);
+    },
+    'run HA DB migrations',
+    makeBackoff([1, 2, 4, 8, 16]),
+    logger,
+    true,
+  );
+}
+
 /**
  * Enables or disables probabilistic transaction dropping on validators and waits for rollout.
  * Wired to env vars P2P_DROP_TX and P2P_DROP_TX_CHANCE via Helm values.
@@ -239,25 +363,46 @@ export async function enableValidatorDynamicBootNode(
  */
 export async function rollAztecPods(namespace: string, clearState: boolean = false) {
   // Pod components use 'validator', but StatefulSets and PVCs use 'sequencer-node' for validators
-  const podComponents = ['p2p-bootstrap', 'prover-node', 'prover-broker', 'prover-agent', 'sequencer-node', 'rpc'];
-  const pvcComponents = ['p2p-bootstrap', 'prover-node', 'prover-broker', 'sequencer-node', 'rpc'];
+  const podComponents = [
+    'p2p-bootstrap',
+    'prover-node',
+    'prover-broker',
+    'prover-agent',
+    'sequencer-node',
+    'rpc',
+    'validator-ha-db',
+  ];
+  const pvcComponents = ['p2p-bootstrap', 'prover-node', 'prover-broker', 'sequencer-node', 'rpc', 'validator-ha-db'];
   // StatefulSet components that need to be scaled down before PVC deletion
   // Note: validators use 'sequencer-node' as component label, not 'validator'
-  const statefulSetComponents = ['p2p-bootstrap', 'prover-node', 'prover-broker', 'sequencer-node', 'rpc'];
+  const statefulSetComponents = [
+    'p2p-bootstrap',
+    'prover-node',
+    'prover-broker',
+    'sequencer-node',
+    'rpc',
+    'validator-ha-db',
+  ];
 
   if (clearState) {
     // To delete PVCs, we must first scale down StatefulSets so pods release the volumes
     // Otherwise PVC deletion will hang waiting for pods to terminate
 
-    // First, save original replica counts
+    // Save original replica counts for all StatefulSets
     const originalReplicas: Map<string, number> = new Map();
     for (const component of statefulSetComponents) {
       try {
-        const getCmd = `kubectl get statefulset -l app.kubernetes.io/component=${component} -n ${namespace} -o jsonpath='{.items[0].spec.replicas}'`;
+        // Get all StatefulSets that match the component label
+        const getCmd = `kubectl get statefulset -l app.kubernetes.io/component=${component} -n ${namespace} -o json`;
         const { stdout } = await execAsync(getCmd);
-        const replicas = parseInt(stdout.replace(/'/g, '').trim(), 10);
-        if (!isNaN(replicas) && replicas > 0) {
-          originalReplicas.set(component, replicas);
+        const result = JSON.parse(stdout);
+        for (const sts of result.items || []) {
+          const name = sts.metadata.name;
+          const replicas = sts.spec.replicas ?? 1;
+          if (replicas > 0) {
+            originalReplicas.set(name, replicas);
+            logger.debug(`Saved replica count for StatefulSet ${name}: ${replicas}`);
+          }
         }
       } catch {
         // Component might not exist, continue
@@ -276,27 +421,81 @@ export async function rollAztecPods(namespace: string, clearState: boolean = fal
       }
     }
 
-    // Wait for pods to terminate
-    await sleep(15 * 1000);
+    // Wait for all pods to fully terminate before deleting PVCs.
+    // terminationGracePeriodSeconds default is 30s.
+    logger.info('Waiting for pods to fully terminate before deleting PVCs...');
+    for (const component of statefulSetComponents) {
+      try {
+        // Wait for all pods with this component label to be deleted
+        const waitCmd = `kubectl wait pods -l app.kubernetes.io/component=${component} --for=delete -n ${namespace} --timeout=2m`;
+        logger.info(`command: ${waitCmd}`);
+        await execAsync(waitCmd);
+      } catch (e) {
+        logger.verbose(`Wait for pod deletion ${component} skipped: ${e}`);
+      }
+    }
+    // Extra buffer to ensure PVC protection finalizers are cleared
+    await sleep(5 * 1000);
 
     // Now delete PVCs (they should no longer be in use)
     for (const component of pvcComponents) {
-      await deleteResourceByLabel({
-        resource: 'persistentvolumeclaims',
-        namespace: namespace,
-        label: `app.kubernetes.io/component=${component}`,
-      });
+      try {
+        await deleteResourceByLabel({
+          resource: 'persistentvolumeclaims',
+          namespace: namespace,
+          label: `app.kubernetes.io/component=${component}`,
+        });
+      } catch (e) {
+        logger.warn(`Failed to delete PVCs for ${component}: ${e}`);
+      }
     }
 
-    // Scale StatefulSets back up to original replica counts
-    for (const component of statefulSetComponents) {
-      const replicas = originalReplicas.get(component) ?? 1;
+    // Verify PVCs are deleted
+    for (const component of pvcComponents) {
       try {
-        const scaleCmd = `kubectl scale statefulset -l app.kubernetes.io/component=${component} -n ${namespace} --replicas=${replicas} --timeout=2m`;
+        const waitCmd = `kubectl wait pvc -l app.kubernetes.io/component=${component} --for=delete -n ${namespace} --timeout=2m`;
+        logger.info(`command: ${waitCmd}`);
+        await execAsync(waitCmd);
+      } catch (e) {
+        logger.verbose(`Wait for PVC deletion ${component} skipped: ${e}`);
+      }
+    }
+
+    const haDbStatefulSets = [...originalReplicas.entries()].filter(([name]) => name.includes('validator-ha-db'));
+    const otherStatefulSets = [...originalReplicas.entries()].filter(([name]) => !name.includes('validator-ha-db'));
+
+    // Bring up HA DB first so we can run migrations before validators start
+    for (const [stsName, replicas] of haDbStatefulSets) {
+      try {
+        const scaleCmd = `kubectl scale statefulset ${stsName} -n ${namespace} --replicas=${replicas} --timeout=2m`;
         logger.info(`command: ${scaleCmd}`);
         await execAsync(scaleCmd);
       } catch (e) {
-        logger.verbose(`Scale up ${component} skipped: ${e}`);
+        logger.verbose(`Scale up ${stsName} skipped: ${e}`);
+      }
+    }
+
+    if (haDbStatefulSets.length > 0) {
+      try {
+        await waitForStatefulSetsReady({
+          namespace,
+          label: 'app.kubernetes.io/component=validator-ha-db',
+          timeoutSeconds: 600,
+        });
+        await initHADb(namespace);
+      } catch (e) {
+        logger.warn(`HA DB migration step skipped or failed: ${e}`);
+      }
+    }
+
+    // Scale remaining StatefulSets back up to original replica counts (by name, not label)
+    for (const [stsName, replicas] of otherStatefulSets) {
+      try {
+        const scaleCmd = `kubectl scale statefulset ${stsName} -n ${namespace} --replicas=${replicas} --timeout=2m`;
+        logger.info(`command: ${scaleCmd}`);
+        await execAsync(scaleCmd);
+      } catch (e) {
+        logger.verbose(`Scale up ${stsName} skipped: ${e}`);
       }
     }
   } else {
@@ -312,8 +511,21 @@ export async function rollAztecPods(namespace: string, clearState: boolean = fal
 
   await sleep(10 * 1000);
 
-  // Wait for pods to come back
-  for (const component of podComponents) {
+  // Wait for StatefulSets to have all replicas ready.
+  for (const component of statefulSetComponents) {
+    try {
+      await waitForStatefulSetsReady({
+        namespace,
+        label: `app.kubernetes.io/component=${component}`,
+        timeoutSeconds: 600, // 10 minutes
+      });
+    } catch (e) {
+      logger.warn(`StatefulSet component ${component} may not be fully ready: ${e}`);
+    }
+  }
+
+  const nonStatefulSetComponents = podComponents.filter(c => !statefulSetComponents.includes(c));
+  for (const component of nonStatefulSetComponents) {
     await waitForResourceByLabel({
       resource: 'pods',
       namespace: namespace,