@rocicorp/zero 1.11.0-canary.36 → 1.11.0-canary.38

package/out/zero-cache/src/server/change-streamer.js.map

@@ -1 +1 @@
-{"version":3,"file":"change-streamer.js","names":[],"sources":["../../../../../zero-cache/src/server/change-streamer.ts"],"sourcesContent":["import {consoleLogSink, LogContext} from '@rocicorp/logger';\nimport {resolver} from '@rocicorp/resolver';\nimport {assert} from '../../../shared/src/asserts.ts';\nimport {must} from '../../../shared/src/must.ts';\nimport {promiseVoid} from '../../../shared/src/resolved-promises.ts';\nimport {DatabaseInitError} from '../../../zqlite/src/db.ts';\nimport {getServerContext} from '../config/server-context.ts';\nimport {getNormalizedZeroConfig} from '../config/zero-config.ts';\nimport {deleteLiteDB} from '../db/delete-lite-db.ts';\nimport {registerSQLiteCorruptionDiagnosticTarget} from '../db/sqlite-corruption.ts';\nimport {warmupConnections} from '../db/warmup.ts';\nimport {initEventSink, publishCriticalEvent} from '../observability/events.ts';\nimport {getOrCreateGauge} from '../observability/metrics.ts';\nimport {InitCleanup} from '../services/change-source/common/init-cleanup.ts';\nimport type {PgChangeLogPurgeLocker} from '../services/change-source/common/replica-restore.ts';\nimport {initializeCustomChangeSource} from '../services/change-source/custom/change-source.ts';\nimport {initializePostgresChangeSource} from '../services/change-source/pg/change-source-init.ts';\nimport {createBackupCleanupMonitor} from '../services/change-streamer/backup-cleanup-monitor-factory.ts';\nimport {ChangeStreamerHttpServer} from '../services/change-streamer/change-streamer-http.ts';\nimport {initializeStreamer} from '../services/change-streamer/change-streamer-service.ts';\nimport type {ChangeStreamerService} from '../services/change-streamer/change-streamer.ts';\nimport {initChangeStreamerSchema} from '../services/change-streamer/schema/init.ts';\nimport {AutoResetSignal} from '../services/change-streamer/schema/tables.ts';\nimport {\n  PurgeLocker,\n  type PurgeLock,\n} from '../services/change-streamer/storer.ts';\nimport {\n  exitAfter,\n  ProcessManager,\n  runUntilKilled,\n} from '../services/life-cycle.ts';\nimport {deleteLitestreamMetaDir} from '../services/litestream/commands.ts';\nimport {\n  changeLogFileName,\n  deleteChangeLogDB,\n} from '../services/replicator/change-log-db.ts';\nimport {\n  replicationStatusError,\n  ReplicationStatusPublisher,\n} from '../services/replicator/replication-status.ts';\nimport {sqliteFileBytes} from '../services/replicator/sqlite-change-log-observability.ts';\nimport {connectPgClient} from '../types/pg.ts';\nimport {\n  broadcastWorker,\n  childWorker,\n  parentWorker,\n  singleProcessMode,\n  type ProfileResponseMessage,\n  type Worker,\n} from '../types/processes.ts';\nimport {installProfileHandler} from '../types/profiler.ts';\nimport {getShardConfig} from '../types/shards.ts';\nimport type {ReplicaFileMode} from '../workers/replicator.ts';\nimport {createLogContext} from './logging.ts';\nimport {startOtelAuto} from './otel-start.ts';\nimport {REPLICATOR_URL} from './worker-urls.ts';\n\n// Default LogContext, overridden in runWorker\nlet lc = new LogContext('info', {}, consoleLogSink);\n\nexport default async function runWorker(\n  parent: Worker,\n  env: NodeJS.ProcessEnv,\n  ...argv: string[]\n): Promise<void> {\n  installProfileHandler(parent, 'change-streamer');\n  const config = getNormalizedZeroConfig({env, argv});\n  const {\n    taskID,\n    changeStreamer: {\n      port,\n      address,\n      protocol,\n      startupDelayMs,\n      backPressureLimitHeapProportion,\n      flowControlConsensusTimeoutProportion,\n      flowControlSlowSubscriberGracePeriodSeconds,\n      pgChangeLogEnabled,\n      sqliteChangeLogMode,\n      sqliteChangeLogReadPercent,\n      sqliteChangeLogColdReadPercent,\n      sqliteChangeLogComparePercent,\n      sqliteChangeLogRetentionMs,\n      sqliteChangeLogReadBatchRows,\n      sqliteChangeLogPurgeBatchRows,\n      sqliteChangeLogBarrierTimeoutMs,\n    },\n    autoReset,\n    replicationLag,\n    litestream,\n    upstream,\n    change,\n    replica,\n    initialSync,\n    keepaliveTimeoutMs,\n    sqliteCorruptionChecks,\n  } = config;\n\n  startOtelAuto(\n    createLogContext(config, 'change-streamer', 0, false),\n    'change-streamer',\n    0,\n  );\n  lc = createLogContext(config, 'change-streamer');\n  registerSQLiteCorruptionDiagnosticTarget(\n    {\n      debugName: 'change-streamer replica',\n      dbPath: replica.file,\n    },\n    sqliteCorruptionChecks,\n  );\n  initEventSink(lc, config);\n\n  // Startup-time client used for for change-streamer initialization\n  // and handoff / takeover. Steady-state clients are managed by the\n  // change-streamer (Storer) implementation.\n  const changeDB = await connectPgClient(\n    lc,\n    change.db,\n    'change-streamer-init',\n    {max: 5},\n    {sendStringAsJson: true},\n  );\n  void warmupConnections(lc, changeDB, 'change').catch(() => {});\n\n  const shard = getShardConfig(config);\n\n  // Ensure the change DB schema is initialized/up-to-date.\n  await initChangeStreamerSchema(lc, changeDB, shard);\n\n  let purgeLock = null as PurgeLock | null;\n  let changeStreamer: ChangeStreamerService | undefined;\n  let backupURL: string | undefined;\n\n  const context = getServerContext(config);\n  const sqliteChangeLogEnabled = sqliteChangeLogMode !== 'off';\n  // The modes are cumulative, so `serve` implies `compare`.\n  const sqliteChangeLogComparing =\n    sqliteChangeLogMode === 'compare' || sqliteChangeLogMode === 'serve';\n  if (sqliteChangeLogEnabled) {\n    const changeLogFile = changeLogFileName(replica.file);\n    registerSQLiteCorruptionDiagnosticTarget(\n      {\n        debugName: 'change-streamer change-log',\n        dbPath: changeLogFile,\n      },\n      sqliteCorruptionChecks,\n    );\n    // The log's bytes are accounted for in the process that writes them. This\n    // is local disk only — the log is excluded from the litestream backup — and\n    // it is a whole-database total, because a table that left the replica left\n    // through its wal: the replica's footprint is what shrank and this is where\n    // it went.\n    getOrCreateGauge('replica', 'sqlite_change_log.file_bytes', {\n      description:\n        `The SQLite change log's total on-disk footprint: its main db file ` +\n        `plus its wal sidecars.`,\n      unit: 'By',\n    }).addCallback(async o =>\n      o.observe(await sqliteFileBytes(lc, changeLogFile)),\n    );\n  }\n\n  let waitForFirstBackupBeforeServing = false;\n  let newBackupLineage = false;\n  for (const first of [true, false]) {\n    // Resources acquired by an initialization attempt (e.g. a claimed\n    // replication slot, the purge lock) to be released if the attempt fails.\n    const cleanup = new InitCleanup(lc);\n    // When restoring from litestream, a lock is acquired to prevent change-log\n    // purges. This ensures that (this) change-streamer will be able to resume\n    // from the backup. The lock is acquired by the change source initialization\n    // just before restoring (i.e. after a replication slot is created, if\n    // applicable), and released once this change-streamer takes over the\n    // change-log.\n    const acquirePurgeLock: PgChangeLogPurgeLocker | undefined =\n      pgChangeLogEnabled && litestream.backupURL && litestream.executable\n        ? async (slotWatermark?: string) => {\n            const lock = await new PurgeLocker(lc, shard, changeDB).acquire(\n              slotWatermark,\n            );\n            purgeLock = lock === 'behind-slot' ? null : lock;\n            if (purgeLock) {\n              const acquired = purgeLock;\n              cleanup.onFailure('purge lock', () => {\n                purgeLock = null;\n                return acquired.release();\n              });\n            }\n            return lock;\n          }\n        : undefined;\n\n    const restoreOptions = {litestream, acquirePurgeLock, cleanup};\n    try {\n      // Note: This performs initial sync of the replica if necessary.\n      const {\n        pgReplicationEpoch: epoch,\n        pgReplicationSlotPerReplica: slotPerReplica,\n        pgResumeOrphanedSlotGracePeriodMs: inactiveReplicaGracePeriodMs,\n      } = upstream;\n      const {\n        changeSource,\n        subscriptionState,\n        destinationBackupURL,\n        replicaID,\n        waitForBackupBeforeServing,\n        newBackupLineage: initNewBackupLineage,\n        pgChangeLogBehindSlot,\n      } = upstream.type === 'pg'\n        ? await initializePostgresChangeSource(\n            lc,\n            upstream.db,\n            shard,\n            replica.file,\n            {\n              ...initialSync,\n              replicationSlotFailover: upstream.pgReplicationSlotFailover,\n              installPartialIndexTriggers: upstream.pgPartialIndexTriggers,\n            },\n            context,\n            replicationLag.reportIntervalMs,\n            restoreOptions,\n            {\n              epoch,\n              slotPerReplica,\n              inactiveReplicaGracePeriodMs,\n              backupV5: litestream.backupUsingV5,\n              pgChangeLog: pgChangeLogEnabled,\n            },\n            upstream.pgStreamInboundTimeoutMs,\n          )\n        : await initializeCustomChangeSource(\n            lc,\n            upstream.db,\n            shard,\n            replica.file,\n            context,\n            restoreOptions,\n          );\n\n      const replicationStatusPublisher =\n        ReplicationStatusPublisher.forReplicaFile(replica.file);\n\n      changeStreamer = await initializeStreamer(\n        lc,\n        shard,\n        taskID,\n        address,\n        protocol,\n        changeDB,\n        changeSource,\n        replicationStatusPublisher,\n        subscriptionState,\n        destinationBackupURL\n          ? {\n              backupURL: destinationBackupURL,\n              litestreamVersion: litestream.backupUsingV5 ? 'v5' : 'legacy',\n              replicaFile: replica.file,\n            }\n          : null,\n        purgeLock,\n        autoReset ?? false,\n        {\n          pgChangeLogEnabled,\n          pgChangeLogBehindSlot: pgChangeLogBehindSlot ?? false,\n          backPressureLimitHeapProportion,\n          flowControlConsensusTimeoutProportion,\n          flowControlSlowSubscriberGracePeriodMs:\n            flowControlSlowSubscriberGracePeriodSeconds > 0\n              ? flowControlSlowSubscriberGracePeriodSeconds * 1000\n              : undefined,\n          statementTimeoutMs: change.statementTimeoutMs,\n          changeLogBatchSize: change.logBatchSize,\n          sqliteCatchup: {\n            changeLogFile: changeLogFileName(replica.file),\n            readBatchRows: sqliteChangeLogReadBatchRows,\n            barrierTimeoutMs: sqliteChangeLogBarrierTimeoutMs,\n          },\n          // The presence of these options is the writer's gate. This process\n          // performs the restore and the initial sync, so the replica's\n          // identity is in hand at the moment the log is opened.\n          sqliteChangeLogWriter: sqliteChangeLogEnabled\n            ? {\n                replicaFile: replica.file,\n                identity: {\n                  // RMv2 supplies the epoch; until then the generation and the\n                  // replica ID are the whole of the identity.\n                  epoch: null,\n                  generation: subscriptionState.replicaVersion,\n                  replicaID,\n                },\n              }\n            : undefined,\n          // The purge scheduler shares the writer's gate -- mode, not the\n          // read path -- because `write` mode, with no read selector at all,\n          // is the configuration it ships in.\n          sqliteChangeLogPurge: sqliteChangeLogEnabled\n            ? {\n                retentionMs: sqliteChangeLogRetentionMs,\n                batchRows: sqliteChangeLogPurgeBatchRows,\n              }\n            : undefined,\n          // Compare mode runs both advisory checks. Postgres remains authoritative.\n          sqliteChangeLogCompare:\n            pgChangeLogEnabled && sqliteChangeLogComparing\n              ? {\n                  replicaFile: replica.file,\n                  comparePercent: sqliteChangeLogComparePercent,\n                  retentionMs: sqliteChangeLogRetentionMs,\n                  readBatchRows: sqliteChangeLogReadBatchRows,\n                }\n              : undefined,\n          // Slice 11 lands dark by default: serve mode constructs the stable\n          // router, while readPercent=0 keeps every catchup on PG and emits\n          // eligibility metrics before any canary traffic is enabled.\n          sqliteChangeLogServe:\n            sqliteChangeLogMode === 'serve'\n              ? {\n                  readPercent: sqliteChangeLogReadPercent,\n                  coldReadPercent: sqliteChangeLogColdReadPercent,\n                  retentionMs: sqliteChangeLogRetentionMs,\n                }\n              : undefined,\n        },\n        setTimeout,\n      );\n      backupURL = destinationBackupURL;\n      waitForFirstBackupBeforeServing = waitForBackupBeforeServing;\n      newBackupLineage = initNewBackupLineage;\n      break;\n    } catch (e) {\n      // Release the resources acquired by the failed attempt, e.g.:\n      // * The claimed replication slot, which would otherwise remain active\n      //   indefinitely, preventing a retry from resuming (or forking) it.\n      // * The purge lock. This is safe because the purge lock exists to\n      //   preserve change-log entries so the new change-streamer can resume\n      //   from the backup replica's watermark. An AutoResetSignal means we\n      //   can't resume from the backup replica (e.g. its replication slot is\n      //   gone), so the change-log entries the lock was protecting are no\n      //   longer needed. The retry performs a fresh initial sync with a new\n      //   replication slot, independent of the old change-log. Releasing is\n      //   also necessary to avoid a self-deadlock when CHANGE_DB ==\n      //   UPSTREAM_DB: CREATE_REPLICATION_SLOT waits for all older\n      //   transactions to finish, including this lock's open transaction.\n      await cleanup.release();\n      if (first && e instanceof AutoResetSignal) {\n        lc.warn?.(`resetting replica ${replica.file}`, e);\n        // TODO: Make deleteLiteDB work with litestream. It will probably have to be\n        //       a semantic wipe instead of a file delete.\n        deleteLiteDB(replica.file);\n        // The change log carries the identity of the replica it was written\n        // beside, and the retry performs a fresh initial sync with a new\n        // replicaVersion. Reconciliation would catch that on its own (as\n        // 'identity-mismatch') but only after the writer opened a file that is\n        // known here to be garbage.\n        deleteChangeLogDB(replica.file);\n        continue; // execute again with a fresh initial-sync\n      }\n      if (e instanceof DatabaseInitError) {\n        throw new Error(\n          `Cannot open ZERO_REPLICA_FILE at \"${replica.file}\". Please check that the path is valid.`,\n          {cause: e},\n        );\n      }\n      throw e;\n    }\n  }\n  // impossible: upstream must have advanced in order for replication to be stuck.\n  assert(changeStreamer, `resetting replica did not advance replicaVersion`);\n\n  const processes = new ProcessManager(lc, parent);\n  const profileSubWorkers: Worker[] = [];\n  const {promise: replicatorReady, resolve} = resolver();\n  if (!backupURL) {\n    resolve(); // No backup replicator to wait for, so resolve immediately.\n  } else {\n    lc.info?.('setting up backup to', backupURL);\n    config.litestream.backupURLOverride = backupURL;\n    if (newBackupLineage) {\n      deleteLitestreamMetaDir(replica.file);\n    }\n    // Start a backup replicator, which starts up the corresponding\n    // litestream backup process.\n    const backupReplicator = processes\n      .addWorker(\n        childWorker(\n          REPLICATOR_URL,\n          {...env, ZERO_LITESTREAM_BACKUP_URL_OVERRIDE: backupURL},\n          'backup' satisfies ReplicaFileMode,\n        ),\n        'supporting',\n        'backup-replicator',\n      )\n      .onceMessageType('ready', () => resolve());\n    profileSubWorkers.push(backupReplicator);\n    // Relay profileResponse messages from backup-replicator up to parent\n    backupReplicator.onMessageType<ProfileResponseMessage>(\n      'profileResponse',\n      res => parent.send(['profileResponse', res]),\n    );\n  }\n\n  const backupMonitor = createBackupCleanupMonitor({\n    lc,\n    config,\n    replicaFile: replica.file,\n    changeStreamer,\n  });\n\n  let backupReady = promiseVoid;\n  if (waitForFirstBackupBeforeServing) {\n    const start = performance.now();\n    lc.info?.(`awaiting initial backup ...`);\n\n    backupReady = backupMonitor.firstBackupReceived().then(() => {\n      const elapsed = performance.now() - start;\n      lc.info?.(`initial backup confirmed after ${elapsed.toFixed(2)}ms`);\n    });\n  }\n\n  // In RMv2, readiness additionally waits for the backup-replicator to catch\n  // up to the replication stream. This is not done in RMv1 (i.e. when the\n  // PG change-log is enabled), in which the change-streamer itself is only\n  // started after the readinessGate (plus a startup delay), and thus the\n  // backup-replicator cannot catch up before then.\n  const readinessGate = pgChangeLogEnabled\n    ? backupReady\n    : Promise.all([backupReady, replicatorReady]);\n\n  // Create the broadcast facade once: each broadcastWorker() adds permanent\n  // 'message' forwarders to every sub worker, so creating one per /profz\n  // request would leak a forwarder per request.\n  const profileWorker =\n    profileSubWorkers.length > 0\n      ? broadcastWorker(profileSubWorkers)\n      : undefined;\n  const getProfileWorker = profileWorker\n    ? () => Promise.resolve(profileWorker)\n    : undefined;\n\n  const changeStreamerWebServer = new ChangeStreamerHttpServer(\n    lc,\n    {\n      port,\n      keepaliveTimeoutMs,\n      // The startup delay is only relevant when taking over the PG change-log\n      // (RMv1 and RMv1.5), and is disabled for RMv2.\n      startupDelayMs: pgChangeLogEnabled ? startupDelayMs : 0,\n      readinessGate,\n      config,\n      getProfileWorker,\n    },\n    parent,\n    changeStreamer,\n  );\n\n  void readinessGate.then(() => parent.send(['ready', {ready: true}]));\n\n  // Note: The changeStreamer itself is not started here; it is started by the\n  //       changeStreamerWebServer after a delay to ensure that routing\n  //       routing elements have registered the server before the\n  //       change-streamer takes over the replication slot.\n  // TODO: Remove this delay and start the changeStreamer normally here once\n  //       transitioned to RMv2, since changeStreamer startup will no longer\n  //       disrupt an existing task.\n  try {\n    await runUntilKilled(lc, parent, changeStreamerWebServer, backupMonitor);\n  } catch (err) {\n    processes.logErrorAndExit(err, 'change-streamer');\n  } finally {\n    await processes.shutdown();\n  }\n}\n\n// fork()\nif (!singleProcessMode()) {\n  void exitAfter(\n    () => lc,\n    () =>\n      runWorker(\n        must(parentWorker),\n        process.env,\n        ...process.argv.slice(2),\n      ).catch(async e => {\n        await publishCriticalEvent(\n          lc,\n          replicationStatusError(lc, 'Initializing', e),\n        );\n        throw e;\n      }),\n  );\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AA2DA,IAAI,KAAK,IAAI,WAAW,QAAQ,CAAC,GAAG,cAAc;AAElD,eAA8B,UAC5B,QACA,KACA,GAAG,MACY;CACf,sBAAsB,QAAQ,iBAAiB;CAC/C,MAAM,SAAS,wBAAwB;EAAC;EAAK;CAAI,CAAC;CAClD,MAAM,EACJ,QACA,gBAAgB,EACd,MACA,SACA,UACA,gBACA,iCACA,uCACA,6CACA,oBACA,qBACA,4BACA,gCACA,+BACA,4BACA,8BACA,+BACA,mCAEF,WACA,gBACA,YACA,UACA,QACA,SACA,aACA,oBACA,2BACE;CAEJ,cACE,iBAAiB,QAAQ,mBAAmB,GAAG,KAAK,GACpD,mBACA,CACF;CACA,KAAK,iBAAiB,QAAQ,iBAAiB;CAC/C,yCACE;EACE,WAAW;EACX,QAAQ,QAAQ;CAClB,GACA,sBACF;CACA,cAAc,IAAI,MAAM;CAKxB,MAAM,WAAW,MAAM,gBACrB,IACA,OAAO,IACP,wBACA,EAAC,KAAK,EAAC,GACP,EAAC,kBAAkB,KAAI,CACzB;CACA,kBAAuB,IAAI,UAAU,QAAQ,EAAE,YAAY,CAAC,CAAC;CAE7D,MAAM,QAAQ,eAAe,MAAM;CAGnC,MAAM,yBAAyB,IAAI,UAAU,KAAK;CAElD,IAAI,YAAY;CAChB,IAAI;CACJ,IAAI;CAEJ,MAAM,UAAU,iBAAiB,MAAM;CACvC,MAAM,yBAAyB,wBAAwB;CAEvD,MAAM,2BACJ,wBAAwB,aAAa,wBAAwB;CAC/D,IAAI,wBAAwB;EAC1B,MAAM,gBAAgB,kBAAkB,QAAQ,IAAI;EACpD,yCACE;GACE,WAAW;GACX,QAAQ;EACV,GACA,sBACF;EAMA,iBAAiB,WAAW,gCAAgC;GAC1D,aACE;GAEF,MAAM;EACR,CAAC,EAAE,YAAY,OAAM,MACnB,EAAE,QAAQ,MAAM,gBAAgB,IAAI,aAAa,CAAC,CACpD;CACF;CAEA,IAAI,kCAAkC;CACtC,IAAI,mBAAmB;CACvB,KAAK,MAAM,SAAS,CAAC,MAAM,KAAK,GAAG;EAGjC,MAAM,UAAU,IAAI,YAAY,EAAE;EAyBlC,MAAM,iBAAiB;GAAC;GAAY,kBAjBlC,sBAAsB,WAAW,aAAa,WAAW,aACrD,OAAO,kBAA2B;IAChC,MAAM,OAAO,MAAM,IAAI,YAAY,IAAI,OAAO,QAAQ,EAAE,QACtD,aACF;IACA,YAAY,SAAS,gBAAgB,OAAO;IAC5C,IAAI,WAAW;KACb,MAAM,WAAW;KACjB,QAAQ,UAAU,oBAAoB;MACpC,YAAY;MACZ,OAAO,SAAS,QAAQ;KAC1B,CAAC;IACH;IACA,OAAO;GACT,IACA,KAAA;GAEgD;EAAO;EAC7D,IAAI;GAEF,MAAM,EACJ,oBAAoB,OACpB,6BAA6B,gBAC7B,mCAAmC,iCACjC;GACJ,MAAM,EACJ,cACA,mBACA,sBACA,WACA,4BACA,kBAAkB,sBAClB,0BACE,SAAS,SAAS,OAClB,MAAM,+BACJ,IACA,SAAS,IACT,OACA,QAAQ,MACR;IACE,GAAG;IACH,yBAAyB,SAAS;IAClC,6BAA6B,SAAS;GACxC,GACA,SACA,eAAe,kBACf,gBACA;IACE;IACA;IACA;IACA,UAAU,WAAW;IACrB,aAAa;GACf,GACA,SAAS,wBACX,IACA,MAAM,6BACJ,IACA,SAAS,IACT,OACA,QAAQ,MACR,SACA,cACF;GAEJ,MAAM,6BACJ,2BAA2B,eAAe,QAAQ,IAAI;GAExD,iBAAiB,MAAM,mBACrB,IACA,OACA,QACA,SACA,UACA,UACA,cACA,4BACA,mBACA,uBACI;IACE,WAAW;IACX,mBAAmB,WAAW,gBAAgB,OAAO;IACrD,aAAa,QAAQ;GACvB,IACA,MACJ,WACA,aAAa,OACb;IACE;IACA,uBAAuB,yBAAyB;IAChD;IACA;IACA,wCACE,8CAA8C,IAC1C,8CAA8C,MAC9C,KAAA;IACN,oBAAoB,OAAO;IAC3B,oBAAoB,OAAO;IAC3B,eAAe;KACb,eAAe,kBAAkB,QAAQ,IAAI;KAC7C,eAAe;KACf,kBAAkB;IACpB;IAIA,uBAAuB,yBACnB;KACE,aAAa,QAAQ;KACrB,UAAU;MAGR,OAAO;MACP,YAAY,kBAAkB;MAC9B;KACF;IACF,IACA,KAAA;IAIJ,sBAAsB,yBAClB;KACE,aAAa;KACb,WAAW;IACb,IACA,KAAA;IAEJ,wBACE,sBAAsB,2BAClB;KACE,aAAa,QAAQ;KACrB,gBAAgB;KAChB,aAAa;KACb,eAAe;IACjB,IACA,KAAA;IAIN,sBACE,wBAAwB,UACpB;KACE,aAAa;KACb,iBAAiB;KACjB,aAAa;IACf,IACA,KAAA;GACR,GACA,UACF;GACA,YAAY;GACZ,kCAAkC;GAClC,mBAAmB;GACnB;EACF,SAAS,GAAG;GAcV,MAAM,QAAQ,QAAQ;GACtB,IAAI,SAAS,aAAa,iBAAiB;IACzC,GAAG,OAAO,qBAAqB,QAAQ,QAAQ,CAAC;IAGhD,aAAa,QAAQ,IAAI;IAMzB,kBAAkB,QAAQ,IAAI;IAC9B;GACF;GACA,IAAI,aAAa,mBACf,MAAM,IAAI,MACR,qCAAqC,QAAQ,KAAK,0CAClD,EAAC,OAAO,EAAC,CACX;GAEF,MAAM;EACR;CACF;CAEA,OAAO,gBAAgB,kDAAkD;CAEzE,MAAM,YAAY,IAAI,eAAe,IAAI,MAAM;CAC/C,MAAM,oBAA8B,CAAC;CACrC,MAAM,EAAC,SAAS,iBAAiB,YAAW,SAAS;CACrD,IAAI,CAAC,WACH,QAAQ;MACH;EACL,GAAG,OAAO,wBAAwB,SAAS;EAC3C,OAAO,WAAW,oBAAoB;EACtC,IAAI,kBACF,wBAAwB,QAAQ,IAAI;EAItC,MAAM,mBAAmB,UACtB,UACC,YACE,gBACA;GAAC,GAAG;GAAK,qCAAqC;EAAS,GACvD,QACF,GACA,cACA,mBACF,EACC,gBAAgB,eAAe,QAAQ,CAAC;EAC3C,kBAAkB,KAAK,gBAAgB;EAEvC,iBAAiB,cACf,oBACA,QAAO,OAAO,KAAK,CAAC,mBAAmB,GAAG,CAAC,CAC7C;CACF;CAEA,MAAM,gBAAgB,2BAA2B;EAC/C;EACA;EACA,aAAa,QAAQ;EACrB;CACF,CAAC;CAED,IAAI,cAAc;CAClB,IAAI,iCAAiC;EACnC,MAAM,QAAQ,YAAY,IAAI;EAC9B,GAAG,OAAO,6BAA6B;EAEvC,cAAc,cAAc,oBAAoB,EAAE,WAAW;GAC3D,MAAM,UAAU,YAAY,IAAI,IAAI;GACpC,GAAG,OAAO,kCAAkC,QAAQ,QAAQ,CAAC,EAAE,GAAG;EACpE,CAAC;CACH;CAOA,MAAM,gBAAgB,qBAClB,cACA,QAAQ,IAAI,CAAC,aAAa,eAAe,CAAC;CAK9C,MAAM,gBACJ,kBAAkB,SAAS,IACvB,gBAAgB,iBAAiB,IACjC,KAAA;CAKN,MAAM,0BAA0B,IAAI,yBAClC,IACA;EACE;EACA;EAGA,gBAAgB,qBAAqB,iBAAiB;EACtD;EACA;EACA,kBAdqB,sBACf,QAAQ,QAAQ,aAAa,IACnC,KAAA;CAaF,GACA,QACA,cACF;CAEA,cAAmB,WAAW,OAAO,KAAK,CAAC,SAAS,EAAC,OAAO,KAAI,CAAC,CAAC,CAAC;CASnE,IAAI;EACF,MAAM,eAAe,IAAI,QAAQ,yBAAyB,aAAa;CACzE,SAAS,KAAK;EACZ,UAAU,gBAAgB,KAAK,iBAAiB;CAClD,UAAU;EACR,MAAM,UAAU,SAAS;CAC3B;AACF;AAGA,IAAI,CAAC,kBAAkB,GACrB,gBACQ,UAEJ,UACE,KAAK,YAAY,GACjB,QAAQ,KACR,GAAG,QAAQ,KAAK,MAAM,CAAC,CACzB,EAAE,MAAM,OAAM,MAAK;CACjB,MAAM,qBACJ,IACA,uBAAuB,IAAI,gBAAgB,CAAC,CAC9C;CACA,MAAM;AACR,CAAC,CACL"}
+{"version":3,"file":"change-streamer.js","names":[],"sources":["../../../../../zero-cache/src/server/change-streamer.ts"],"sourcesContent":["import {consoleLogSink, LogContext} from '@rocicorp/logger';\nimport {resolver} from '@rocicorp/resolver';\nimport {assert} from '../../../shared/src/asserts.ts';\nimport {must} from '../../../shared/src/must.ts';\nimport {promiseVoid} from '../../../shared/src/resolved-promises.ts';\nimport {DatabaseInitError} from '../../../zqlite/src/db.ts';\nimport {getServerContext} from '../config/server-context.ts';\nimport {getNormalizedZeroConfig} from '../config/zero-config.ts';\nimport {deleteLiteDB} from '../db/delete-lite-db.ts';\nimport {registerSQLiteCorruptionDiagnosticTarget} from '../db/sqlite-corruption.ts';\nimport {warmupConnections} from '../db/warmup.ts';\nimport {initEventSink, publishCriticalEvent} from '../observability/events.ts';\nimport {getOrCreateGauge} from '../observability/metrics.ts';\nimport {InitCleanup} from '../services/change-source/common/init-cleanup.ts';\nimport type {PgChangeLogPurgeLocker} from '../services/change-source/common/replica-restore.ts';\nimport {initializeCustomChangeSource} from '../services/change-source/custom/change-source.ts';\nimport {initializePostgresChangeSource} from '../services/change-source/pg/change-source-init.ts';\nimport {createBackupCleanupMonitor} from '../services/change-streamer/backup-cleanup-monitor-factory.ts';\nimport {ChangeStreamerHttpServer} from '../services/change-streamer/change-streamer-http.ts';\nimport {initializeStreamer} from '../services/change-streamer/change-streamer-service.ts';\nimport type {ChangeStreamerService} from '../services/change-streamer/change-streamer.ts';\nimport {initChangeStreamerSchema} from '../services/change-streamer/schema/init.ts';\nimport {AutoResetSignal} from '../services/change-streamer/schema/tables.ts';\nimport {\n  PurgeLocker,\n  type PurgeLock,\n} from '../services/change-streamer/storer.ts';\nimport {\n  exitAfter,\n  ProcessManager,\n  runUntilKilled,\n} from '../services/life-cycle.ts';\nimport {deleteLitestreamMetaDir} from '../services/litestream/commands.ts';\nimport {\n  changeLogFileName,\n  deleteChangeLogDB,\n} from '../services/replicator/change-log-db.ts';\nimport {\n  replicationStatusError,\n  ReplicationStatusPublisher,\n} from '../services/replicator/replication-status.ts';\nimport {sqliteFileBytes} from '../services/replicator/sqlite-change-log-observability.ts';\nimport {connectPgClient} from '../types/pg.ts';\nimport {\n  broadcastWorker,\n  childWorker,\n  parentWorker,\n  singleProcessMode,\n  type ProfileResponseMessage,\n  type Worker,\n} from '../types/processes.ts';\nimport {installProfileHandler} from '../types/profiler.ts';\nimport {getShardConfig} from '../types/shards.ts';\nimport type {ReplicaFileMode} from '../workers/replicator.ts';\nimport {createLogContext} from './logging.ts';\nimport {startOtelAuto} from './otel-start.ts';\nimport {REPLICATOR_URL} from './worker-urls.ts';\n\n// Default LogContext, overridden in runWorker\nlet lc = new LogContext('info', {}, consoleLogSink);\n\nexport default async function runWorker(\n  parent: Worker,\n  env: NodeJS.ProcessEnv,\n  ...argv: string[]\n): Promise<void> {\n  installProfileHandler(parent, 'change-streamer');\n  const config = getNormalizedZeroConfig({env, argv});\n  const {\n    taskID,\n    changeStreamer: {\n      port,\n      address,\n      protocol,\n      startupDelayMs,\n      backPressureLimitHeapProportion,\n      flowControlConsensusTimeoutProportion,\n      flowControlSlowSubscriberGracePeriodSeconds,\n      maxConcurrentSwitchoverRestores,\n      pgChangeLogEnabled,\n      sqliteChangeLogMode,\n      sqliteChangeLogReadPercent,\n      sqliteChangeLogColdReadPercent,\n      sqliteChangeLogComparePercent,\n      sqliteChangeLogRetentionMs,\n      sqliteChangeLogReadBatchRows,\n      sqliteChangeLogPurgeBatchRows,\n      sqliteChangeLogBarrierTimeoutMs,\n    },\n    autoReset,\n    replicationLag,\n    litestream,\n    upstream,\n    change,\n    replica,\n    initialSync,\n    keepaliveTimeoutMs,\n    sqliteCorruptionChecks,\n  } = config;\n\n  startOtelAuto(\n    createLogContext(config, 'change-streamer', 0, false),\n    'change-streamer',\n    0,\n  );\n  lc = createLogContext(config, 'change-streamer');\n  registerSQLiteCorruptionDiagnosticTarget(\n    {\n      debugName: 'change-streamer replica',\n      dbPath: replica.file,\n    },\n    sqliteCorruptionChecks,\n  );\n  initEventSink(lc, config);\n\n  // Startup-time client used for for change-streamer initialization\n  // and handoff / takeover. Steady-state clients are managed by the\n  // change-streamer (Storer) implementation.\n  const changeDB = await connectPgClient(\n    lc,\n    change.db,\n    'change-streamer-init',\n    {max: 5},\n    {sendStringAsJson: true},\n  );\n  void warmupConnections(lc, changeDB, 'change').catch(() => {});\n\n  const shard = getShardConfig(config);\n\n  // Ensure the change DB schema is initialized/up-to-date.\n  await initChangeStreamerSchema(lc, changeDB, shard);\n\n  let purgeLock = null as PurgeLock | null;\n  let changeStreamer: ChangeStreamerService | undefined;\n  let backupURL: string | undefined;\n\n  const context = getServerContext(config);\n  const sqliteChangeLogEnabled = sqliteChangeLogMode !== 'off';\n  // The modes are cumulative, so `serve` implies `compare`.\n  const sqliteChangeLogComparing =\n    sqliteChangeLogMode === 'compare' || sqliteChangeLogMode === 'serve';\n  if (sqliteChangeLogEnabled) {\n    const changeLogFile = changeLogFileName(replica.file);\n    registerSQLiteCorruptionDiagnosticTarget(\n      {\n        debugName: 'change-streamer change-log',\n        dbPath: changeLogFile,\n      },\n      sqliteCorruptionChecks,\n    );\n    // The log's bytes are accounted for in the process that writes them. This\n    // is local disk only — the log is excluded from the litestream backup — and\n    // it is a whole-database total, because a table that left the replica left\n    // through its wal: the replica's footprint is what shrank and this is where\n    // it went.\n    getOrCreateGauge('replica', 'sqlite_change_log.file_bytes', {\n      description:\n        `The SQLite change log's total on-disk footprint: its main db file ` +\n        `plus its wal sidecars.`,\n      unit: 'By',\n    }).addCallback(async o =>\n      o.observe(await sqliteFileBytes(lc, changeLogFile)),\n    );\n  }\n\n  let waitForFirstBackupBeforeServing = false;\n  let newBackupLineage = false;\n  for (const first of [true, false]) {\n    // Resources acquired by an initialization attempt (e.g. a claimed\n    // replication slot, the purge lock) to be released if the attempt fails.\n    const cleanup = new InitCleanup(lc);\n    // When restoring from litestream, a lock is acquired to prevent change-log\n    // purges. This ensures that (this) change-streamer will be able to resume\n    // from the backup. The lock is acquired by the change source initialization\n    // just before restoring (i.e. after a replication slot is created, if\n    // applicable), and released once this change-streamer takes over the\n    // change-log.\n    const acquirePurgeLock: PgChangeLogPurgeLocker | undefined =\n      pgChangeLogEnabled && litestream.backupURL && litestream.executable\n        ? async (slotWatermark?: string) => {\n            const lock = await new PurgeLocker(lc, shard, changeDB).acquire(\n              slotWatermark,\n            );\n            purgeLock = lock === 'behind-slot' ? null : lock;\n            if (purgeLock) {\n              const acquired = purgeLock;\n              cleanup.onFailure('purge lock', () => {\n                purgeLock = null;\n                return acquired.release();\n              });\n            }\n            return lock;\n          }\n        : undefined;\n\n    const restoreOptions = {litestream, acquirePurgeLock, cleanup};\n    try {\n      // Note: This performs initial sync of the replica if necessary.\n      const {\n        pgReplicationEpoch: epoch,\n        pgReplicationSlotPerReplica: slotPerReplica,\n        pgResumeOrphanedSlotGracePeriodMs: inactiveReplicaGracePeriodMs,\n      } = upstream;\n      const {\n        changeSource,\n        subscriptionState,\n        destinationBackupURL,\n        replicaID,\n        waitForBackupBeforeServing,\n        newBackupLineage: initNewBackupLineage,\n        pgChangeLogBehindSlot,\n      } = upstream.type === 'pg'\n        ? await initializePostgresChangeSource(\n            lc,\n            upstream.db,\n            shard,\n            replica.file,\n            {\n              ...initialSync,\n              replicationSlotFailover: upstream.pgReplicationSlotFailover,\n              installPartialIndexTriggers: upstream.pgPartialIndexTriggers,\n            },\n            context,\n            replicationLag.reportIntervalMs,\n            restoreOptions,\n            {\n              epoch,\n              slotPerReplica,\n              inactiveReplicaGracePeriodMs,\n              backupV5: litestream.backupUsingV5,\n              pgChangeLog: pgChangeLogEnabled,\n            },\n            upstream.pgStreamInboundTimeoutMs,\n          )\n        : await initializeCustomChangeSource(\n            lc,\n            upstream.db,\n            shard,\n            replica.file,\n            context,\n            restoreOptions,\n          );\n\n      const replicationStatusPublisher =\n        ReplicationStatusPublisher.forReplicaFile(replica.file);\n\n      changeStreamer = await initializeStreamer(\n        lc,\n        shard,\n        taskID,\n        address,\n        protocol,\n        changeDB,\n        changeSource,\n        replicationStatusPublisher,\n        subscriptionState,\n        destinationBackupURL\n          ? {\n              backupURL: destinationBackupURL,\n              litestreamVersion: litestream.backupUsingV5 ? 'v5' : 'legacy',\n              replicaFile: replica.file,\n            }\n          : null,\n        purgeLock,\n        autoReset ?? false,\n        {\n          pgChangeLogEnabled,\n          pgChangeLogBehindSlot: pgChangeLogBehindSlot ?? false,\n          backPressureLimitHeapProportion,\n          flowControlConsensusTimeoutProportion,\n          flowControlSlowSubscriberGracePeriodMs:\n            flowControlSlowSubscriberGracePeriodSeconds > 0\n              ? flowControlSlowSubscriberGracePeriodSeconds * 1000\n              : undefined,\n          maxConcurrentSwitchoverRestores,\n          statementTimeoutMs: change.statementTimeoutMs,\n          changeLogBatchSize: change.logBatchSize,\n          sqliteCatchup: {\n            changeLogFile: changeLogFileName(replica.file),\n            readBatchRows: sqliteChangeLogReadBatchRows,\n            barrierTimeoutMs: sqliteChangeLogBarrierTimeoutMs,\n          },\n          // The presence of these options is the writer's gate. This process\n          // performs the restore and the initial sync, so the replica's\n          // identity is in hand at the moment the log is opened.\n          sqliteChangeLogWriter: sqliteChangeLogEnabled\n            ? {\n                replicaFile: replica.file,\n                identity: {\n                  // RMv2 supplies the epoch; until then the generation and the\n                  // replica ID are the whole of the identity.\n                  epoch: null,\n                  generation: subscriptionState.replicaVersion,\n                  replicaID,\n                },\n              }\n            : undefined,\n          // The purge scheduler shares the writer's gate -- mode, not the\n          // read path -- because `write` mode, with no read selector at all,\n          // is the configuration it ships in.\n          sqliteChangeLogPurge: sqliteChangeLogEnabled\n            ? {\n                retentionMs: sqliteChangeLogRetentionMs,\n                batchRows: sqliteChangeLogPurgeBatchRows,\n              }\n            : undefined,\n          // Compare mode runs both advisory checks. Postgres remains authoritative.\n          sqliteChangeLogCompare:\n            pgChangeLogEnabled && sqliteChangeLogComparing\n              ? {\n                  replicaFile: replica.file,\n                  comparePercent: sqliteChangeLogComparePercent,\n                  retentionMs: sqliteChangeLogRetentionMs,\n                  readBatchRows: sqliteChangeLogReadBatchRows,\n                }\n              : undefined,\n          // Slice 11 lands dark by default: serve mode constructs the stable\n          // router, while readPercent=0 keeps every catchup on PG and emits\n          // eligibility metrics before any canary traffic is enabled.\n          sqliteChangeLogServe:\n            sqliteChangeLogMode === 'serve'\n              ? {\n                  readPercent: sqliteChangeLogReadPercent,\n                  coldReadPercent: sqliteChangeLogColdReadPercent,\n                  retentionMs: sqliteChangeLogRetentionMs,\n                }\n              : undefined,\n        },\n        setTimeout,\n      );\n      backupURL = destinationBackupURL;\n      waitForFirstBackupBeforeServing = waitForBackupBeforeServing;\n      newBackupLineage = initNewBackupLineage;\n      break;\n    } catch (e) {\n      // Release the resources acquired by the failed attempt, e.g.:\n      // * The claimed replication slot, which would otherwise remain active\n      //   indefinitely, preventing a retry from resuming (or forking) it.\n      // * The purge lock. This is safe because the purge lock exists to\n      //   preserve change-log entries so the new change-streamer can resume\n      //   from the backup replica's watermark. An AutoResetSignal means we\n      //   can't resume from the backup replica (e.g. its replication slot is\n      //   gone), so the change-log entries the lock was protecting are no\n      //   longer needed. The retry performs a fresh initial sync with a new\n      //   replication slot, independent of the old change-log. Releasing is\n      //   also necessary to avoid a self-deadlock when CHANGE_DB ==\n      //   UPSTREAM_DB: CREATE_REPLICATION_SLOT waits for all older\n      //   transactions to finish, including this lock's open transaction.\n      await cleanup.release();\n      if (first && e instanceof AutoResetSignal) {\n        lc.warn?.(`resetting replica ${replica.file}`, e);\n        // TODO: Make deleteLiteDB work with litestream. It will probably have to be\n        //       a semantic wipe instead of a file delete.\n        deleteLiteDB(replica.file);\n        // The change log carries the identity of the replica it was written\n        // beside, and the retry performs a fresh initial sync with a new\n        // replicaVersion. Reconciliation would catch that on its own (as\n        // 'identity-mismatch') but only after the writer opened a file that is\n        // known here to be garbage.\n        deleteChangeLogDB(replica.file);\n        continue; // execute again with a fresh initial-sync\n      }\n      if (e instanceof DatabaseInitError) {\n        throw new Error(\n          `Cannot open ZERO_REPLICA_FILE at \"${replica.file}\". Please check that the path is valid.`,\n          {cause: e},\n        );\n      }\n      throw e;\n    }\n  }\n  // impossible: upstream must have advanced in order for replication to be stuck.\n  assert(changeStreamer, `resetting replica did not advance replicaVersion`);\n\n  const processes = new ProcessManager(lc, parent);\n  const profileSubWorkers: Worker[] = [];\n  const {promise: replicatorReady, resolve} = resolver();\n  if (!backupURL) {\n    resolve(); // No backup replicator to wait for, so resolve immediately.\n  } else {\n    lc.info?.('setting up backup to', backupURL);\n    config.litestream.backupURLOverride = backupURL;\n    if (newBackupLineage) {\n      deleteLitestreamMetaDir(replica.file);\n    }\n    // Start a backup replicator, which starts up the corresponding\n    // litestream backup process.\n    const backupReplicator = processes\n      .addWorker(\n        childWorker(\n          REPLICATOR_URL,\n          {...env, ZERO_LITESTREAM_BACKUP_URL_OVERRIDE: backupURL},\n          'backup' satisfies ReplicaFileMode,\n        ),\n        'supporting',\n        'backup-replicator',\n      )\n      .onceMessageType('ready', () => resolve());\n    profileSubWorkers.push(backupReplicator);\n    // Relay profileResponse messages from backup-replicator up to parent\n    backupReplicator.onMessageType<ProfileResponseMessage>(\n      'profileResponse',\n      res => parent.send(['profileResponse', res]),\n    );\n  }\n\n  const backupMonitor = createBackupCleanupMonitor({\n    lc,\n    config,\n    replicaFile: replica.file,\n    changeStreamer,\n  });\n\n  let backupReady = promiseVoid;\n  if (waitForFirstBackupBeforeServing) {\n    const start = performance.now();\n    lc.info?.(`awaiting initial backup ...`);\n\n    backupReady = backupMonitor.firstBackupReceived().then(() => {\n      const elapsed = performance.now() - start;\n      lc.info?.(`initial backup confirmed after ${elapsed.toFixed(2)}ms`);\n    });\n  }\n\n  // In RMv2, readiness additionally waits for the backup-replicator to catch\n  // up to the replication stream. This is not done in RMv1 (i.e. when the\n  // PG change-log is enabled), in which the change-streamer itself is only\n  // started after the readinessGate (plus a startup delay), and thus the\n  // backup-replicator cannot catch up before then.\n  const readinessGate = pgChangeLogEnabled\n    ? backupReady\n    : Promise.all([backupReady, replicatorReady]);\n\n  // Create the broadcast facade once: each broadcastWorker() adds permanent\n  // 'message' forwarders to every sub worker, so creating one per /profz\n  // request would leak a forwarder per request.\n  const profileWorker =\n    profileSubWorkers.length > 0\n      ? broadcastWorker(profileSubWorkers)\n      : undefined;\n  const getProfileWorker = profileWorker\n    ? () => Promise.resolve(profileWorker)\n    : undefined;\n\n  const changeStreamerWebServer = new ChangeStreamerHttpServer(\n    lc,\n    {\n      port,\n      keepaliveTimeoutMs,\n      // The startup delay is only relevant when taking over the PG change-log\n      // (RMv1 and RMv1.5), and is disabled for RMv2.\n      startupDelayMs: pgChangeLogEnabled ? startupDelayMs : 0,\n      readinessGate,\n      config,\n      getProfileWorker,\n    },\n    parent,\n    changeStreamer,\n  );\n\n  void readinessGate.then(() => parent.send(['ready', {ready: true}]));\n\n  // Note: The changeStreamer itself is not started here; it is started by the\n  //       changeStreamerWebServer after a delay to ensure that routing\n  //       routing elements have registered the server before the\n  //       change-streamer takes over the replication slot.\n  // TODO: Remove this delay and start the changeStreamer normally here once\n  //       transitioned to RMv2, since changeStreamer startup will no longer\n  //       disrupt an existing task.\n  try {\n    await runUntilKilled(lc, parent, changeStreamerWebServer, backupMonitor);\n  } catch (err) {\n    processes.logErrorAndExit(err, 'change-streamer');\n  } finally {\n    await processes.shutdown();\n  }\n}\n\n// fork()\nif (!singleProcessMode()) {\n  void exitAfter(\n    () => lc,\n    () =>\n      runWorker(\n        must(parentWorker),\n        process.env,\n        ...process.argv.slice(2),\n      ).catch(async e => {\n        await publishCriticalEvent(\n          lc,\n          replicationStatusError(lc, 'Initializing', e),\n        );\n        throw e;\n      }),\n  );\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AA2DA,IAAI,KAAK,IAAI,WAAW,QAAQ,CAAC,GAAG,cAAc;AAElD,eAA8B,UAC5B,QACA,KACA,GAAG,MACY;CACf,sBAAsB,QAAQ,iBAAiB;CAC/C,MAAM,SAAS,wBAAwB;EAAC;EAAK;CAAI,CAAC;CAClD,MAAM,EACJ,QACA,gBAAgB,EACd,MACA,SACA,UACA,gBACA,iCACA,uCACA,6CACA,iCACA,oBACA,qBACA,4BACA,gCACA,+BACA,4BACA,8BACA,+BACA,mCAEF,WACA,gBACA,YACA,UACA,QACA,SACA,aACA,oBACA,2BACE;CAEJ,cACE,iBAAiB,QAAQ,mBAAmB,GAAG,KAAK,GACpD,mBACA,CACF;CACA,KAAK,iBAAiB,QAAQ,iBAAiB;CAC/C,yCACE;EACE,WAAW;EACX,QAAQ,QAAQ;CAClB,GACA,sBACF;CACA,cAAc,IAAI,MAAM;CAKxB,MAAM,WAAW,MAAM,gBACrB,IACA,OAAO,IACP,wBACA,EAAC,KAAK,EAAC,GACP,EAAC,kBAAkB,KAAI,CACzB;CACA,kBAAuB,IAAI,UAAU,QAAQ,EAAE,YAAY,CAAC,CAAC;CAE7D,MAAM,QAAQ,eAAe,MAAM;CAGnC,MAAM,yBAAyB,IAAI,UAAU,KAAK;CAElD,IAAI,YAAY;CAChB,IAAI;CACJ,IAAI;CAEJ,MAAM,UAAU,iBAAiB,MAAM;CACvC,MAAM,yBAAyB,wBAAwB;CAEvD,MAAM,2BACJ,wBAAwB,aAAa,wBAAwB;CAC/D,IAAI,wBAAwB;EAC1B,MAAM,gBAAgB,kBAAkB,QAAQ,IAAI;EACpD,yCACE;GACE,WAAW;GACX,QAAQ;EACV,GACA,sBACF;EAMA,iBAAiB,WAAW,gCAAgC;GAC1D,aACE;GAEF,MAAM;EACR,CAAC,EAAE,YAAY,OAAM,MACnB,EAAE,QAAQ,MAAM,gBAAgB,IAAI,aAAa,CAAC,CACpD;CACF;CAEA,IAAI,kCAAkC;CACtC,IAAI,mBAAmB;CACvB,KAAK,MAAM,SAAS,CAAC,MAAM,KAAK,GAAG;EAGjC,MAAM,UAAU,IAAI,YAAY,EAAE;EAyBlC,MAAM,iBAAiB;GAAC;GAAY,kBAjBlC,sBAAsB,WAAW,aAAa,WAAW,aACrD,OAAO,kBAA2B;IAChC,MAAM,OAAO,MAAM,IAAI,YAAY,IAAI,OAAO,QAAQ,EAAE,QACtD,aACF;IACA,YAAY,SAAS,gBAAgB,OAAO;IAC5C,IAAI,WAAW;KACb,MAAM,WAAW;KACjB,QAAQ,UAAU,oBAAoB;MACpC,YAAY;MACZ,OAAO,SAAS,QAAQ;KAC1B,CAAC;IACH;IACA,OAAO;GACT,IACA,KAAA;GAEgD;EAAO;EAC7D,IAAI;GAEF,MAAM,EACJ,oBAAoB,OACpB,6BAA6B,gBAC7B,mCAAmC,iCACjC;GACJ,MAAM,EACJ,cACA,mBACA,sBACA,WACA,4BACA,kBAAkB,sBAClB,0BACE,SAAS,SAAS,OAClB,MAAM,+BACJ,IACA,SAAS,IACT,OACA,QAAQ,MACR;IACE,GAAG;IACH,yBAAyB,SAAS;IAClC,6BAA6B,SAAS;GACxC,GACA,SACA,eAAe,kBACf,gBACA;IACE;IACA;IACA;IACA,UAAU,WAAW;IACrB,aAAa;GACf,GACA,SAAS,wBACX,IACA,MAAM,6BACJ,IACA,SAAS,IACT,OACA,QAAQ,MACR,SACA,cACF;GAEJ,MAAM,6BACJ,2BAA2B,eAAe,QAAQ,IAAI;GAExD,iBAAiB,MAAM,mBACrB,IACA,OACA,QACA,SACA,UACA,UACA,cACA,4BACA,mBACA,uBACI;IACE,WAAW;IACX,mBAAmB,WAAW,gBAAgB,OAAO;IACrD,aAAa,QAAQ;GACvB,IACA,MACJ,WACA,aAAa,OACb;IACE;IACA,uBAAuB,yBAAyB;IAChD;IACA;IACA,wCACE,8CAA8C,IAC1C,8CAA8C,MAC9C,KAAA;IACN;IACA,oBAAoB,OAAO;IAC3B,oBAAoB,OAAO;IAC3B,eAAe;KACb,eAAe,kBAAkB,QAAQ,IAAI;KAC7C,eAAe;KACf,kBAAkB;IACpB;IAIA,uBAAuB,yBACnB;KACE,aAAa,QAAQ;KACrB,UAAU;MAGR,OAAO;MACP,YAAY,kBAAkB;MAC9B;KACF;IACF,IACA,KAAA;IAIJ,sBAAsB,yBAClB;KACE,aAAa;KACb,WAAW;IACb,IACA,KAAA;IAEJ,wBACE,sBAAsB,2BAClB;KACE,aAAa,QAAQ;KACrB,gBAAgB;KAChB,aAAa;KACb,eAAe;IACjB,IACA,KAAA;IAIN,sBACE,wBAAwB,UACpB;KACE,aAAa;KACb,iBAAiB;KACjB,aAAa;IACf,IACA,KAAA;GACR,GACA,UACF;GACA,YAAY;GACZ,kCAAkC;GAClC,mBAAmB;GACnB;EACF,SAAS,GAAG;GAcV,MAAM,QAAQ,QAAQ;GACtB,IAAI,SAAS,aAAa,iBAAiB;IACzC,GAAG,OAAO,qBAAqB,QAAQ,QAAQ,CAAC;IAGhD,aAAa,QAAQ,IAAI;IAMzB,kBAAkB,QAAQ,IAAI;IAC9B;GACF;GACA,IAAI,aAAa,mBACf,MAAM,IAAI,MACR,qCAAqC,QAAQ,KAAK,0CAClD,EAAC,OAAO,EAAC,CACX;GAEF,MAAM;EACR;CACF;CAEA,OAAO,gBAAgB,kDAAkD;CAEzE,MAAM,YAAY,IAAI,eAAe,IAAI,MAAM;CAC/C,MAAM,oBAA8B,CAAC;CACrC,MAAM,EAAC,SAAS,iBAAiB,YAAW,SAAS;CACrD,IAAI,CAAC,WACH,QAAQ;MACH;EACL,GAAG,OAAO,wBAAwB,SAAS;EAC3C,OAAO,WAAW,oBAAoB;EACtC,IAAI,kBACF,wBAAwB,QAAQ,IAAI;EAItC,MAAM,mBAAmB,UACtB,UACC,YACE,gBACA;GAAC,GAAG;GAAK,qCAAqC;EAAS,GACvD,QACF,GACA,cACA,mBACF,EACC,gBAAgB,eAAe,QAAQ,CAAC;EAC3C,kBAAkB,KAAK,gBAAgB;EAEvC,iBAAiB,cACf,oBACA,QAAO,OAAO,KAAK,CAAC,mBAAmB,GAAG,CAAC,CAC7C;CACF;CAEA,MAAM,gBAAgB,2BAA2B;EAC/C;EACA;EACA,aAAa,QAAQ;EACrB;CACF,CAAC;CAED,IAAI,cAAc;CAClB,IAAI,iCAAiC;EACnC,MAAM,QAAQ,YAAY,IAAI;EAC9B,GAAG,OAAO,6BAA6B;EAEvC,cAAc,cAAc,oBAAoB,EAAE,WAAW;GAC3D,MAAM,UAAU,YAAY,IAAI,IAAI;GACpC,GAAG,OAAO,kCAAkC,QAAQ,QAAQ,CAAC,EAAE,GAAG;EACpE,CAAC;CACH;CAOA,MAAM,gBAAgB,qBAClB,cACA,QAAQ,IAAI,CAAC,aAAa,eAAe,CAAC;CAK9C,MAAM,gBACJ,kBAAkB,SAAS,IACvB,gBAAgB,iBAAiB,IACjC,KAAA;CAKN,MAAM,0BAA0B,IAAI,yBAClC,IACA;EACE;EACA;EAGA,gBAAgB,qBAAqB,iBAAiB;EACtD;EACA;EACA,kBAdqB,sBACf,QAAQ,QAAQ,aAAa,IACnC,KAAA;CAaF,GACA,QACA,cACF;CAEA,cAAmB,WAAW,OAAO,KAAK,CAAC,SAAS,EAAC,OAAO,KAAI,CAAC,CAAC,CAAC;CASnE,IAAI;EACF,MAAM,eAAe,IAAI,QAAQ,yBAAyB,aAAa;CACzE,SAAS,KAAK;EACZ,UAAU,gBAAgB,KAAK,iBAAiB;CAClD,UAAU;EACR,MAAM,UAAU,SAAS;CAC3B;AACF;AAGA,IAAI,CAAC,kBAAkB,GACrB,gBACQ,UAEJ,UACE,KAAK,YAAY,GACjB,QAAQ,KACR,GAAG,QAAQ,KAAK,MAAM,CAAC,CACzB,EAAE,MAAM,OAAM,MAAK;CACjB,MAAM,qBACJ,IACA,uBAAuB,IAAI,gBAAgB,CAAC,CAC9C;CACA,MAAM;AACR,CAAC,CACL"}

package/out/zero-cache/src/server/main.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"main.d.ts","sourceRoot":"","sources":["../../../../../zero-cache/src/server/main.ts"],"names":[],"mappings":"AAeA,OAAO,EAML,KAAK,MAAM,EACZ,MAAM,uBAAuB,CAAC;AAwB/B,wBAA8B,SAAS,CACrC,MAAM,EAAE,MAAM,EACd,GAAG,EAAE,MAAM,CAAC,UAAU,GACrB,OAAO,CAAC,IAAI,CAAC,CA+Kf"}
+{"version":3,"file":"main.d.ts","sourceRoot":"","sources":["../../../../../zero-cache/src/server/main.ts"],"names":[],"mappings":"AAeA,OAAO,EAOL,KAAK,MAAM,EACZ,MAAM,uBAAuB,CAAC;AAwB/B,wBAA8B,SAAS,CACrC,MAAM,EAAE,MAAM,EACd,GAAG,EAAE,MAAM,CAAC,UAAU,GACrB,OAAO,CAAC,IAAI,CAAC,CAyLf"}

package/out/zero-cache/src/server/main.js

@@ -50,7 +50,7 @@ async function runWorker(parent, env) {
 	let changeStreamer;
 	if (runChangeStreamer) {
 		const { promise: changeStreamerReady, resolve: changeStreamerStarted } = resolver();
-		changeStreamer = loadWorker(CHANGE_STREAMER_URL, "supporting").once("message", changeStreamerStarted);
+		changeStreamer = loadWorker(CHANGE_STREAMER_URL, numSyncers > 0 ? "supporting" : "user-facing").once("message", changeStreamerStarted);
 		await changeStreamerReady;
 	}
 	if (numSyncers > 0) {
@@ -71,6 +71,10 @@ async function runWorker(parent, env) {
 			subscribeTo(lc, replicator);
 			resolve();
 		});
+		replicator.onMessageType("drainRequested", (msg) => {
+			lc.info?.(`drain requested by the replicator`);
+			parent.send(["drainRequested", msg]);
+		});
 		await replicaReady;
 		const notifier = createNotifierFrom(lc, replicator);
 		for (let i = 0; i < numSyncers; i++) syncers.push(loadWorker(SYNCER_URL, "user-facing", i, mode, String(i)));

package/out/zero-cache/src/server/main.js.map

@@ -1 +1 @@
-{"version":3,"file":"main.js","names":[],"sources":["../../../../../zero-cache/src/server/main.ts"],"sourcesContent":["import path from 'node:path';\nimport {consoleLogSink, LogContext} from '@rocicorp/logger';\nimport {resolver} from '@rocicorp/resolver';\nimport {must} from '../../../shared/src/must.ts';\nimport {runsChangeStreamer} from '../config/normalize.ts';\nimport {getNormalizedZeroConfig} from '../config/zero-config.ts';\nimport {registerSQLiteCorruptionDiagnosticTarget} from '../db/sqlite-corruption.ts';\nimport {initEventSink} from '../observability/events.ts';\nimport {\n  exitAfter,\n  ProcessManager,\n  recordStartupDurationMs,\n  runUntilKilled,\n  type WorkerType,\n} from '../services/life-cycle.ts';\nimport {\n  childWorker,\n  parentWorker,\n  singleProcessMode,\n  type ProfileMessage,\n  type ProfileResponseMessage,\n  type Worker,\n} from '../types/processes.ts';\nimport {\n  createNotifierFrom,\n  handleSubscriptionsFrom,\n  type ReplicaFileMode,\n  subscribeTo,\n} from '../workers/replicator.ts';\nimport {createLogContext} from './logging.ts';\nimport {startOtelAuto} from './otel-start.ts';\nimport {WorkerDispatcher} from './worker-dispatcher.ts';\nimport {\n  CHANGE_STREAMER_URL,\n  MUTATOR_URL,\n  REAPER_URL,\n  REPLICATOR_URL,\n  SHADOW_SYNCER_URL,\n  SYNCER_URL,\n} from './worker-urls.ts';\n\nconst clientConnectionBifurcated = false;\n\n// Default LogContext, overridden in runWorker\nlet lc = new LogContext('info', {}, consoleLogSink);\n\nexport default async function runWorker(\n  parent: Worker,\n  env: NodeJS.ProcessEnv,\n): Promise<void> {\n  const startMs = Date.now();\n  const config = getNormalizedZeroConfig({env});\n\n  startOtelAuto(\n    createLogContext(config, 'dispatcher', 0, false),\n    'dispatcher',\n    0,\n  );\n  lc = createLogContext(config, 'dispatcher');\n  registerSQLiteCorruptionDiagnosticTarget(\n    {\n      debugName: 'dispatcher replica',\n      dbPath: config.replica.file,\n    },\n    config.sqliteCorruptionChecks,\n  );\n  initEventSink(lc, config);\n\n  const processes = new ProcessManager(lc, parent);\n\n  const {numSyncWorkers: numSyncers} = config;\n  if (config.enableCrudMutations && config.upstream.maxConns < numSyncers) {\n    throw new Error(\n      `Insufficient upstream connections (${config.upstream.maxConns}) for ${numSyncers} syncers.` +\n        `Increase ZERO_UPSTREAM_MAX_CONNS or decrease ZERO_NUM_SYNC_WORKERS (which defaults to available cores).`,\n    );\n  }\n  if (config.cvr.maxConns < numSyncers) {\n    throw new Error(\n      `Insufficient cvr connections (${config.cvr.maxConns}) for ${numSyncers} syncers.` +\n        `Increase ZERO_CVR_MAX_CONNS or decrease ZERO_NUM_SYNC_WORKERS (which defaults to available cores).`,\n    );\n  }\n\n  const internalFlags: string[] =\n    numSyncers === 0\n      ? []\n      : [\n          '--upstream-max-conns-per-worker',\n          String(Math.floor(config.upstream.maxConns / numSyncers)),\n          '--cvr-max-conns-per-worker',\n          String(Math.floor(config.cvr.maxConns / numSyncers)),\n        ];\n\n  const allSubWorkers: Worker[] = [];\n  function loadWorker(\n    moduleUrl: URL,\n    type: WorkerType,\n    id?: string | number,\n    ...args: string[]\n  ): Worker {\n    const worker = childWorker(moduleUrl, env, ...args, ...internalFlags);\n    const name = path.basename(moduleUrl.pathname) + (id ? ` (${id})` : '');\n    const w = processes.addWorker(worker, type, name);\n    allSubWorkers.push(w);\n    w.onMessageType<ProfileResponseMessage>('profileResponse', res =>\n      parent.send(['profileResponse', res]),\n    );\n    return w;\n  }\n\n  const {taskID, litestream} = config;\n  const runChangeStreamer = runsChangeStreamer(config);\n  const sqliteChangeLogEnabled =\n    config.changeStreamer.sqliteChangeLogMode !== 'off';\n  if (sqliteChangeLogEnabled && !runChangeStreamer) {\n    // A multi-node deployment configures every task from one environment, so\n    // the change log's options reach the view-syncers too. The log lives in\n    // the change-streamer, and a task that connects to one simply never writes\n    // a log. Warning rather than refusing to start is what makes the\n    // fleet-wide environment -- often the only knob an operator has -- able to\n    // express the rollout at all.\n    lc.warn?.(\n      `ignoring --change-streamer-sqlite-change-log-mode=` +\n        `${config.changeStreamer.sqliteChangeLogMode}: this task connects to ` +\n        `a change-streamer rather than running one, and the SQLite change log ` +\n        `lives in the change-streamer`,\n    );\n  }\n\n  let changeStreamer: Worker | undefined;\n\n  if (runChangeStreamer) {\n    const {promise: changeStreamerReady, resolve: changeStreamerStarted} =\n      resolver();\n    changeStreamer = loadWorker(CHANGE_STREAMER_URL, 'supporting').once(\n      'message',\n      changeStreamerStarted,\n    );\n    // Wait for the change-streamer to be ready to guarantee that a replica\n    // file is present.\n    await changeStreamerReady;\n  }\n\n  if (numSyncers > 0) {\n    const {promise: reaperReady, resolve: reaperStarted} = resolver();\n    loadWorker(REAPER_URL, 'supporting').once('message', reaperStarted);\n    // Before starting the view-syncers, ensure that the reaper has started\n    // up, indicating that any CVR db migrations have been performed.\n    await reaperReady;\n  }\n\n  // Only run the shadow-sync canary on the replication-manager (or in\n  // single-node mode, where it also owns upstream). Running on every\n  // view-syncer would hammer the upstream with N redundant canaries.\n  if (config.shadowSync.enabled && runChangeStreamer) {\n    const {promise: shadowReady, resolve: shadowStarted} = resolver();\n    loadWorker(SHADOW_SYNCER_URL, 'supporting').once('message', shadowStarted);\n    await shadowReady;\n  }\n\n  const syncers: Worker[] = [];\n  if (numSyncers) {\n    const mode: ReplicaFileMode =\n      runChangeStreamer && litestream.backupURL ? 'serving-copy' : 'serving';\n    const {promise: replicaReady, resolve} = resolver();\n    const replicator = loadWorker(\n      REPLICATOR_URL,\n      'supporting',\n      mode,\n      mode,\n    ).once('message', () => {\n      subscribeTo(lc, replicator);\n      resolve();\n    });\n    await replicaReady;\n\n    const notifier = createNotifierFrom(lc, replicator);\n    for (let i = 0; i < numSyncers; i++) {\n      syncers.push(loadWorker(SYNCER_URL, 'user-facing', i, mode, String(i)));\n    }\n    syncers.forEach(syncer => handleSubscriptionsFrom(lc, syncer, notifier));\n  }\n  let mutator: Worker | undefined;\n  if (clientConnectionBifurcated) {\n    mutator = loadWorker(MUTATOR_URL, 'supporting', 'mutator');\n  }\n\n  lc.info?.('waiting for workers to be ready ...');\n  const logWaiting = setInterval(\n    () => lc.info?.(`still waiting for ${processes.initializing().join(', ')}`),\n    10_000,\n  );\n  await processes.allWorkersReady();\n  clearInterval(logWaiting);\n  const startupDurationMs = Date.now() - startMs;\n  lc.info?.(`all workers ready (${startupDurationMs} ms)`);\n  recordStartupDurationMs(startupDurationMs);\n\n  parent.send(['ready', {ready: true}]);\n  parent.onMessageType<ProfileMessage>('profile', req => {\n    for (const w of allSubWorkers) {\n      w.send(['profile', req]);\n    }\n  });\n\n  try {\n    await runUntilKilled(\n      lc,\n      parent,\n      new WorkerDispatcher(\n        lc,\n        taskID,\n        parent,\n        syncers,\n        mutator,\n        changeStreamer,\n      ),\n    );\n  } catch (err) {\n    processes.logErrorAndExit(err, 'dispatcher');\n  } finally {\n    await processes.shutdown();\n  }\n}\n\nif (!singleProcessMode()) {\n  void exitAfter(\n    () => lc,\n    () => runWorker(must(parentWorker), process.env),\n  );\n}\n"],"mappings":";;;;;;;;;;;;;;;;AA4CA,IAAI,KAAK,IAAI,WAAW,QAAQ,CAAC,GAAG,cAAc;AAElD,eAA8B,UAC5B,QACA,KACe;CACf,MAAM,UAAU,KAAK,IAAI;CACzB,MAAM,SAAS,wBAAwB,EAAC,IAAG,CAAC;CAE5C,cACE,iBAAiB,QAAQ,cAAc,GAAG,KAAK,GAC/C,cACA,CACF;CACA,KAAK,iBAAiB,QAAQ,YAAY;CAC1C,yCACE;EACE,WAAW;EACX,QAAQ,OAAO,QAAQ;CACzB,GACA,OAAO,sBACT;CACA,cAAc,IAAI,MAAM;CAExB,MAAM,YAAY,IAAI,eAAe,IAAI,MAAM;CAE/C,MAAM,EAAC,gBAAgB,eAAc;CACrC,IAAI,OAAO,uBAAuB,OAAO,SAAS,WAAW,YAC3D,MAAM,IAAI,MACR,sCAAsC,OAAO,SAAS,SAAS,QAAQ,WAAW,iHAEpF;CAEF,IAAI,OAAO,IAAI,WAAW,YACxB,MAAM,IAAI,MACR,iCAAiC,OAAO,IAAI,SAAS,QAAQ,WAAW,4GAE1E;CAGF,MAAM,gBACJ,eAAe,IACX,CAAC,IACD;EACE;EACA,OAAO,KAAK,MAAM,OAAO,SAAS,WAAW,UAAU,CAAC;EACxD;EACA,OAAO,KAAK,MAAM,OAAO,IAAI,WAAW,UAAU,CAAC;CACrD;CAEN,MAAM,gBAA0B,CAAC;CACjC,SAAS,WACP,WACA,MACA,IACA,GAAG,MACK;EACR,MAAM,SAAS,YAAY,WAAW,KAAK,GAAG,MAAM,GAAG,aAAa;EACpE,MAAM,OAAO,KAAK,SAAS,UAAU,QAAQ,KAAK,KAAK,KAAK,GAAG,KAAK;EACpE,MAAM,IAAI,UAAU,UAAU,QAAQ,MAAM,IAAI;EAChD,cAAc,KAAK,CAAC;EACpB,EAAE,cAAsC,oBAAmB,QACzD,OAAO,KAAK,CAAC,mBAAmB,GAAG,CAAC,CACtC;EACA,OAAO;CACT;CAEA,MAAM,EAAC,QAAQ,eAAc;CAC7B,MAAM,oBAAoB,mBAAmB,MAAM;CAGnD,IADE,OAAO,eAAe,wBAAwB,SAClB,CAAC,mBAO7B,GAAG,OACD,qDACK,OAAO,eAAe,oBAAoB,0HAGjD;CAGF,IAAI;CAEJ,IAAI,mBAAmB;EACrB,MAAM,EAAC,SAAS,qBAAqB,SAAS,0BAC5C,SAAS;EACX,iBAAiB,WAAW,qBAAqB,YAAY,EAAE,KAC7D,WACA,qBACF;EAGA,MAAM;CACR;CAEA,IAAI,aAAa,GAAG;EAClB,MAAM,EAAC,SAAS,aAAa,SAAS,kBAAiB,SAAS;EAChE,WAAW,YAAY,YAAY,EAAE,KAAK,WAAW,aAAa;EAGlE,MAAM;CACR;CAKA,IAAI,OAAO,WAAW,WAAW,mBAAmB;EAClD,MAAM,EAAC,SAAS,aAAa,SAAS,kBAAiB,SAAS;EAChE,WAAW,mBAAmB,YAAY,EAAE,KAAK,WAAW,aAAa;EACzE,MAAM;CACR;CAEA,MAAM,UAAoB,CAAC;CAC3B,IAAI,YAAY;EACd,MAAM,OACJ,qBAAqB,WAAW,YAAY,iBAAiB;EAC/D,MAAM,EAAC,SAAS,cAAc,YAAW,SAAS;EAClD,MAAM,aAAa,WACjB,gBACA,cACA,MACA,IACF,EAAE,KAAK,iBAAiB;GACtB,YAAY,IAAI,UAAU;GAC1B,QAAQ;EACV,CAAC;EACD,MAAM;EAEN,MAAM,WAAW,mBAAmB,IAAI,UAAU;EAClD,KAAK,IAAI,IAAI,GAAG,IAAI,YAAY,KAC9B,QAAQ,KAAK,WAAW,YAAY,eAAe,GAAG,MAAM,OAAO,CAAC,CAAC,CAAC;EAExE,QAAQ,SAAQ,WAAU,wBAAwB,IAAI,QAAQ,QAAQ,CAAC;CACzE;CACA,IAAI;CAKJ,GAAG,OAAO,qCAAqC;CAC/C,MAAM,aAAa,kBACX,GAAG,OAAO,qBAAqB,UAAU,aAAa,EAAE,KAAK,IAAI,GAAG,GAC1E,GACF;CACA,MAAM,UAAU,gBAAgB;CAChC,cAAc,UAAU;CACxB,MAAM,oBAAoB,KAAK,IAAI,IAAI;CACvC,GAAG,OAAO,sBAAsB,kBAAkB,KAAK;CACvD,wBAAwB,iBAAiB;CAEzC,OAAO,KAAK,CAAC,SAAS,EAAC,OAAO,KAAI,CAAC,CAAC;CACpC,OAAO,cAA8B,YAAW,QAAO;EACrD,KAAK,MAAM,KAAK,eACd,EAAE,KAAK,CAAC,WAAW,GAAG,CAAC;CAE3B,CAAC;CAED,IAAI;EACF,MAAM,eACJ,IACA,QACA,IAAI,iBACF,IACA,QACA,QACA,SACA,SACA,cACF,CACF;CACF,SAAS,KAAK;EACZ,UAAU,gBAAgB,KAAK,YAAY;CAC7C,UAAU;EACR,MAAM,UAAU,SAAS;CAC3B;AACF;AAEA,IAAI,CAAC,kBAAkB,GACrB,gBACQ,UACA,UAAU,KAAK,YAAY,GAAG,QAAQ,GAAG,CACjD"}
+{"version":3,"file":"main.js","names":[],"sources":["../../../../../zero-cache/src/server/main.ts"],"sourcesContent":["import path from 'node:path';\nimport {consoleLogSink, LogContext} from '@rocicorp/logger';\nimport {resolver} from '@rocicorp/resolver';\nimport {must} from '../../../shared/src/must.ts';\nimport {runsChangeStreamer} from '../config/normalize.ts';\nimport {getNormalizedZeroConfig} from '../config/zero-config.ts';\nimport {registerSQLiteCorruptionDiagnosticTarget} from '../db/sqlite-corruption.ts';\nimport {initEventSink} from '../observability/events.ts';\nimport {\n  exitAfter,\n  ProcessManager,\n  recordStartupDurationMs,\n  runUntilKilled,\n  type WorkerType,\n} from '../services/life-cycle.ts';\nimport {\n  childWorker,\n  parentWorker,\n  singleProcessMode,\n  type DrainRequestedMessage,\n  type ProfileMessage,\n  type ProfileResponseMessage,\n  type Worker,\n} from '../types/processes.ts';\nimport {\n  createNotifierFrom,\n  handleSubscriptionsFrom,\n  subscribeTo,\n  type ReplicaFileMode,\n} from '../workers/replicator.ts';\nimport {createLogContext} from './logging.ts';\nimport {startOtelAuto} from './otel-start.ts';\nimport {WorkerDispatcher} from './worker-dispatcher.ts';\nimport {\n  CHANGE_STREAMER_URL,\n  MUTATOR_URL,\n  REAPER_URL,\n  REPLICATOR_URL,\n  SHADOW_SYNCER_URL,\n  SYNCER_URL,\n} from './worker-urls.ts';\n\nconst clientConnectionBifurcated = false;\n\n// Default LogContext, overridden in runWorker\nlet lc = new LogContext('info', {}, consoleLogSink);\n\nexport default async function runWorker(\n  parent: Worker,\n  env: NodeJS.ProcessEnv,\n): Promise<void> {\n  const startMs = Date.now();\n  const config = getNormalizedZeroConfig({env});\n\n  startOtelAuto(\n    createLogContext(config, 'dispatcher', 0, false),\n    'dispatcher',\n    0,\n  );\n  lc = createLogContext(config, 'dispatcher');\n  registerSQLiteCorruptionDiagnosticTarget(\n    {\n      debugName: 'dispatcher replica',\n      dbPath: config.replica.file,\n    },\n    config.sqliteCorruptionChecks,\n  );\n  initEventSink(lc, config);\n\n  const processes = new ProcessManager(lc, parent);\n\n  const {numSyncWorkers: numSyncers} = config;\n  if (config.enableCrudMutations && config.upstream.maxConns < numSyncers) {\n    throw new Error(\n      `Insufficient upstream connections (${config.upstream.maxConns}) for ${numSyncers} syncers.` +\n        `Increase ZERO_UPSTREAM_MAX_CONNS or decrease ZERO_NUM_SYNC_WORKERS (which defaults to available cores).`,\n    );\n  }\n  if (config.cvr.maxConns < numSyncers) {\n    throw new Error(\n      `Insufficient cvr connections (${config.cvr.maxConns}) for ${numSyncers} syncers.` +\n        `Increase ZERO_CVR_MAX_CONNS or decrease ZERO_NUM_SYNC_WORKERS (which defaults to available cores).`,\n    );\n  }\n\n  const internalFlags: string[] =\n    numSyncers === 0\n      ? []\n      : [\n          '--upstream-max-conns-per-worker',\n          String(Math.floor(config.upstream.maxConns / numSyncers)),\n          '--cvr-max-conns-per-worker',\n          String(Math.floor(config.cvr.maxConns / numSyncers)),\n        ];\n\n  const allSubWorkers: Worker[] = [];\n  function loadWorker(\n    moduleUrl: URL,\n    type: WorkerType,\n    id?: string | number,\n    ...args: string[]\n  ): Worker {\n    const worker = childWorker(moduleUrl, env, ...args, ...internalFlags);\n    const name = path.basename(moduleUrl.pathname) + (id ? ` (${id})` : '');\n    const w = processes.addWorker(worker, type, name);\n    allSubWorkers.push(w);\n    w.onMessageType<ProfileResponseMessage>('profileResponse', res =>\n      parent.send(['profileResponse', res]),\n    );\n    return w;\n  }\n\n  const {taskID, litestream} = config;\n  const runChangeStreamer = runsChangeStreamer(config);\n  const sqliteChangeLogEnabled =\n    config.changeStreamer.sqliteChangeLogMode !== 'off';\n  if (sqliteChangeLogEnabled && !runChangeStreamer) {\n    // A multi-node deployment configures every task from one environment, so\n    // the change log's options reach the view-syncers too. The log lives in\n    // the change-streamer, and a task that connects to one simply never writes\n    // a log. Warning rather than refusing to start is what makes the\n    // fleet-wide environment -- often the only knob an operator has -- able to\n    // express the rollout at all.\n    lc.warn?.(\n      `ignoring --change-streamer-sqlite-change-log-mode=` +\n        `${config.changeStreamer.sqliteChangeLogMode}: this task connects to ` +\n        `a change-streamer rather than running one, and the SQLite change log ` +\n        `lives in the change-streamer`,\n    );\n  }\n\n  let changeStreamer: Worker | undefined;\n\n  if (runChangeStreamer) {\n    const {promise: changeStreamerReady, resolve: changeStreamerStarted} =\n      resolver();\n    // In a dedicated replication-manager (i.e. no syncers), the\n    // change-streamer is the user-facing worker: it receives the SIGTERM\n    // first so that it can drain subscribers before other supporting\n    // workers are stopped. When syncers run in the same task, the\n    // change-streamer supports them and thus must outlive them.\n    changeStreamer = loadWorker(\n      CHANGE_STREAMER_URL,\n      numSyncers > 0 ? 'supporting' : 'user-facing',\n    ).once('message', changeStreamerStarted);\n    // Wait for the change-streamer to be ready to guarantee that a replica\n    // file is present.\n    await changeStreamerReady;\n  }\n\n  if (numSyncers > 0) {\n    const {promise: reaperReady, resolve: reaperStarted} = resolver();\n    loadWorker(REAPER_URL, 'supporting').once('message', reaperStarted);\n    // Before starting the view-syncers, ensure that the reaper has started\n    // up, indicating that any CVR db migrations have been performed.\n    await reaperReady;\n  }\n\n  // Only run the shadow-sync canary on the replication-manager (or in\n  // single-node mode, where it also owns upstream). Running on every\n  // view-syncer would hammer the upstream with N redundant canaries.\n  if (config.shadowSync.enabled && runChangeStreamer) {\n    const {promise: shadowReady, resolve: shadowStarted} = resolver();\n    loadWorker(SHADOW_SYNCER_URL, 'supporting').once('message', shadowStarted);\n    await shadowReady;\n  }\n\n  const syncers: Worker[] = [];\n  if (numSyncers) {\n    const mode: ReplicaFileMode =\n      runChangeStreamer && litestream.backupURL ? 'serving-copy' : 'serving';\n    const {promise: replicaReady, resolve} = resolver();\n    const replicator = loadWorker(\n      REPLICATOR_URL,\n      'supporting',\n      mode,\n      mode,\n    ).once('message', () => {\n      subscribeTo(lc, replicator);\n      resolve();\n    });\n    // Relay drain requests to the runner, which coordinates the SelfDrain.\n    replicator.onMessageType<DrainRequestedMessage>('drainRequested', msg => {\n      lc.info?.(`drain requested by the replicator`);\n      parent.send<DrainRequestedMessage>(['drainRequested', msg]);\n    });\n    await replicaReady;\n\n    const notifier = createNotifierFrom(lc, replicator);\n    for (let i = 0; i < numSyncers; i++) {\n      syncers.push(loadWorker(SYNCER_URL, 'user-facing', i, mode, String(i)));\n    }\n    syncers.forEach(syncer => handleSubscriptionsFrom(lc, syncer, notifier));\n  }\n  let mutator: Worker | undefined;\n  if (clientConnectionBifurcated) {\n    mutator = loadWorker(MUTATOR_URL, 'supporting', 'mutator');\n  }\n\n  lc.info?.('waiting for workers to be ready ...');\n  const logWaiting = setInterval(\n    () => lc.info?.(`still waiting for ${processes.initializing().join(', ')}`),\n    10_000,\n  );\n  await processes.allWorkersReady();\n  clearInterval(logWaiting);\n  const startupDurationMs = Date.now() - startMs;\n  lc.info?.(`all workers ready (${startupDurationMs} ms)`);\n  recordStartupDurationMs(startupDurationMs);\n\n  parent.send(['ready', {ready: true}]);\n  parent.onMessageType<ProfileMessage>('profile', req => {\n    for (const w of allSubWorkers) {\n      w.send(['profile', req]);\n    }\n  });\n\n  try {\n    await runUntilKilled(\n      lc,\n      parent,\n      new WorkerDispatcher(\n        lc,\n        taskID,\n        parent,\n        syncers,\n        mutator,\n        changeStreamer,\n      ),\n    );\n  } catch (err) {\n    processes.logErrorAndExit(err, 'dispatcher');\n  } finally {\n    await processes.shutdown();\n  }\n}\n\nif (!singleProcessMode()) {\n  void exitAfter(\n    () => lc,\n    () => runWorker(must(parentWorker), process.env),\n  );\n}\n"],"mappings":";;;;;;;;;;;;;;;;AA6CA,IAAI,KAAK,IAAI,WAAW,QAAQ,CAAC,GAAG,cAAc;AAElD,eAA8B,UAC5B,QACA,KACe;CACf,MAAM,UAAU,KAAK,IAAI;CACzB,MAAM,SAAS,wBAAwB,EAAC,IAAG,CAAC;CAE5C,cACE,iBAAiB,QAAQ,cAAc,GAAG,KAAK,GAC/C,cACA,CACF;CACA,KAAK,iBAAiB,QAAQ,YAAY;CAC1C,yCACE;EACE,WAAW;EACX,QAAQ,OAAO,QAAQ;CACzB,GACA,OAAO,sBACT;CACA,cAAc,IAAI,MAAM;CAExB,MAAM,YAAY,IAAI,eAAe,IAAI,MAAM;CAE/C,MAAM,EAAC,gBAAgB,eAAc;CACrC,IAAI,OAAO,uBAAuB,OAAO,SAAS,WAAW,YAC3D,MAAM,IAAI,MACR,sCAAsC,OAAO,SAAS,SAAS,QAAQ,WAAW,iHAEpF;CAEF,IAAI,OAAO,IAAI,WAAW,YACxB,MAAM,IAAI,MACR,iCAAiC,OAAO,IAAI,SAAS,QAAQ,WAAW,4GAE1E;CAGF,MAAM,gBACJ,eAAe,IACX,CAAC,IACD;EACE;EACA,OAAO,KAAK,MAAM,OAAO,SAAS,WAAW,UAAU,CAAC;EACxD;EACA,OAAO,KAAK,MAAM,OAAO,IAAI,WAAW,UAAU,CAAC;CACrD;CAEN,MAAM,gBAA0B,CAAC;CACjC,SAAS,WACP,WACA,MACA,IACA,GAAG,MACK;EACR,MAAM,SAAS,YAAY,WAAW,KAAK,GAAG,MAAM,GAAG,aAAa;EACpE,MAAM,OAAO,KAAK,SAAS,UAAU,QAAQ,KAAK,KAAK,KAAK,GAAG,KAAK;EACpE,MAAM,IAAI,UAAU,UAAU,QAAQ,MAAM,IAAI;EAChD,cAAc,KAAK,CAAC;EACpB,EAAE,cAAsC,oBAAmB,QACzD,OAAO,KAAK,CAAC,mBAAmB,GAAG,CAAC,CACtC;EACA,OAAO;CACT;CAEA,MAAM,EAAC,QAAQ,eAAc;CAC7B,MAAM,oBAAoB,mBAAmB,MAAM;CAGnD,IADE,OAAO,eAAe,wBAAwB,SAClB,CAAC,mBAO7B,GAAG,OACD,qDACK,OAAO,eAAe,oBAAoB,0HAGjD;CAGF,IAAI;CAEJ,IAAI,mBAAmB;EACrB,MAAM,EAAC,SAAS,qBAAqB,SAAS,0BAC5C,SAAS;EAMX,iBAAiB,WACf,qBACA,aAAa,IAAI,eAAe,aAClC,EAAE,KAAK,WAAW,qBAAqB;EAGvC,MAAM;CACR;CAEA,IAAI,aAAa,GAAG;EAClB,MAAM,EAAC,SAAS,aAAa,SAAS,kBAAiB,SAAS;EAChE,WAAW,YAAY,YAAY,EAAE,KAAK,WAAW,aAAa;EAGlE,MAAM;CACR;CAKA,IAAI,OAAO,WAAW,WAAW,mBAAmB;EAClD,MAAM,EAAC,SAAS,aAAa,SAAS,kBAAiB,SAAS;EAChE,WAAW,mBAAmB,YAAY,EAAE,KAAK,WAAW,aAAa;EACzE,MAAM;CACR;CAEA,MAAM,UAAoB,CAAC;CAC3B,IAAI,YAAY;EACd,MAAM,OACJ,qBAAqB,WAAW,YAAY,iBAAiB;EAC/D,MAAM,EAAC,SAAS,cAAc,YAAW,SAAS;EAClD,MAAM,aAAa,WACjB,gBACA,cACA,MACA,IACF,EAAE,KAAK,iBAAiB;GACtB,YAAY,IAAI,UAAU;GAC1B,QAAQ;EACV,CAAC;EAED,WAAW,cAAqC,mBAAkB,QAAO;GACvE,GAAG,OAAO,mCAAmC;GAC7C,OAAO,KAA4B,CAAC,kBAAkB,GAAG,CAAC;EAC5D,CAAC;EACD,MAAM;EAEN,MAAM,WAAW,mBAAmB,IAAI,UAAU;EAClD,KAAK,IAAI,IAAI,GAAG,IAAI,YAAY,KAC9B,QAAQ,KAAK,WAAW,YAAY,eAAe,GAAG,MAAM,OAAO,CAAC,CAAC,CAAC;EAExE,QAAQ,SAAQ,WAAU,wBAAwB,IAAI,QAAQ,QAAQ,CAAC;CACzE;CACA,IAAI;CAKJ,GAAG,OAAO,qCAAqC;CAC/C,MAAM,aAAa,kBACX,GAAG,OAAO,qBAAqB,UAAU,aAAa,EAAE,KAAK,IAAI,GAAG,GAC1E,GACF;CACA,MAAM,UAAU,gBAAgB;CAChC,cAAc,UAAU;CACxB,MAAM,oBAAoB,KAAK,IAAI,IAAI;CACvC,GAAG,OAAO,sBAAsB,kBAAkB,KAAK;CACvD,wBAAwB,iBAAiB;CAEzC,OAAO,KAAK,CAAC,SAAS,EAAC,OAAO,KAAI,CAAC,CAAC;CACpC,OAAO,cAA8B,YAAW,QAAO;EACrD,KAAK,MAAM,KAAK,eACd,EAAE,KAAK,CAAC,WAAW,GAAG,CAAC;CAE3B,CAAC;CAED,IAAI;EACF,MAAM,eACJ,IACA,QACA,IAAI,iBACF,IACA,QACA,QACA,SACA,SACA,cACF,CACF;CACF,SAAS,KAAK;EACZ,UAAU,gBAAgB,KAAK,YAAY;CAC7C,UAAU;EACR,MAAM,UAAU,SAAS;CAC3B;AACF;AAEA,IAAI,CAAC,kBAAkB,GACrB,gBACQ,UACA,UAAU,KAAK,YAAY,GAAG,QAAQ,GAAG,CACjD"}

package/out/zero-cache/src/server/replicator.d.ts

@@ -1,6 +1,10 @@
 import { LogContext } from '@rocicorp/logger';
+import type { NormalizedZeroConfig } from '../config/normalize.ts';
+import { type ReservationFollowup, type SnapshotReserver } from '../services/change-streamer/change-streamer-http.ts';
 import { type Worker } from '../types/processes.ts';
 import { type WalMode } from '../workers/replicator.ts';
 export default function runWorker(parent: Worker, env: NodeJS.ProcessEnv, ...args: string[]): Promise<void>;
 export declare function setupMetrics(lc: LogContext, file: string, walMode: WalMode, pageSize: number): void;
+/** Exported for testing. */
+export declare function restoreReplica(lc: LogContext, config: NormalizedZeroConfig, changeStreamer: SnapshotReserver): Promise<ReservationFollowup>;
 //# sourceMappingURL=replicator.d.ts.map

package/out/zero-cache/src/server/replicator.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"replicator.d.ts","sourceRoot":"","sources":["../../../../../zero-cache/src/server/replicator.ts"],"names":[],"mappings":"AAGA,OAAO,EAAiB,UAAU,EAAC,MAAM,kBAAkB,CAAC;AA+C5D,OAAO,EAGL,KAAK,MAAM,EACZ,MAAM,uBAAuB,CAAC;AAG/B,OAAO,EAML,KAAK,OAAO,EACb,MAAM,0BAA0B,CAAC;AAOlC,wBAA8B,SAAS,CACrC,MAAM,EAAE,MAAM,EACd,GAAG,EAAE,MAAM,CAAC,UAAU,EACtB,GAAG,IAAI,EAAE,MAAM,EAAE,GAChB,OAAO,CAAC,IAAI,CAAC,CAmJf;AAED,wBAAgB,YAAY,CAC1B,EAAE,EAAE,UAAU,EACd,IAAI,EAAE,MAAM,EACZ,OAAO,EAAE,OAAO,EAChB,QAAQ,EAAE,MAAM,QAwDjB"}
+{"version":3,"file":"replicator.d.ts","sourceRoot":"","sources":["../../../../../zero-cache/src/server/replicator.ts"],"names":[],"mappings":"AAGA,OAAO,EAAiB,UAAU,EAAC,MAAM,kBAAkB,CAAC;AAM5D,OAAO,KAAK,EAEV,oBAAoB,EACrB,MAAM,wBAAwB,CAAC;AAKhC,OAAO,EAEL,KAAK,mBAAmB,EACxB,KAAK,gBAAgB,EACtB,MAAM,qDAAqD,CAAC;AA8B7D,OAAO,EAIL,KAAK,MAAM,EACZ,MAAM,uBAAuB,CAAC;AAG/B,OAAO,EAML,KAAK,OAAO,EACb,MAAM,0BAA0B,CAAC;AAOlC,wBAA8B,SAAS,CACrC,MAAM,EAAE,MAAM,EACd,GAAG,EAAE,MAAM,CAAC,UAAU,EACtB,GAAG,IAAI,EAAE,MAAM,EAAE,GAChB,OAAO,CAAC,IAAI,CAAC,CA4Jf;AAED,wBAAgB,YAAY,CAC1B,EAAE,EAAE,UAAU,EACd,IAAI,EAAE,MAAM,EACZ,OAAO,EAAE,OAAO,EAChB,QAAQ,EAAE,MAAM,QAwDjB;AAyFD,4BAA4B;AAC5B,wBAAsB,cAAc,CAClC,EAAE,EAAE,UAAU,EACd,MAAM,EAAE,oBAAoB,EAC5B,cAAc,EAAE,gBAAgB,GAC/B,OAAO,CAAC,mBAAmB,CAAC,CAuF9B"}

package/out/zero-cache/src/server/replicator.js

@@ -14,7 +14,7 @@ import { installProfileHandler } from "../types/profiler.js";
 import { initEventSink } from "../observability/events.js";
 import { ReplicationStatusPublisher } from "../services/replicator/replication-status.js";
 import { litestreamRestoreDuration, litestreamRestoreMetricAttrs, litestreamRestoreRuns } from "../services/litestream/metrics.js";
-import { startReplicaBackupProcess, tryRestore } from "../services/litestream/commands.js";
+import { readLocalReplica, startReplicaBackupProcess, tryRestore } from "../services/litestream/commands.js";
 import { ChangeStreamerHttpClient } from "../services/change-streamer/change-streamer-http.js";
 import { sqliteFileBytes } from "../services/replicator/sqlite-change-log-observability.js";
 import { startOtelAuto } from "./otel-start.js";
@@ -71,7 +71,10 @@ async function runWorker(parent, env, ...args) {
 	const replicator = new ReplicatorService(lc, taskID, `${workerName}-${pid}`, mode, changeStreamer, workerClient, runningLocalChangeStreamer ? ReplicationStatusPublisher.forReplicaFile(dbPath) : null, {
 		readinessThresholdMs: initialReadinessThresholdMs,
 		maxReadinessDelayMs
-	}, initialConnection);
+	}, initialConnection, runningLocalChangeStreamer ? void 0 : {
+		onRestoreGranted: () => parent.send(["drainRequested", {}]),
+		replicaFile: dbPath
+	});
 	setUpMessageHandlers(lc, replicator, parent);
 	replicator.ready().then(() => {
 		parent.send(["ready", { ready: true }]);
@@ -158,14 +161,25 @@ function observeSQLiteFileBytes(lc, file) {
 	return async (o) => o.observe(await sqliteFileBytes(lc, file));
 }
 var RETRY_INTERVAL_MS = 3e3;
+var MAX_OLDER_REPLICA_VERSION_RETRIES = 20;
+/** Exported for testing. */
 async function restoreReplica(lc, config, changeStreamer) {
 	const start = performance.now();
 	let backupURL;
 	let result;
 	const progress = new RestoreProgressReporter(lc, config.replica.file);
+	let olderReplicaVersionRetries = 0;
+	const localReplica = readLocalReplica(lc, config.replica.file);
+	const localReplicaVersion = localReplica?.replicaVersion;
 	try {
 		for (;;) {
-			const { reserved, followup } = await reserveAndGetSnapshotStatus(lc, config.taskID, changeStreamer);
+			const { reserved, followup } = await reserveAndGetSnapshotStatus(lc, config.taskID, changeStreamer, localReplica);
+			if (localReplicaVersion !== void 0 && reserved.replicaVersion < localReplicaVersion && olderReplicaVersionRetries++ < MAX_OLDER_REPLICA_VERSION_RETRIES) {
+				followup.cancel();
+				lc.info?.(`replication-manager is at an older replica version (${reserved.replicaVersion}) than the local replica (${localReplicaVersion}). retrying in ${RETRY_INTERVAL_MS / 1e3} seconds`);
+				await sleep(RETRY_INTERVAL_MS);
+				continue;
+			}
 			({backupURL} = reserved);
 			const litestream = {
 				...config.litestream,
@@ -206,6 +220,6 @@ async function restoreReplica(lc, config, changeStreamer) {
 }
 if (!singleProcessMode()) exitAfter(() => lc, () => runWorker(must(parentWorker), process.env, ...process.argv.slice(2)));
 //#endregion
-export { runWorker as default, setupMetrics };
+export { runWorker as default, restoreReplica, setupMetrics };
 
 //# sourceMappingURL=replicator.js.map

package/out/zero-cache/src/server/replicator.js.map

@@ -1 +1 @@
-{"version":3,"file":"replicator.js","names":[],"sources":["../../../../../zero-cache/src/server/replicator.ts"],"sourcesContent":["import {stat} from 'node:fs/promises';\nimport {pid} from 'node:process';\nimport type {ObservableCallback} from '@opentelemetry/api';\nimport {consoleLogSink, LogContext} from '@rocicorp/logger';\nimport {assert} from '../../../shared/src/asserts.ts';\nimport {must} from '../../../shared/src/must.ts';\nimport {sleep} from '../../../shared/src/sleep.ts';\nimport * as v from '../../../shared/src/valita.ts';\nimport {Database} from '../../../zqlite/src/db.ts';\nimport type {\n  LitestreamConfig,\n  NormalizedZeroConfig,\n} from '../config/normalize.ts';\nimport {getNormalizedZeroConfig} from '../config/zero-config.ts';\nimport {registerSQLiteCorruptionDiagnosticTarget} from '../db/sqlite-corruption.ts';\nimport {initEventSink} from '../observability/events.ts';\nimport {getOrCreateGauge} from '../observability/metrics.ts';\nimport {\n  ChangeStreamerHttpClient,\n  type ReservationFollowup,\n  type SnapshotReserver,\n} from '../services/change-streamer/change-streamer-http.ts';\nimport {reserveAndGetSnapshotStatus} from '../services/change-streamer/snapshot.ts';\nimport {\n  exitAfter,\n  ProcessManager,\n  runUntilKilled,\n} from '../services/life-cycle.ts';\nimport {\n  startReplicaBackupProcess,\n  tryRestore,\n  type RestoreAttempt,\n  type RestoreResult,\n} from '../services/litestream/commands.ts';\nimport {\n  litestreamRestoreDuration,\n  litestreamRestoreMetricAttrs,\n  litestreamRestoreRuns,\n} from '../services/litestream/metrics.ts';\nimport {RestoreProgressReporter} from '../services/litestream/restore-progress.ts';\nimport {ReplicationStatusPublisher} from '../services/replicator/replication-status.ts';\nimport {\n  ReplicatorService,\n  type ReplicatorMode,\n} from '../services/replicator/replicator.ts';\nimport {sqliteFileBytes} from '../services/replicator/sqlite-change-log-observability.ts';\nimport {\n  ThreadWriteWorkerClient,\n  type ForceCheckpointConfig,\n} from '../services/replicator/write-worker-client.ts';\nimport {\n  parentWorker,\n  singleProcessMode,\n  type Worker,\n} from '../types/processes.ts';\nimport {installProfileHandler} from '../types/profiler.ts';\nimport {getShardConfig} from '../types/shards.ts';\nimport {\n  deleteStaleChangeLog,\n  getPragmaConfig,\n  replicaFileModeSchema,\n  setUpMessageHandlers,\n  setupReplica,\n  type WalMode,\n} from '../workers/replicator.ts';\nimport {createLogContext} from './logging.ts';\nimport {startOtelAuto} from './otel-start.ts';\n\n// Default LogContext, overridden in runWorker\nlet lc = new LogContext('info', {}, consoleLogSink);\n\nexport default async function runWorker(\n  parent: Worker,\n  env: NodeJS.ProcessEnv,\n  ...args: string[]\n): Promise<void> {\n  assert(args.length > 0, `replicator mode not specified`);\n  const fileMode = v.parse(args[0], replicaFileModeSchema);\n\n  const config = getNormalizedZeroConfig({env, argv: args.slice(1)});\n  const mode: ReplicatorMode = fileMode === 'backup' ? 'backup' : 'serving';\n  const runningLocalChangeStreamer =\n    config.changeStreamer.mode === 'dedicated' && !config.changeStreamer.uri;\n  const workerName = `${mode}-replicator`;\n  installProfileHandler(parent, workerName);\n  startOtelAuto(createLogContext(config, workerName, 0, false), workerName, 0);\n  lc = createLogContext(config, workerName);\n  const unregisterInitialCorruptionDiagnosticTarget =\n    registerSQLiteCorruptionDiagnosticTarget(\n      {\n        debugName: `${workerName} replica`,\n        dbPath: config.replica.file,\n      },\n      config.sqliteCorruptionChecks,\n    );\n  initEventSink(lc, config);\n\n  const shard = getShardConfig(config);\n  const {\n    taskID,\n    change,\n    changeStreamer: {\n      port,\n      uri: changeStreamerURI = runningLocalChangeStreamer\n        ? `http://localhost:${port}/`\n        : undefined,\n    },\n    replicationLag: {initialReadinessThresholdMs, maxReadinessDelayMs},\n  } = config;\n  const changeStreamer = new ChangeStreamerHttpClient(\n    lc,\n    shard,\n    change.db,\n    changeStreamerURI,\n  );\n\n  // Remote view-syncers restore from the replication-manager, which supplies\n  // the backup URL. A local change-streamer owns the canonical replica and\n  // must not infer backup configuration from bundled executable paths.\n  //\n  // The reservation's connection (`initialConnection`) is threaded through to\n  // IncrementalSyncer's first subscribe(), so the replication-manager that\n  // reserved the change log for this restore is the one that serves the\n  // subscription (single-connection handoff).\n  let initialConnection: ReservationFollowup | undefined;\n  if (\n    fileMode === 'serving' &&\n    !runningLocalChangeStreamer &&\n    (config.litestream.executable || config.litestream.executableV5)\n  ) {\n    initialConnection = await restoreReplica(lc, config, changeStreamer);\n  }\n\n  const {\n    file: dbPath,\n    walMode,\n    pageSize,\n  } = await setupReplica(lc, fileMode, config.replica);\n  unregisterInitialCorruptionDiagnosticTarget();\n  registerSQLiteCorruptionDiagnosticTarget(\n    {\n      debugName: `${workerName} replica`,\n      dbPath,\n    },\n    config.sqliteCorruptionChecks,\n  );\n\n  // A file left behind by a run with the writer enabled would otherwise hold\n  // local disk indefinitely. See `replicatorDeletesStaleChangeLog` for why\n  // this is keyed on the config flag and nothing else.\n  deleteStaleChangeLog(config.changeStreamer.sqliteChangeLogMode, dbPath);\n\n  setupMetrics(lc, dbPath, walMode, pageSize);\n\n  // Create the write worker for async SQLite writes. Enable write-path\n  // checkpoint backpressure only for the backup replicator on litestream v5,\n  // which owns the replica litestream backs up.\n  const pragmas = getPragmaConfig(fileMode);\n  const checkpoint: ForceCheckpointConfig | null =\n    mode === 'backup' &&\n    config.litestream.backupURL &&\n    config.litestream.backupUsingV5 &&\n    config.litestream.forceCheckpointThresholdMB\n      ? {\n          checkpointThresholdPages:\n            (config.litestream.forceCheckpointThresholdMB * 1024 ** 2) /\n            pageSize,\n          maxWalPages: !config.litestream.maxWalSizeMB\n            ? undefined\n            : (config.litestream.maxWalSizeMB * 1024 ** 2) / pageSize,\n        }\n      : null;\n\n  const workerClient = new ThreadWriteWorkerClient();\n  await workerClient.init(dbPath, mode, pragmas, config.log, checkpoint);\n\n  // The backup-replicator manages the litestream backup process. It is\n  // started after the write worker has prepared the db file in order to\n  // avoid contending on the sqlite lock.\n  const processes =\n    mode === 'backup' ? new ProcessManager(lc, parent) : undefined;\n  processes?.addSubprocess(\n    startReplicaBackupProcess(lc, config.litestream, dbPath),\n    'supporting',\n    'litestream',\n  );\n\n  const replicator = new ReplicatorService(\n    lc,\n    taskID,\n    `${workerName}-${pid}`,\n    mode,\n    changeStreamer,\n    workerClient,\n    runningLocalChangeStreamer\n      ? // publish ReplicationStatusEvents from backup-replicator only\n        ReplicationStatusPublisher.forReplicaFile(dbPath)\n      : null,\n    {\n      readinessThresholdMs: initialReadinessThresholdMs,\n      maxReadinessDelayMs,\n    },\n    initialConnection,\n  );\n\n  setUpMessageHandlers(lc, replicator, parent);\n\n  void replicator.ready().then(() => {\n    parent.send(['ready', {ready: true}]);\n  });\n\n  if (!processes) {\n    return runUntilKilled(lc, parent, replicator);\n  }\n\n  try {\n    await runUntilKilled(lc, parent, replicator);\n  } catch (err) {\n    processes.logErrorAndExit(err, workerName);\n  } finally {\n    await processes.shutdown();\n  }\n}\n\nexport function setupMetrics(\n  lc: LogContext,\n  file: string,\n  walMode: WalMode,\n  pageSize: number,\n) {\n  let db: Database | undefined;\n  const getDb = () => {\n    if (!db) {\n      try {\n        db = new Database(lc, file, {readonly: true});\n      } catch (e) {\n        lc.warn?.(`unable to open ${file} for wal metrics`, e);\n      }\n    }\n    return db;\n  };\n\n  getOrCreateGauge('replica', 'db_size', {\n    description:\n      `The size of the replica's main db file, ` +\n      `which does not include the wal file(s)`,\n    unit: 'bytes',\n  }).addCallback(observeFileSize(lc, file));\n\n  getOrCreateGauge('replica', 'wal_size', {\n    description: `The size of the replica's wal file`,\n    unit: 'bytes',\n  }).addCallback(observeFileSize(lc, `${file}-wal`));\n\n  if (walMode === 'wal2') {\n    getOrCreateGauge('replica', 'wal2_size', {\n      description: `The size of the replica's wal2 file`,\n      unit: 'bytes',\n    }).addCallback(observeFileSize(lc, `${file}-wal2`));\n  }\n\n  getOrCreateGauge('replica', 'wal_active_bytes', {\n    description:\n      `The actual size of active WAL frames currently in use ` +\n      `(excluding pre-allocated or reclaimed space).`,\n    unit: 'bytes',\n  }).addCallback(observeWalActiveBytes(lc, getDb, pageSize));\n\n  getOrCreateGauge('replica', 'wal_uncheckpointed_bytes', {\n    description:\n      `The size of un-checkpointed WAL frames that have not yet ` +\n      `been backfilled into the main database.`,\n    unit: 'bytes',\n  }).addCallback(observeWalUncheckpointedBytes(lc, getDb, pageSize));\n\n  // A whole-database total rather than a per-file one. Its counterpart,\n  // `sqlite_change_log.file_bytes`, is emitted by the change-streamer, which is\n  // the process that writes the log.\n  getOrCreateGauge('replica', 'file_bytes', {\n    description:\n      `The replica's total on-disk footprint: its main db file plus its ` +\n      `wal sidecars.`,\n    unit: 'By',\n  }).addCallback(observeSQLiteFileBytes(lc, file));\n}\n\nfunction observeFileSize(lc: LogContext, file: string): ObservableCallback {\n  return async o => {\n    try {\n      const stats = await stat(file);\n      o.observe(stats.size);\n    } catch (e) {\n      lc.warn?.(`unable to stat ${file} for size metrics`, e);\n    }\n  };\n}\n\nfunction observeWalActiveBytes(\n  lc: LogContext,\n  getDb: () => Database | undefined,\n  pageSize: number,\n): ObservableCallback {\n  return o => {\n    try {\n      const db = getDb();\n      if (!db) {\n        return;\n      }\n      const [{log}] = db.pragma<{log: number}>('wal_checkpoint(NOOP)');\n      if (typeof log === 'number') {\n        o.observe(Math.max(0, log) * pageSize);\n      }\n    } catch (e) {\n      lc.warn?.('unable to inspect wal_checkpoint for active bytes metric', e);\n    }\n  };\n}\n\nfunction observeWalUncheckpointedBytes(\n  lc: LogContext,\n  getDb: () => Database | undefined,\n  pageSize: number,\n): ObservableCallback {\n  return o => {\n    try {\n      const db = getDb();\n      if (!db) {\n        return;\n      }\n      const [{log, checkpointed}] = db.pragma<{\n        log: number;\n        checkpointed: number;\n      }>('wal_checkpoint(NOOP)');\n      if (typeof log === 'number' && typeof checkpointed === 'number') {\n        o.observe(Math.max(0, log - checkpointed) * pageSize);\n      }\n    } catch (e) {\n      lc.warn?.(\n        'unable to inspect wal_checkpoint for uncheckpointed bytes metric',\n        e,\n      );\n    }\n  };\n}\n\nfunction observeSQLiteFileBytes(\n  lc: LogContext,\n  file: string,\n): ObservableCallback {\n  return async o => o.observe(await sqliteFileBytes(lc, file));\n}\n\nconst RETRY_INTERVAL_MS = 3000;\n\n// View-syncers wait indefinitely for the replication-manager to publish a\n// restorable backup. On a fresh stack the first backup is not durable until\n// the initial sync completes and litestream uploads the initial snapshot,\n// which can take many minutes for a large replica. The platform's startup\n// probe budget (which scales with replica size) is the backstop, so\n// restoreReplica must not impose its own shorter cap and self-terminate\n// while the backup is still being produced.\n//\n// Returns the `followup` of the reservation whose backup was actually\n// restored, still open, so the caller can hand it to IncrementalSyncer's\n// first subscribe() via `followup.subscribe()`: the same replication-manager\n// that reserved the change log is the one that serves the subscription.\nasync function restoreReplica(\n  lc: LogContext,\n  config: NormalizedZeroConfig,\n  changeStreamer: SnapshotReserver,\n): Promise<ReservationFollowup> {\n  const start = performance.now();\n  let backupURL: string | undefined;\n  let result: RestoreResult | undefined;\n  const progress = new RestoreProgressReporter(lc, config.replica.file);\n  try {\n    for (;;) {\n      const {reserved, followup} = await reserveAndGetSnapshotStatus(\n        lc,\n        config.taskID,\n        changeStreamer,\n      );\n      // The backupURL comes from the replication-manager's snapshot response.\n      ({backupURL} = reserved);\n      const litestream: LitestreamConfig = {...config.litestream, backupURL};\n      progress.start(reserved.replicaSize);\n      let attempt: RestoreAttempt;\n      try {\n        attempt = await tryRestore(\n          lc,\n          litestream,\n          config.replica.file,\n          reserved,\n          'view_syncer',\n          followup.signal, // abort the restore if the reservation disconnects\n        );\n      } catch (e) {\n        if (followup.signal.aborted) {\n          lc.info?.(\n            `reservation disconnected during restore. retrying`,\n            String(e),\n          );\n          continue;\n        }\n        followup.cancel();\n        throw e;\n      } finally {\n        progress.stop();\n      }\n      if (attempt.restored) {\n        result = attempt.result;\n        progress.done();\n        return followup; // connection to be converted to a subscription\n      }\n      // This attempt's reservation is stale; a fresh one is made on retry.\n      followup.cancel();\n      lc.info?.(\n        `replica not found. retrying in ${RETRY_INTERVAL_MS / 1000} seconds`,\n      );\n      await sleep(RETRY_INTERVAL_MS);\n    }\n  } finally {\n    progress.stop();\n    const attrs = litestreamRestoreMetricAttrs(\n      config.litestream,\n      'view_syncer',\n      backupURL,\n    );\n    const labels = {...attrs, result: result ?? 'error'};\n    litestreamRestoreRuns().add(1, labels);\n    litestreamRestoreDuration().recordMs(performance.now() - start, labels);\n  }\n}\n\n// fork()\nif (!singleProcessMode()) {\n  void exitAfter(\n    () => lc,\n    () => runWorker(must(parentWorker), process.env, ...process.argv.slice(2)),\n  );\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAqEA,IAAI,KAAK,IAAI,WAAW,QAAQ,CAAC,GAAG,cAAc;AAElD,eAA8B,UAC5B,QACA,KACA,GAAG,MACY;CACf,OAAO,KAAK,SAAS,GAAG,+BAA+B;CACvD,MAAM,WAAW,MAAQ,KAAK,IAAI,qBAAqB;CAEvD,MAAM,SAAS,wBAAwB;EAAC;EAAK,MAAM,KAAK,MAAM,CAAC;CAAC,CAAC;CACjE,MAAM,OAAuB,aAAa,WAAW,WAAW;CAChE,MAAM,6BACJ,OAAO,eAAe,SAAS,eAAe,CAAC,OAAO,eAAe;CACvE,MAAM,aAAa,GAAG,KAAK;CAC3B,sBAAsB,QAAQ,UAAU;CACxC,cAAc,iBAAiB,QAAQ,YAAY,GAAG,KAAK,GAAG,YAAY,CAAC;CAC3E,KAAK,iBAAiB,QAAQ,UAAU;CACxC,MAAM,8CACJ,yCACE;EACE,WAAW,GAAG,WAAW;EACzB,QAAQ,OAAO,QAAQ;CACzB,GACA,OAAO,sBACT;CACF,cAAc,IAAI,MAAM;CAExB,MAAM,QAAQ,eAAe,MAAM;CACnC,MAAM,EACJ,QACA,QACA,gBAAgB,EACd,MACA,KAAK,oBAAoB,6BACrB,oBAAoB,KAAK,KACzB,KAAA,KAEN,gBAAgB,EAAC,6BAA6B,0BAC5C;CACJ,MAAM,iBAAiB,IAAI,yBACzB,IACA,OACA,OAAO,IACP,iBACF;CAUA,IAAI;CACJ,IACE,aAAa,aACb,CAAC,+BACA,OAAO,WAAW,cAAc,OAAO,WAAW,eAEnD,oBAAoB,MAAM,eAAe,IAAI,QAAQ,cAAc;CAGrE,MAAM,EACJ,MAAM,QACN,SACA,aACE,MAAM,aAAa,IAAI,UAAU,OAAO,OAAO;CACnD,4CAA4C;CAC5C,yCACE;EACE,WAAW,GAAG,WAAW;EACzB;CACF,GACA,OAAO,sBACT;CAKA,qBAAqB,OAAO,eAAe,qBAAqB,MAAM;CAEtE,aAAa,IAAI,QAAQ,SAAS,QAAQ;CAK1C,MAAM,UAAU,gBAAgB,QAAQ;CACxC,MAAM,aACJ,SAAS,YACT,OAAO,WAAW,aAClB,OAAO,WAAW,iBAClB,OAAO,WAAW,6BACd;EACE,0BACG,OAAO,WAAW,6BAA6B,QAAQ,IACxD;EACF,aAAa,CAAC,OAAO,WAAW,eAC5B,KAAA,IACC,OAAO,WAAW,eAAe,QAAQ,IAAK;CACrD,IACA;CAEN,MAAM,eAAe,IAAI,wBAAwB;CACjD,MAAM,aAAa,KAAK,QAAQ,MAAM,SAAS,OAAO,KAAK,UAAU;CAKrE,MAAM,YACJ,SAAS,WAAW,IAAI,eAAe,IAAI,MAAM,IAAI,KAAA;CACvD,WAAW,cACT,0BAA0B,IAAI,OAAO,YAAY,MAAM,GACvD,cACA,YACF;CAEA,MAAM,aAAa,IAAI,kBACrB,IACA,QACA,GAAG,WAAW,GAAG,OACjB,MACA,gBACA,cACA,6BAEI,2BAA2B,eAAe,MAAM,IAChD,MACJ;EACE,sBAAsB;EACtB;CACF,GACA,iBACF;CAEA,qBAAqB,IAAI,YAAY,MAAM;CAE3C,WAAgB,MAAM,EAAE,WAAW;EACjC,OAAO,KAAK,CAAC,SAAS,EAAC,OAAO,KAAI,CAAC,CAAC;CACtC,CAAC;CAED,IAAI,CAAC,WACH,OAAO,eAAe,IAAI,QAAQ,UAAU;CAG9C,IAAI;EACF,MAAM,eAAe,IAAI,QAAQ,UAAU;CAC7C,SAAS,KAAK;EACZ,UAAU,gBAAgB,KAAK,UAAU;CAC3C,UAAU;EACR,MAAM,UAAU,SAAS;CAC3B;AACF;AAEA,SAAgB,aACd,IACA,MACA,SACA,UACA;CACA,IAAI;CACJ,MAAM,cAAc;EAClB,IAAI,CAAC,IACH,IAAI;GACF,KAAK,IAAI,SAAS,IAAI,MAAM,EAAC,UAAU,KAAI,CAAC;EAC9C,SAAS,GAAG;GACV,GAAG,OAAO,kBAAkB,KAAK,mBAAmB,CAAC;EACvD;EAEF,OAAO;CACT;CAEA,iBAAiB,WAAW,WAAW;EACrC,aACE;EAEF,MAAM;CACR,CAAC,EAAE,YAAY,gBAAgB,IAAI,IAAI,CAAC;CAExC,iBAAiB,WAAW,YAAY;EACtC,aAAa;EACb,MAAM;CACR,CAAC,EAAE,YAAY,gBAAgB,IAAI,GAAG,KAAK,KAAK,CAAC;CAEjD,IAAI,YAAY,QACd,iBAAiB,WAAW,aAAa;EACvC,aAAa;EACb,MAAM;CACR,CAAC,EAAE,YAAY,gBAAgB,IAAI,GAAG,KAAK,MAAM,CAAC;CAGpD,iBAAiB,WAAW,oBAAoB;EAC9C,aACE;EAEF,MAAM;CACR,CAAC,EAAE,YAAY,sBAAsB,IAAI,OAAO,QAAQ,CAAC;CAEzD,iBAAiB,WAAW,4BAA4B;EACtD,aACE;EAEF,MAAM;CACR,CAAC,EAAE,YAAY,8BAA8B,IAAI,OAAO,QAAQ,CAAC;CAKjE,iBAAiB,WAAW,cAAc;EACxC,aACE;EAEF,MAAM;CACR,CAAC,EAAE,YAAY,uBAAuB,IAAI,IAAI,CAAC;AACjD;AAEA,SAAS,gBAAgB,IAAgB,MAAkC;CACzE,OAAO,OAAM,MAAK;EAChB,IAAI;GACF,MAAM,QAAQ,MAAM,KAAK,IAAI;GAC7B,EAAE,QAAQ,MAAM,IAAI;EACtB,SAAS,GAAG;GACV,GAAG,OAAO,kBAAkB,KAAK,oBAAoB,CAAC;EACxD;CACF;AACF;AAEA,SAAS,sBACP,IACA,OACA,UACoB;CACpB,QAAO,MAAK;EACV,IAAI;GACF,MAAM,KAAK,MAAM;GACjB,IAAI,CAAC,IACH;GAEF,MAAM,CAAC,EAAC,SAAQ,GAAG,OAAsB,sBAAsB;GAC/D,IAAI,OAAO,QAAQ,UACjB,EAAE,QAAQ,KAAK,IAAI,GAAG,GAAG,IAAI,QAAQ;EAEzC,SAAS,GAAG;GACV,GAAG,OAAO,4DAA4D,CAAC;EACzE;CACF;AACF;AAEA,SAAS,8BACP,IACA,OACA,UACoB;CACpB,QAAO,MAAK;EACV,IAAI;GACF,MAAM,KAAK,MAAM;GACjB,IAAI,CAAC,IACH;GAEF,MAAM,CAAC,EAAC,KAAK,kBAAiB,GAAG,OAG9B,sBAAsB;GACzB,IAAI,OAAO,QAAQ,YAAY,OAAO,iBAAiB,UACrD,EAAE,QAAQ,KAAK,IAAI,GAAG,MAAM,YAAY,IAAI,QAAQ;EAExD,SAAS,GAAG;GACV,GAAG,OACD,oEACA,CACF;EACF;CACF;AACF;AAEA,SAAS,uBACP,IACA,MACoB;CACpB,OAAO,OAAM,MAAK,EAAE,QAAQ,MAAM,gBAAgB,IAAI,IAAI,CAAC;AAC7D;AAEA,IAAM,oBAAoB;AAc1B,eAAe,eACb,IACA,QACA,gBAC8B;CAC9B,MAAM,QAAQ,YAAY,IAAI;CAC9B,IAAI;CACJ,IAAI;CACJ,MAAM,WAAW,IAAI,wBAAwB,IAAI,OAAO,QAAQ,IAAI;CACpE,IAAI;EACF,SAAS;GACP,MAAM,EAAC,UAAU,aAAY,MAAM,4BACjC,IACA,OAAO,QACP,cACF;GAEA,CAAC,CAAC,aAAa;GACf,MAAM,aAA+B;IAAC,GAAG,OAAO;IAAY;GAAS;GACrE,SAAS,MAAM,SAAS,WAAW;GACnC,IAAI;GACJ,IAAI;IACF,UAAU,MAAM,WACd,IACA,YACA,OAAO,QAAQ,MACf,UACA,eACA,SAAS,MACX;GACF,SAAS,GAAG;IACV,IAAI,SAAS,OAAO,SAAS;KAC3B,GAAG,OACD,qDACA,OAAO,CAAC,CACV;KACA;IACF;IACA,SAAS,OAAO;IAChB,MAAM;GACR,UAAU;IACR,SAAS,KAAK;GAChB;GACA,IAAI,QAAQ,UAAU;IACpB,SAAS,QAAQ;IACjB,SAAS,KAAK;IACd,OAAO;GACT;GAEA,SAAS,OAAO;GAChB,GAAG,OACD,kCAAkC,oBAAoB,IAAK,SAC7D;GACA,MAAM,MAAM,iBAAiB;EAC/B;CACF,UAAU;EACR,SAAS,KAAK;EAMd,MAAM,SAAS;GAAC,GALF,6BACZ,OAAO,YACP,eACA,SAEiB;GAAO,QAAQ,UAAU;EAAO;EACnD,sBAAsB,EAAE,IAAI,GAAG,MAAM;EACrC,0BAA0B,EAAE,SAAS,YAAY,IAAI,IAAI,OAAO,MAAM;CACxE;AACF;AAGA,IAAI,CAAC,kBAAkB,GACrB,gBACQ,UACA,UAAU,KAAK,YAAY,GAAG,QAAQ,KAAK,GAAG,QAAQ,KAAK,MAAM,CAAC,CAAC,CAC3E"}
+{"version":3,"file":"replicator.js","names":[],"sources":["../../../../../zero-cache/src/server/replicator.ts"],"sourcesContent":["import {stat} from 'node:fs/promises';\nimport {pid} from 'node:process';\nimport type {ObservableCallback} from '@opentelemetry/api';\nimport {consoleLogSink, LogContext} from '@rocicorp/logger';\nimport {assert} from '../../../shared/src/asserts.ts';\nimport {must} from '../../../shared/src/must.ts';\nimport {sleep} from '../../../shared/src/sleep.ts';\nimport * as v from '../../../shared/src/valita.ts';\nimport {Database} from '../../../zqlite/src/db.ts';\nimport type {\n  LitestreamConfig,\n  NormalizedZeroConfig,\n} from '../config/normalize.ts';\nimport {getNormalizedZeroConfig} from '../config/zero-config.ts';\nimport {registerSQLiteCorruptionDiagnosticTarget} from '../db/sqlite-corruption.ts';\nimport {initEventSink} from '../observability/events.ts';\nimport {getOrCreateGauge} from '../observability/metrics.ts';\nimport {\n  ChangeStreamerHttpClient,\n  type ReservationFollowup,\n  type SnapshotReserver,\n} from '../services/change-streamer/change-streamer-http.ts';\nimport {reserveAndGetSnapshotStatus} from '../services/change-streamer/snapshot.ts';\nimport {\n  exitAfter,\n  ProcessManager,\n  runUntilKilled,\n} from '../services/life-cycle.ts';\nimport {\n  readLocalReplica,\n  startReplicaBackupProcess,\n  tryRestore,\n  type RestoreAttempt,\n  type RestoreResult,\n} from '../services/litestream/commands.ts';\nimport {\n  litestreamRestoreDuration,\n  litestreamRestoreMetricAttrs,\n  litestreamRestoreRuns,\n} from '../services/litestream/metrics.ts';\nimport {RestoreProgressReporter} from '../services/litestream/restore-progress.ts';\nimport {ReplicationStatusPublisher} from '../services/replicator/replication-status.ts';\nimport {\n  ReplicatorService,\n  type ReplicatorMode,\n} from '../services/replicator/replicator.ts';\nimport {sqliteFileBytes} from '../services/replicator/sqlite-change-log-observability.ts';\nimport {\n  ThreadWriteWorkerClient,\n  type ForceCheckpointConfig,\n} from '../services/replicator/write-worker-client.ts';\nimport {\n  parentWorker,\n  singleProcessMode,\n  type DrainRequestedMessage,\n  type Worker,\n} from '../types/processes.ts';\nimport {installProfileHandler} from '../types/profiler.ts';\nimport {getShardConfig} from '../types/shards.ts';\nimport {\n  deleteStaleChangeLog,\n  getPragmaConfig,\n  replicaFileModeSchema,\n  setUpMessageHandlers,\n  setupReplica,\n  type WalMode,\n} from '../workers/replicator.ts';\nimport {createLogContext} from './logging.ts';\nimport {startOtelAuto} from './otel-start.ts';\n\n// Default LogContext, overridden in runWorker\nlet lc = new LogContext('info', {}, consoleLogSink);\n\nexport default async function runWorker(\n  parent: Worker,\n  env: NodeJS.ProcessEnv,\n  ...args: string[]\n): Promise<void> {\n  assert(args.length > 0, `replicator mode not specified`);\n  const fileMode = v.parse(args[0], replicaFileModeSchema);\n\n  const config = getNormalizedZeroConfig({env, argv: args.slice(1)});\n  const mode: ReplicatorMode = fileMode === 'backup' ? 'backup' : 'serving';\n  const runningLocalChangeStreamer =\n    config.changeStreamer.mode === 'dedicated' && !config.changeStreamer.uri;\n  const workerName = `${mode}-replicator`;\n  installProfileHandler(parent, workerName);\n  startOtelAuto(createLogContext(config, workerName, 0, false), workerName, 0);\n  lc = createLogContext(config, workerName);\n  const unregisterInitialCorruptionDiagnosticTarget =\n    registerSQLiteCorruptionDiagnosticTarget(\n      {\n        debugName: `${workerName} replica`,\n        dbPath: config.replica.file,\n      },\n      config.sqliteCorruptionChecks,\n    );\n  initEventSink(lc, config);\n\n  const shard = getShardConfig(config);\n  const {\n    taskID,\n    change,\n    changeStreamer: {\n      port,\n      uri: changeStreamerURI = runningLocalChangeStreamer\n        ? `http://localhost:${port}/`\n        : undefined,\n    },\n    replicationLag: {initialReadinessThresholdMs, maxReadinessDelayMs},\n  } = config;\n  const changeStreamer = new ChangeStreamerHttpClient(\n    lc,\n    shard,\n    change.db,\n    changeStreamerURI,\n  );\n\n  // Remote view-syncers restore from the replication-manager, which supplies\n  // the backup URL. A local change-streamer owns the canonical replica and\n  // must not infer backup configuration from bundled executable paths.\n  //\n  // The reservation's connection (`initialConnection`) is threaded through to\n  // IncrementalSyncer's first subscribe(), so the replication-manager that\n  // reserved the change log for this restore is the one that serves the\n  // subscription (single-connection handoff).\n  let initialConnection: ReservationFollowup | undefined;\n  if (\n    fileMode === 'serving' &&\n    !runningLocalChangeStreamer &&\n    (config.litestream.executable || config.litestream.executableV5)\n  ) {\n    initialConnection = await restoreReplica(lc, config, changeStreamer);\n  }\n\n  const {\n    file: dbPath,\n    walMode,\n    pageSize,\n  } = await setupReplica(lc, fileMode, config.replica);\n  unregisterInitialCorruptionDiagnosticTarget();\n  registerSQLiteCorruptionDiagnosticTarget(\n    {\n      debugName: `${workerName} replica`,\n      dbPath,\n    },\n    config.sqliteCorruptionChecks,\n  );\n\n  // A file left behind by a run with the writer enabled would otherwise hold\n  // local disk indefinitely. See `replicatorDeletesStaleChangeLog` for why\n  // this is keyed on the config flag and nothing else.\n  deleteStaleChangeLog(config.changeStreamer.sqliteChangeLogMode, dbPath);\n\n  setupMetrics(lc, dbPath, walMode, pageSize);\n\n  // Create the write worker for async SQLite writes. Enable write-path\n  // checkpoint backpressure only for the backup replicator on litestream v5,\n  // which owns the replica litestream backs up.\n  const pragmas = getPragmaConfig(fileMode);\n  const checkpoint: ForceCheckpointConfig | null =\n    mode === 'backup' &&\n    config.litestream.backupURL &&\n    config.litestream.backupUsingV5 &&\n    config.litestream.forceCheckpointThresholdMB\n      ? {\n          checkpointThresholdPages:\n            (config.litestream.forceCheckpointThresholdMB * 1024 ** 2) /\n            pageSize,\n          maxWalPages: !config.litestream.maxWalSizeMB\n            ? undefined\n            : (config.litestream.maxWalSizeMB * 1024 ** 2) / pageSize,\n        }\n      : null;\n\n  const workerClient = new ThreadWriteWorkerClient();\n  await workerClient.init(dbPath, mode, pragmas, config.log, checkpoint);\n\n  // The backup-replicator manages the litestream backup process. It is\n  // started after the write worker has prepared the db file in order to\n  // avoid contending on the sqlite lock.\n  const processes =\n    mode === 'backup' ? new ProcessManager(lc, parent) : undefined;\n  processes?.addSubprocess(\n    startReplicaBackupProcess(lc, config.litestream, dbPath),\n    'supporting',\n    'litestream',\n  );\n\n  const replicator = new ReplicatorService(\n    lc,\n    taskID,\n    `${workerName}-${pid}`,\n    mode,\n    changeStreamer,\n    workerClient,\n    runningLocalChangeStreamer\n      ? // publish ReplicationStatusEvents from backup-replicator only\n        ReplicationStatusPublisher.forReplicaFile(dbPath)\n      : null,\n    {\n      readinessThresholdMs: initialReadinessThresholdMs,\n      maxReadinessDelayMs,\n    },\n    initialConnection,\n    // View-syncers (i.e. with a remote change-streamer) switch over to\n    // another replication-manager when theirs drains.\n    runningLocalChangeStreamer\n      ? undefined\n      : {\n          onRestoreGranted: () =>\n            parent.send<DrainRequestedMessage>(['drainRequested', {}]),\n          replicaFile: dbPath,\n        },\n  );\n\n  setUpMessageHandlers(lc, replicator, parent);\n\n  void replicator.ready().then(() => {\n    parent.send(['ready', {ready: true}]);\n  });\n\n  if (!processes) {\n    return runUntilKilled(lc, parent, replicator);\n  }\n\n  try {\n    await runUntilKilled(lc, parent, replicator);\n  } catch (err) {\n    processes.logErrorAndExit(err, workerName);\n  } finally {\n    await processes.shutdown();\n  }\n}\n\nexport function setupMetrics(\n  lc: LogContext,\n  file: string,\n  walMode: WalMode,\n  pageSize: number,\n) {\n  let db: Database | undefined;\n  const getDb = () => {\n    if (!db) {\n      try {\n        db = new Database(lc, file, {readonly: true});\n      } catch (e) {\n        lc.warn?.(`unable to open ${file} for wal metrics`, e);\n      }\n    }\n    return db;\n  };\n\n  getOrCreateGauge('replica', 'db_size', {\n    description:\n      `The size of the replica's main db file, ` +\n      `which does not include the wal file(s)`,\n    unit: 'bytes',\n  }).addCallback(observeFileSize(lc, file));\n\n  getOrCreateGauge('replica', 'wal_size', {\n    description: `The size of the replica's wal file`,\n    unit: 'bytes',\n  }).addCallback(observeFileSize(lc, `${file}-wal`));\n\n  if (walMode === 'wal2') {\n    getOrCreateGauge('replica', 'wal2_size', {\n      description: `The size of the replica's wal2 file`,\n      unit: 'bytes',\n    }).addCallback(observeFileSize(lc, `${file}-wal2`));\n  }\n\n  getOrCreateGauge('replica', 'wal_active_bytes', {\n    description:\n      `The actual size of active WAL frames currently in use ` +\n      `(excluding pre-allocated or reclaimed space).`,\n    unit: 'bytes',\n  }).addCallback(observeWalActiveBytes(lc, getDb, pageSize));\n\n  getOrCreateGauge('replica', 'wal_uncheckpointed_bytes', {\n    description:\n      `The size of un-checkpointed WAL frames that have not yet ` +\n      `been backfilled into the main database.`,\n    unit: 'bytes',\n  }).addCallback(observeWalUncheckpointedBytes(lc, getDb, pageSize));\n\n  // A whole-database total rather than a per-file one. Its counterpart,\n  // `sqlite_change_log.file_bytes`, is emitted by the change-streamer, which is\n  // the process that writes the log.\n  getOrCreateGauge('replica', 'file_bytes', {\n    description:\n      `The replica's total on-disk footprint: its main db file plus its ` +\n      `wal sidecars.`,\n    unit: 'By',\n  }).addCallback(observeSQLiteFileBytes(lc, file));\n}\n\nfunction observeFileSize(lc: LogContext, file: string): ObservableCallback {\n  return async o => {\n    try {\n      const stats = await stat(file);\n      o.observe(stats.size);\n    } catch (e) {\n      lc.warn?.(`unable to stat ${file} for size metrics`, e);\n    }\n  };\n}\n\nfunction observeWalActiveBytes(\n  lc: LogContext,\n  getDb: () => Database | undefined,\n  pageSize: number,\n): ObservableCallback {\n  return o => {\n    try {\n      const db = getDb();\n      if (!db) {\n        return;\n      }\n      const [{log}] = db.pragma<{log: number}>('wal_checkpoint(NOOP)');\n      if (typeof log === 'number') {\n        o.observe(Math.max(0, log) * pageSize);\n      }\n    } catch (e) {\n      lc.warn?.('unable to inspect wal_checkpoint for active bytes metric', e);\n    }\n  };\n}\n\nfunction observeWalUncheckpointedBytes(\n  lc: LogContext,\n  getDb: () => Database | undefined,\n  pageSize: number,\n): ObservableCallback {\n  return o => {\n    try {\n      const db = getDb();\n      if (!db) {\n        return;\n      }\n      const [{log, checkpointed}] = db.pragma<{\n        log: number;\n        checkpointed: number;\n      }>('wal_checkpoint(NOOP)');\n      if (typeof log === 'number' && typeof checkpointed === 'number') {\n        o.observe(Math.max(0, log - checkpointed) * pageSize);\n      }\n    } catch (e) {\n      lc.warn?.(\n        'unable to inspect wal_checkpoint for uncheckpointed bytes metric',\n        e,\n      );\n    }\n  };\n}\n\nfunction observeSQLiteFileBytes(\n  lc: LogContext,\n  file: string,\n): ObservableCallback {\n  return async o => o.observe(await sqliteFileBytes(lc, file));\n}\n\nconst RETRY_INTERVAL_MS = 3000;\n\n// The number of times a view-syncer retries a snapshot reservation from a\n// replication-manager with an older replica version than its local replica\n// (e.g. a not-yet-terminated replication-manager during a resync rollout),\n// before accepting the older version (e.g. if the upstream was legitimately\n// replaced by one with a lower replica version).\nconst MAX_OLDER_REPLICA_VERSION_RETRIES = 20;\n\n// View-syncers wait indefinitely for the replication-manager to publish a\n// restorable backup. On a fresh stack the first backup is not durable until\n// the initial sync completes and litestream uploads the initial snapshot,\n// which can take many minutes for a large replica. The platform's startup\n// probe budget (which scales with replica size) is the backstop, so\n// restoreReplica must not impose its own shorter cap and self-terminate\n// while the backup is still being produced.\n//\n// Returns the `followup` of the reservation whose backup was actually\n// restored, still open, so the caller can hand it to IncrementalSyncer's\n// first subscribe() via `followup.subscribe()`: the same replication-manager\n// that reserved the change log is the one that serves the subscription.\n/** Exported for testing. */\nexport async function restoreReplica(\n  lc: LogContext,\n  config: NormalizedZeroConfig,\n  changeStreamer: SnapshotReserver,\n): Promise<ReservationFollowup> {\n  const start = performance.now();\n  let backupURL: string | undefined;\n  let result: RestoreResult | undefined;\n  const progress = new RestoreProgressReporter(lc, config.replica.file);\n  let olderReplicaVersionRetries = 0;\n  // Describing the local replica (if any) allows the replication-manager to\n  // account for replicas that are being replaced (see RestoreAdmission). It\n  // is read once, before the first attempt, as a failed attempt may have\n  // already deleted it; it continues to describe the replica being replaced\n  // (and the version not to downgrade from) on every retry.\n  const localReplica = readLocalReplica(lc, config.replica.file);\n  const localReplicaVersion = localReplica?.replicaVersion;\n  try {\n    for (;;) {\n      const {reserved, followup} = await reserveAndGetSnapshotStatus(\n        lc,\n        config.taskID,\n        changeStreamer,\n        localReplica,\n      );\n      if (\n        localReplicaVersion !== undefined &&\n        reserved.replicaVersion < localReplicaVersion &&\n        olderReplicaVersionRetries++ < MAX_OLDER_REPLICA_VERSION_RETRIES\n      ) {\n        // Don't replace a newer local replica with an older one, e.g. from\n        // a replication-manager that has yet to be replaced by a resync.\n        followup.cancel();\n        lc.info?.(\n          `replication-manager is at an older replica version ` +\n            `(${reserved.replicaVersion}) than the local replica ` +\n            `(${localReplicaVersion}). retrying in ${RETRY_INTERVAL_MS / 1000} seconds`,\n        );\n        await sleep(RETRY_INTERVAL_MS);\n        continue;\n      }\n      // The backupURL comes from the replication-manager's snapshot response.\n      ({backupURL} = reserved);\n      const litestream: LitestreamConfig = {...config.litestream, backupURL};\n      progress.start(reserved.replicaSize);\n      let attempt: RestoreAttempt;\n      try {\n        attempt = await tryRestore(\n          lc,\n          litestream,\n          config.replica.file,\n          reserved,\n          'view_syncer',\n          followup.signal, // abort the restore if the reservation disconnects\n        );\n      } catch (e) {\n        if (followup.signal.aborted) {\n          lc.info?.(\n            `reservation disconnected during restore. retrying`,\n            String(e),\n          );\n          continue;\n        }\n        followup.cancel();\n        throw e;\n      } finally {\n        progress.stop();\n      }\n      if (attempt.restored) {\n        result = attempt.result;\n        progress.done();\n        return followup; // connection to be converted to a subscription\n      }\n      // This attempt's reservation is stale; a fresh one is made on retry.\n      followup.cancel();\n      lc.info?.(\n        `replica not found. retrying in ${RETRY_INTERVAL_MS / 1000} seconds`,\n      );\n      await sleep(RETRY_INTERVAL_MS);\n    }\n  } finally {\n    progress.stop();\n    const attrs = litestreamRestoreMetricAttrs(\n      config.litestream,\n      'view_syncer',\n      backupURL,\n    );\n    const labels = {...attrs, result: result ?? 'error'};\n    litestreamRestoreRuns().add(1, labels);\n    litestreamRestoreDuration().recordMs(performance.now() - start, labels);\n  }\n}\n\n// fork()\nif (!singleProcessMode()) {\n  void exitAfter(\n    () => lc,\n    () => runWorker(must(parentWorker), process.env, ...process.argv.slice(2)),\n  );\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAuEA,IAAI,KAAK,IAAI,WAAW,QAAQ,CAAC,GAAG,cAAc;AAElD,eAA8B,UAC5B,QACA,KACA,GAAG,MACY;CACf,OAAO,KAAK,SAAS,GAAG,+BAA+B;CACvD,MAAM,WAAW,MAAQ,KAAK,IAAI,qBAAqB;CAEvD,MAAM,SAAS,wBAAwB;EAAC;EAAK,MAAM,KAAK,MAAM,CAAC;CAAC,CAAC;CACjE,MAAM,OAAuB,aAAa,WAAW,WAAW;CAChE,MAAM,6BACJ,OAAO,eAAe,SAAS,eAAe,CAAC,OAAO,eAAe;CACvE,MAAM,aAAa,GAAG,KAAK;CAC3B,sBAAsB,QAAQ,UAAU;CACxC,cAAc,iBAAiB,QAAQ,YAAY,GAAG,KAAK,GAAG,YAAY,CAAC;CAC3E,KAAK,iBAAiB,QAAQ,UAAU;CACxC,MAAM,8CACJ,yCACE;EACE,WAAW,GAAG,WAAW;EACzB,QAAQ,OAAO,QAAQ;CACzB,GACA,OAAO,sBACT;CACF,cAAc,IAAI,MAAM;CAExB,MAAM,QAAQ,eAAe,MAAM;CACnC,MAAM,EACJ,QACA,QACA,gBAAgB,EACd,MACA,KAAK,oBAAoB,6BACrB,oBAAoB,KAAK,KACzB,KAAA,KAEN,gBAAgB,EAAC,6BAA6B,0BAC5C;CACJ,MAAM,iBAAiB,IAAI,yBACzB,IACA,OACA,OAAO,IACP,iBACF;CAUA,IAAI;CACJ,IACE,aAAa,aACb,CAAC,+BACA,OAAO,WAAW,cAAc,OAAO,WAAW,eAEnD,oBAAoB,MAAM,eAAe,IAAI,QAAQ,cAAc;CAGrE,MAAM,EACJ,MAAM,QACN,SACA,aACE,MAAM,aAAa,IAAI,UAAU,OAAO,OAAO;CACnD,4CAA4C;CAC5C,yCACE;EACE,WAAW,GAAG,WAAW;EACzB;CACF,GACA,OAAO,sBACT;CAKA,qBAAqB,OAAO,eAAe,qBAAqB,MAAM;CAEtE,aAAa,IAAI,QAAQ,SAAS,QAAQ;CAK1C,MAAM,UAAU,gBAAgB,QAAQ;CACxC,MAAM,aACJ,SAAS,YACT,OAAO,WAAW,aAClB,OAAO,WAAW,iBAClB,OAAO,WAAW,6BACd;EACE,0BACG,OAAO,WAAW,6BAA6B,QAAQ,IACxD;EACF,aAAa,CAAC,OAAO,WAAW,eAC5B,KAAA,IACC,OAAO,WAAW,eAAe,QAAQ,IAAK;CACrD,IACA;CAEN,MAAM,eAAe,IAAI,wBAAwB;CACjD,MAAM,aAAa,KAAK,QAAQ,MAAM,SAAS,OAAO,KAAK,UAAU;CAKrE,MAAM,YACJ,SAAS,WAAW,IAAI,eAAe,IAAI,MAAM,IAAI,KAAA;CACvD,WAAW,cACT,0BAA0B,IAAI,OAAO,YAAY,MAAM,GACvD,cACA,YACF;CAEA,MAAM,aAAa,IAAI,kBACrB,IACA,QACA,GAAG,WAAW,GAAG,OACjB,MACA,gBACA,cACA,6BAEI,2BAA2B,eAAe,MAAM,IAChD,MACJ;EACE,sBAAsB;EACtB;CACF,GACA,mBAGA,6BACI,KAAA,IACA;EACE,wBACE,OAAO,KAA4B,CAAC,kBAAkB,CAAC,CAAC,CAAC;EAC3D,aAAa;CACf,CACN;CAEA,qBAAqB,IAAI,YAAY,MAAM;CAE3C,WAAgB,MAAM,EAAE,WAAW;EACjC,OAAO,KAAK,CAAC,SAAS,EAAC,OAAO,KAAI,CAAC,CAAC;CACtC,CAAC;CAED,IAAI,CAAC,WACH,OAAO,eAAe,IAAI,QAAQ,UAAU;CAG9C,IAAI;EACF,MAAM,eAAe,IAAI,QAAQ,UAAU;CAC7C,SAAS,KAAK;EACZ,UAAU,gBAAgB,KAAK,UAAU;CAC3C,UAAU;EACR,MAAM,UAAU,SAAS;CAC3B;AACF;AAEA,SAAgB,aACd,IACA,MACA,SACA,UACA;CACA,IAAI;CACJ,MAAM,cAAc;EAClB,IAAI,CAAC,IACH,IAAI;GACF,KAAK,IAAI,SAAS,IAAI,MAAM,EAAC,UAAU,KAAI,CAAC;EAC9C,SAAS,GAAG;GACV,GAAG,OAAO,kBAAkB,KAAK,mBAAmB,CAAC;EACvD;EAEF,OAAO;CACT;CAEA,iBAAiB,WAAW,WAAW;EACrC,aACE;EAEF,MAAM;CACR,CAAC,EAAE,YAAY,gBAAgB,IAAI,IAAI,CAAC;CAExC,iBAAiB,WAAW,YAAY;EACtC,aAAa;EACb,MAAM;CACR,CAAC,EAAE,YAAY,gBAAgB,IAAI,GAAG,KAAK,KAAK,CAAC;CAEjD,IAAI,YAAY,QACd,iBAAiB,WAAW,aAAa;EACvC,aAAa;EACb,MAAM;CACR,CAAC,EAAE,YAAY,gBAAgB,IAAI,GAAG,KAAK,MAAM,CAAC;CAGpD,iBAAiB,WAAW,oBAAoB;EAC9C,aACE;EAEF,MAAM;CACR,CAAC,EAAE,YAAY,sBAAsB,IAAI,OAAO,QAAQ,CAAC;CAEzD,iBAAiB,WAAW,4BAA4B;EACtD,aACE;EAEF,MAAM;CACR,CAAC,EAAE,YAAY,8BAA8B,IAAI,OAAO,QAAQ,CAAC;CAKjE,iBAAiB,WAAW,cAAc;EACxC,aACE;EAEF,MAAM;CACR,CAAC,EAAE,YAAY,uBAAuB,IAAI,IAAI,CAAC;AACjD;AAEA,SAAS,gBAAgB,IAAgB,MAAkC;CACzE,OAAO,OAAM,MAAK;EAChB,IAAI;GACF,MAAM,QAAQ,MAAM,KAAK,IAAI;GAC7B,EAAE,QAAQ,MAAM,IAAI;EACtB,SAAS,GAAG;GACV,GAAG,OAAO,kBAAkB,KAAK,oBAAoB,CAAC;EACxD;CACF;AACF;AAEA,SAAS,sBACP,IACA,OACA,UACoB;CACpB,QAAO,MAAK;EACV,IAAI;GACF,MAAM,KAAK,MAAM;GACjB,IAAI,CAAC,IACH;GAEF,MAAM,CAAC,EAAC,SAAQ,GAAG,OAAsB,sBAAsB;GAC/D,IAAI,OAAO,QAAQ,UACjB,EAAE,QAAQ,KAAK,IAAI,GAAG,GAAG,IAAI,QAAQ;EAEzC,SAAS,GAAG;GACV,GAAG,OAAO,4DAA4D,CAAC;EACzE;CACF;AACF;AAEA,SAAS,8BACP,IACA,OACA,UACoB;CACpB,QAAO,MAAK;EACV,IAAI;GACF,MAAM,KAAK,MAAM;GACjB,IAAI,CAAC,IACH;GAEF,MAAM,CAAC,EAAC,KAAK,kBAAiB,GAAG,OAG9B,sBAAsB;GACzB,IAAI,OAAO,QAAQ,YAAY,OAAO,iBAAiB,UACrD,EAAE,QAAQ,KAAK,IAAI,GAAG,MAAM,YAAY,IAAI,QAAQ;EAExD,SAAS,GAAG;GACV,GAAG,OACD,oEACA,CACF;EACF;CACF;AACF;AAEA,SAAS,uBACP,IACA,MACoB;CACpB,OAAO,OAAM,MAAK,EAAE,QAAQ,MAAM,gBAAgB,IAAI,IAAI,CAAC;AAC7D;AAEA,IAAM,oBAAoB;AAO1B,IAAM,oCAAoC;;AAe1C,eAAsB,eACpB,IACA,QACA,gBAC8B;CAC9B,MAAM,QAAQ,YAAY,IAAI;CAC9B,IAAI;CACJ,IAAI;CACJ,MAAM,WAAW,IAAI,wBAAwB,IAAI,OAAO,QAAQ,IAAI;CACpE,IAAI,6BAA6B;CAMjC,MAAM,eAAe,iBAAiB,IAAI,OAAO,QAAQ,IAAI;CAC7D,MAAM,sBAAsB,cAAc;CAC1C,IAAI;EACF,SAAS;GACP,MAAM,EAAC,UAAU,aAAY,MAAM,4BACjC,IACA,OAAO,QACP,gBACA,YACF;GACA,IACE,wBAAwB,KAAA,KACxB,SAAS,iBAAiB,uBAC1B,+BAA+B,mCAC/B;IAGA,SAAS,OAAO;IAChB,GAAG,OACD,uDACM,SAAS,eAAe,4BACxB,oBAAoB,iBAAiB,oBAAoB,IAAK,SACtE;IACA,MAAM,MAAM,iBAAiB;IAC7B;GACF;GAEA,CAAC,CAAC,aAAa;GACf,MAAM,aAA+B;IAAC,GAAG,OAAO;IAAY;GAAS;GACrE,SAAS,MAAM,SAAS,WAAW;GACnC,IAAI;GACJ,IAAI;IACF,UAAU,MAAM,WACd,IACA,YACA,OAAO,QAAQ,MACf,UACA,eACA,SAAS,MACX;GACF,SAAS,GAAG;IACV,IAAI,SAAS,OAAO,SAAS;KAC3B,GAAG,OACD,qDACA,OAAO,CAAC,CACV;KACA;IACF;IACA,SAAS,OAAO;IAChB,MAAM;GACR,UAAU;IACR,SAAS,KAAK;GAChB;GACA,IAAI,QAAQ,UAAU;IACpB,SAAS,QAAQ;IACjB,SAAS,KAAK;IACd,OAAO;GACT;GAEA,SAAS,OAAO;GAChB,GAAG,OACD,kCAAkC,oBAAoB,IAAK,SAC7D;GACA,MAAM,MAAM,iBAAiB;EAC/B;CACF,UAAU;EACR,SAAS,KAAK;EAMd,MAAM,SAAS;GAAC,GALF,6BACZ,OAAO,YACP,eACA,SAEiB;GAAO,QAAQ,UAAU;EAAO;EACnD,sBAAsB,EAAE,IAAI,GAAG,MAAM;EACrC,0BAA0B,EAAE,SAAS,YAAY,IAAI,IAAI,OAAO,MAAM;CACxE;AACF;AAGA,IAAI,CAAC,kBAAkB,GACrB,gBACQ,UACA,UAAU,KAAK,YAAY,GAAG,QAAQ,KAAK,GAAG,QAAQ,KAAK,MAAM,CAAC,CAAC,CAC3E"}

package/out/zero-cache/src/server/runner/run-worker.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"run-worker.d.ts","sourceRoot":"","sources":["../../../../../../zero-cache/src/server/runner/run-worker.ts"],"names":[],"mappings":"AAAA,OAAO,kCAAkC,CAAC;AAS1C,OAAO,EAAc,KAAK,MAAM,EAAC,MAAM,0BAA0B,CAAC;AAsBlE;;;;;;GAMG;AACH,wBAAsB,SAAS,CAC7B,MAAM,EAAE,MAAM,GAAG,IAAI,EACrB,GAAG,EAAE,MAAM,CAAC,UAAU,GACrB,OAAO,CAAC,IAAI,CAAC,CAgEf"}
+{"version":3,"file":"run-worker.d.ts","sourceRoot":"","sources":["../../../../../../zero-cache/src/server/runner/run-worker.ts"],"names":[],"mappings":"AAAA,OAAO,kCAAkC,CAAC;AAa1C,OAAO,EAGL,KAAK,MAAM,EACZ,MAAM,0BAA0B,CAAC;AAsBlC;;;;;;GAMG;AACH,wBAAsB,SAAS,CAC7B,MAAM,EAAE,MAAM,GAAG,IAAI,EACrB,GAAG,EAAE,MAAM,CAAC,UAAU,GACrB,OAAO,CAAC,IAAI,CAAC,CAqEf"}

package/out/zero-cache/src/server/runner/run-worker.js

@@ -4,7 +4,7 @@ import { colorConsole } from "../../../../shared/src/logging.js";
 import { childWorker } from "../../types/processes.js";
 import { normalizeZeroConfig } from "../../config/normalize.js";
 import { getServerVersion, getZeroConfig } from "../../config/zero-config.js";
-import { ProcessManager, runUntilKilled } from "../../services/life-cycle.js";
+import { ProcessManager, SelfDrain, runUntilKilled } from "../../services/life-cycle.js";
 import { createLogContext } from "../logging.js";
 import { MAIN_URL } from "../worker-urls.js";
 import { getTaskID } from "./runtime.js";
@@ -33,7 +33,7 @@ async function runWorker(parent, env) {
 	const lc = createLogContext(cfg, "runner");
 	const config = normalizeZeroConfig(lc, cfg, env, await getTaskID(lc));
 	const processes = new ProcessManager(lc, parent ?? process);
-	const { port, keepaliveTimeoutMs, lazyStartup } = config;
+	const { port, keepaliveTimeoutMs, lazyStartup, selfDrainDelayMs } = config;
 	const serverVersion = getServerVersion(config);
 	lc.info?.(`starting server${!serverVersion ? "" : `@${serverVersion}`} `, {
 		protocolVersion: 53,
@@ -42,6 +42,14 @@ async function runWorker(parent, env) {
 		shard: config.shard,
 		port: config.port
 	});
+	const dispatcher = new ZeroDispatcher(config, lc, {
+		port,
+		keepaliveTimeoutMs
+	}, startZeroCache, () => printStartupMessage(env));
+	const selfDrain = new SelfDrain(lc, parent ?? process, {
+		delayMs: selfDrainDelayMs,
+		onRequested: () => dispatcher.failHealthChecks()
+	});
 	let zeroCache;
 	function startZeroCache() {
 		if (zeroCache === void 0) {
@@ -52,6 +60,7 @@ async function runWorker(parent, env) {
 				r.resolve(w);
 				lc.info?.(`zero-cache ready (${performance.now() - startMs} ms)`);
 			}).once("error", r.reject);
+			w.onMessageType("drainRequested", () => selfDrain.request());
 			processes.addWorker(w, "user-facing", "zero-cache");
 		}
 		return zeroCache.promise;
@@ -60,10 +69,7 @@ async function runWorker(parent, env) {
 	await processes.allWorkersReady();
 	parent?.send(["ready", { ready: true }]);
 	try {
-		await runUntilKilled(lc, parent ?? process, new ZeroDispatcher(config, lc, {
-			port,
-			keepaliveTimeoutMs
-		}, startZeroCache, () => printStartupMessage(env)));
+		await runUntilKilled(lc, parent ?? process, dispatcher);
 	} catch (err) {
 		processes.logErrorAndExit(err, "main");
 	} finally {

package/out/zero-cache/src/server/runner/run-worker.js.map

@@ -1 +1 @@
-{"version":3,"file":"run-worker.js","names":[],"sources":["../../../../../../zero-cache/src/server/runner/run-worker.ts"],"sourcesContent":["import '../../../../shared/src/dotenv.ts';\n\nimport {styleText} from 'node:util';\nimport {resolver, type Resolver} from '@rocicorp/resolver';\nimport {colorConsole} from '../../../../shared/src/logging.ts';\nimport {PROTOCOL_VERSION} from '../../../../zero-protocol/src/protocol-version.ts';\nimport {normalizeZeroConfig} from '../../config/normalize.ts';\nimport {getServerVersion, getZeroConfig} from '../../config/zero-config.ts';\nimport {ProcessManager, runUntilKilled} from '../../services/life-cycle.ts';\nimport {childWorker, type Worker} from '../../types/processes.ts';\nimport {createLogContext} from '../logging.ts';\nimport {MAIN_URL} from '../worker-urls.ts';\nimport {getTaskID} from './runtime.ts';\nimport {ZeroDispatcher} from './zero-dispatcher.ts';\n\nconst startupMessageEnv = 'ZERO_ENABLE_STARTUP_MESSAGE';\n\nfunction printStartupMessage(env: NodeJS.ProcessEnv) {\n  if (env[startupMessageEnv] !== '1') {\n    return;\n  }\n\n  colorConsole.log(\n    `\\nBTW, ${styleText(['bold', 'cyan'], 'Cloud Zero')} ` +\n      'is now available - professional Zero hosting from the team that built it.\\n' +\n      `Get started now: ${styleText(['blue', 'underline'], 'https://zero.rocicorp.dev/cloud')}\\n\\n` +\n      styleText('dim', `Disable this message with ${startupMessageEnv}=0`) +\n      '\\n',\n  );\n}\n\n/**\n * Top-level `runner` entry point to the zero-cache. This layer is responsible for:\n * * runtime-based config normalization\n * * lazy startup\n * * serving /statsz\n * * auto-reset restarts (TODO)\n */\nexport async function runWorker(\n  parent: Worker | null,\n  env: NodeJS.ProcessEnv,\n): Promise<void> {\n  // Note: Deprecation warnings are only emitted at this top-level parse;\n  //       they are suppressed when parsed in subprocesses.\n  const cfg = getZeroConfig({env, emitDeprecationWarnings: true});\n  const lc = createLogContext(cfg, 'runner');\n\n  const defaultTaskID = await getTaskID(lc);\n  const config = normalizeZeroConfig(lc, cfg, env, defaultTaskID);\n  const processes = new ProcessManager(lc, parent ?? process);\n\n  const {port, keepaliveTimeoutMs, lazyStartup} = config;\n  const serverVersion = getServerVersion(config);\n  lc.info?.(`starting server${!serverVersion ? '' : `@${serverVersion}`} `, {\n    protocolVersion: PROTOCOL_VERSION,\n    taskID: config.taskID,\n    app: config.app,\n    shard: config.shard,\n    port: config.port,\n  });\n\n  let zeroCache: Resolver<Worker> | undefined;\n  function startZeroCache(): Promise<Worker> {\n    if (zeroCache === undefined) {\n      const startMs = performance.now();\n      lc.info?.('starting zero-cache');\n\n      const r = (zeroCache = resolver<Worker>());\n      const w = childWorker(MAIN_URL, env)\n        .once('message', () => {\n          r.resolve(w);\n          lc.info?.(`zero-cache ready (${performance.now() - startMs} ms)`);\n        })\n        .once('error', r.reject);\n\n      processes.addWorker(w, 'user-facing', 'zero-cache');\n    }\n    return zeroCache.promise;\n  }\n\n  // Eagerly start the zero-cache if it was not configured with --lazy-startup.\n  if (!lazyStartup) {\n    void startZeroCache();\n  }\n\n  await processes.allWorkersReady();\n  parent?.send(['ready', {ready: true}]);\n\n  try {\n    await runUntilKilled(\n      lc,\n      parent ?? process,\n      new ZeroDispatcher(\n        config,\n        lc,\n        {port, keepaliveTimeoutMs},\n        startZeroCache,\n        () => printStartupMessage(env),\n      ),\n    );\n  } catch (err) {\n    processes.logErrorAndExit(err, 'main');\n  } finally {\n    await processes.shutdown();\n  }\n}\n"],"mappings":";;;;;;;;;;;;;;AAeA,IAAM,oBAAoB;AAE1B,SAAS,oBAAoB,KAAwB;CACnD,IAAI,IAAI,uBAAuB,KAC7B;CAGF,aAAa,IACX,UAAU,UAAU,CAAC,QAAQ,MAAM,GAAG,YAAY,EAAE;mBAE9B,UAAU,CAAC,QAAQ,WAAW,GAAG,iCAAiC,EAAE,QACxF,UAAU,OAAO,6BAA6B,kBAAkB,GAAG,IACnE,IACJ;AACF;;;;;;;;AASA,eAAsB,UACpB,QACA,KACe;CAGf,MAAM,MAAM,cAAc;EAAC;EAAK,yBAAyB;CAAI,CAAC;CAC9D,MAAM,KAAK,iBAAiB,KAAK,QAAQ;CAGzC,MAAM,SAAS,oBAAoB,IAAI,KAAK,KAAK,MADrB,UAAU,EAAE,CACsB;CAC9D,MAAM,YAAY,IAAI,eAAe,IAAI,UAAU,OAAO;CAE1D,MAAM,EAAC,MAAM,oBAAoB,gBAAe;CAChD,MAAM,gBAAgB,iBAAiB,MAAM;CAC7C,GAAG,OAAO,kBAAkB,CAAC,gBAAgB,KAAK,IAAI,gBAAgB,IAAI;EACxE,iBAAA;EACA,QAAQ,OAAO;EACf,KAAK,OAAO;EACZ,OAAO,OAAO;EACd,MAAM,OAAO;CACf,CAAC;CAED,IAAI;CACJ,SAAS,iBAAkC;EACzC,IAAI,cAAc,KAAA,GAAW;GAC3B,MAAM,UAAU,YAAY,IAAI;GAChC,GAAG,OAAO,qBAAqB;GAE/B,MAAM,IAAK,YAAY,SAAiB;GACxC,MAAM,IAAI,YAAY,UAAU,GAAG,EAChC,KAAK,iBAAiB;IACrB,EAAE,QAAQ,CAAC;IACX,GAAG,OAAO,qBAAqB,YAAY,IAAI,IAAI,QAAQ,KAAK;GAClE,CAAC,EACA,KAAK,SAAS,EAAE,MAAM;GAEzB,UAAU,UAAU,GAAG,eAAe,YAAY;EACpD;EACA,OAAO,UAAU;CACnB;CAGA,IAAI,CAAC,aACH,eAAoB;CAGtB,MAAM,UAAU,gBAAgB;CAChC,QAAQ,KAAK,CAAC,SAAS,EAAC,OAAO,KAAI,CAAC,CAAC;CAErC,IAAI;EACF,MAAM,eACJ,IACA,UAAU,SACV,IAAI,eACF,QACA,IACA;GAAC;GAAM;EAAkB,GACzB,sBACM,oBAAoB,GAAG,CAC/B,CACF;CACF,SAAS,KAAK;EACZ,UAAU,gBAAgB,KAAK,MAAM;CACvC,UAAU;EACR,MAAM,UAAU,SAAS;CAC3B;AACF"}
+{"version":3,"file":"run-worker.js","names":[],"sources":["../../../../../../zero-cache/src/server/runner/run-worker.ts"],"sourcesContent":["import '../../../../shared/src/dotenv.ts';\n\nimport {styleText} from 'node:util';\nimport {resolver, type Resolver} from '@rocicorp/resolver';\nimport {colorConsole} from '../../../../shared/src/logging.ts';\nimport {PROTOCOL_VERSION} from '../../../../zero-protocol/src/protocol-version.ts';\nimport {normalizeZeroConfig} from '../../config/normalize.ts';\nimport {getServerVersion, getZeroConfig} from '../../config/zero-config.ts';\nimport {\n  ProcessManager,\n  runUntilKilled,\n  SelfDrain,\n} from '../../services/life-cycle.ts';\nimport {\n  childWorker,\n  type DrainRequestedMessage,\n  type Worker,\n} from '../../types/processes.ts';\nimport {createLogContext} from '../logging.ts';\nimport {MAIN_URL} from '../worker-urls.ts';\nimport {getTaskID} from './runtime.ts';\nimport {ZeroDispatcher} from './zero-dispatcher.ts';\n\nconst startupMessageEnv = 'ZERO_ENABLE_STARTUP_MESSAGE';\n\nfunction printStartupMessage(env: NodeJS.ProcessEnv) {\n  if (env[startupMessageEnv] !== '1') {\n    return;\n  }\n\n  colorConsole.log(\n    `\\nBTW, ${styleText(['bold', 'cyan'], 'Cloud Zero')} ` +\n      'is now available - professional Zero hosting from the team that built it.\\n' +\n      `Get started now: ${styleText(['blue', 'underline'], 'https://zero.rocicorp.dev/cloud')}\\n\\n` +\n      styleText('dim', `Disable this message with ${startupMessageEnv}=0`) +\n      '\\n',\n  );\n}\n\n/**\n * Top-level `runner` entry point to the zero-cache. This layer is responsible for:\n * * runtime-based config normalization\n * * lazy startup\n * * serving /statsz\n * * auto-reset restarts (TODO)\n */\nexport async function runWorker(\n  parent: Worker | null,\n  env: NodeJS.ProcessEnv,\n): Promise<void> {\n  // Note: Deprecation warnings are only emitted at this top-level parse;\n  //       they are suppressed when parsed in subprocesses.\n  const cfg = getZeroConfig({env, emitDeprecationWarnings: true});\n  const lc = createLogContext(cfg, 'runner');\n\n  const defaultTaskID = await getTaskID(lc);\n  const config = normalizeZeroConfig(lc, cfg, env, defaultTaskID);\n  const processes = new ProcessManager(lc, parent ?? process);\n\n  const {port, keepaliveTimeoutMs, lazyStartup, selfDrainDelayMs} = config;\n  const serverVersion = getServerVersion(config);\n  lc.info?.(`starting server${!serverVersion ? '' : `@${serverVersion}`} `, {\n    protocolVersion: PROTOCOL_VERSION,\n    taskID: config.taskID,\n    app: config.app,\n    shard: config.shard,\n    port: config.port,\n  });\n\n  const dispatcher = new ZeroDispatcher(\n    config,\n    lc,\n    {port, keepaliveTimeoutMs},\n    startZeroCache,\n    () => printStartupMessage(env),\n  );\n  const selfDrain = new SelfDrain(lc, parent ?? process, {\n    delayMs: selfDrainDelayMs,\n    onRequested: () => dispatcher.failHealthChecks(),\n  });\n\n  let zeroCache: Resolver<Worker> | undefined;\n  function startZeroCache(): Promise<Worker> {\n    if (zeroCache === undefined) {\n      const startMs = performance.now();\n      lc.info?.('starting zero-cache');\n\n      const r = (zeroCache = resolver<Worker>());\n      const w = childWorker(MAIN_URL, env)\n        .once('message', () => {\n          r.resolve(w);\n          lc.info?.(`zero-cache ready (${performance.now() - startMs} ms)`);\n        })\n        .once('error', r.reject);\n      w.onMessageType<DrainRequestedMessage>('drainRequested', () =>\n        selfDrain.request(),\n      );\n\n      processes.addWorker(w, 'user-facing', 'zero-cache');\n    }\n    return zeroCache.promise;\n  }\n\n  // Eagerly start the zero-cache if it was not configured with --lazy-startup.\n  if (!lazyStartup) {\n    void startZeroCache();\n  }\n\n  await processes.allWorkersReady();\n  parent?.send(['ready', {ready: true}]);\n\n  try {\n    await runUntilKilled(lc, parent ?? process, dispatcher);\n  } catch (err) {\n    processes.logErrorAndExit(err, 'main');\n  } finally {\n    await processes.shutdown();\n  }\n}\n"],"mappings":";;;;;;;;;;;;;;AAuBA,IAAM,oBAAoB;AAE1B,SAAS,oBAAoB,KAAwB;CACnD,IAAI,IAAI,uBAAuB,KAC7B;CAGF,aAAa,IACX,UAAU,UAAU,CAAC,QAAQ,MAAM,GAAG,YAAY,EAAE;mBAE9B,UAAU,CAAC,QAAQ,WAAW,GAAG,iCAAiC,EAAE,QACxF,UAAU,OAAO,6BAA6B,kBAAkB,GAAG,IACnE,IACJ;AACF;;;;;;;;AASA,eAAsB,UACpB,QACA,KACe;CAGf,MAAM,MAAM,cAAc;EAAC;EAAK,yBAAyB;CAAI,CAAC;CAC9D,MAAM,KAAK,iBAAiB,KAAK,QAAQ;CAGzC,MAAM,SAAS,oBAAoB,IAAI,KAAK,KAAK,MADrB,UAAU,EAAE,CACsB;CAC9D,MAAM,YAAY,IAAI,eAAe,IAAI,UAAU,OAAO;CAE1D,MAAM,EAAC,MAAM,oBAAoB,aAAa,qBAAoB;CAClE,MAAM,gBAAgB,iBAAiB,MAAM;CAC7C,GAAG,OAAO,kBAAkB,CAAC,gBAAgB,KAAK,IAAI,gBAAgB,IAAI;EACxE,iBAAA;EACA,QAAQ,OAAO;EACf,KAAK,OAAO;EACZ,OAAO,OAAO;EACd,MAAM,OAAO;CACf,CAAC;CAED,MAAM,aAAa,IAAI,eACrB,QACA,IACA;EAAC;EAAM;CAAkB,GACzB,sBACM,oBAAoB,GAAG,CAC/B;CACA,MAAM,YAAY,IAAI,UAAU,IAAI,UAAU,SAAS;EACrD,SAAS;EACT,mBAAmB,WAAW,iBAAiB;CACjD,CAAC;CAED,IAAI;CACJ,SAAS,iBAAkC;EACzC,IAAI,cAAc,KAAA,GAAW;GAC3B,MAAM,UAAU,YAAY,IAAI;GAChC,GAAG,OAAO,qBAAqB;GAE/B,MAAM,IAAK,YAAY,SAAiB;GACxC,MAAM,IAAI,YAAY,UAAU,GAAG,EAChC,KAAK,iBAAiB;IACrB,EAAE,QAAQ,CAAC;IACX,GAAG,OAAO,qBAAqB,YAAY,IAAI,IAAI,QAAQ,KAAK;GAClE,CAAC,EACA,KAAK,SAAS,EAAE,MAAM;GACzB,EAAE,cAAqC,wBACrC,UAAU,QAAQ,CACpB;GAEA,UAAU,UAAU,GAAG,eAAe,YAAY;EACpD;EACA,OAAO,UAAU;CACnB;CAGA,IAAI,CAAC,aACH,eAAoB;CAGtB,MAAM,UAAU,gBAAgB;CAChC,QAAQ,KAAK,CAAC,SAAS,EAAC,OAAO,KAAI,CAAC,CAAC;CAErC,IAAI;EACF,MAAM,eAAe,IAAI,UAAU,SAAS,UAAU;CACxD,SAAS,KAAK;EACZ,UAAU,gBAAgB,KAAK,MAAM;CACvC,UAAU;EACR,MAAM,UAAU,SAAS;CAC3B;AACF"}

package/out/zero-cache/src/services/change-streamer/change-streamer-http.d.ts

@@ -7,7 +7,7 @@ import { type Sized, type Source } from '../../types/streams.ts';
 import { Subscription } from '../../types/subscription.ts';
 import { HttpService, type Options as HttpOptions } from '../http-service.ts';
 import { type ChangeStreamer, type ChangeStreamerService, type SubscriberContext } from './change-streamer.ts';
-import { type SnapshotStatus, type SubscribeDownstream, type SubscribeUpstream } from './subscribe.ts';
+import { type LocalReplica, type SnapshotStatus, type SubscribeDownstream, type SubscribeUpstream } from './subscribe.ts';
 type Options = HttpOptions & {
     startupDelayMs: number;
     config?: Pick<NormalizedZeroConfig, 'adminPassword' | 'operatorPassword'> | undefined;
@@ -18,10 +18,16 @@ export declare class ChangeStreamerHttpServer extends HttpService {
     readonly id = "change-streamer-http-server";
     constructor(lc: LogContext, opts: Options, parent: Worker, changeStreamer: ChangeStreamerService);
     protected _onStart(): void;
+    /**
+     * Signals (drainable) subscribers to switch over to another
+     * replication-manager, resolving when they have disconnected. Subscribers
+     * continue to be served in the meantime.
+     */
+    drain(): Promise<void>;
     protected _onStop(): Promise<void>;
 }
 export interface SnapshotReserver {
-    reserveSnapshot(taskID: string): Promise<{
+    reserveSnapshot(taskID: string, localReplica?: LocalReplica | undefined): Promise<{
         reserved: SnapshotStatus;
         followup: ReservationFollowup;
     }>;
@@ -46,7 +52,7 @@ export interface ReservationFollowup extends SubscriptionStarter {
 export declare class ChangeStreamerHttpClient implements ChangeStreamer, SnapshotReserver, SubscriptionStarter {
     #private;
     constructor(lc: LogContext, shardID: ShardID, changeDB: string, changeStreamerURI: string | undefined);
-    reserveSnapshot(taskID: string): Promise<{
+    reserveSnapshot(taskID: string, localReplica?: LocalReplica | undefined): Promise<{
         reserved: SnapshotStatus;
         followup: ReservationFollowup;
     }>;

package/out/zero-cache/src/services/change-streamer/change-streamer-http.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"change-streamer-http.d.ts","sourceRoot":"","sources":["../../../../../../zero-cache/src/services/change-streamer/change-streamer-http.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAC,eAAe,EAAC,MAAM,WAAW,CAAC;AAE/C,OAAO,KAAK,EAAC,UAAU,EAAC,MAAM,kBAAkB,CAAC;AAIjD,OAAO,KAAK,EAAC,oBAAoB,EAAC,MAAM,2BAA2B,CAAC;AAGpE,OAAO,EAAC,KAAK,MAAM,EAAC,MAAM,0BAA0B,CAAC;AACrD,OAAO,EAAC,KAAK,OAAO,EAAC,MAAM,uBAAuB,CAAC;AACnD,OAAO,EAKL,KAAK,KAAK,EACV,KAAK,MAAM,EACZ,MAAM,wBAAwB,CAAC;AAChC,OAAO,EAAC,YAAY,EAAC,MAAM,6BAA6B,CAAC;AAIzD,OAAO,EAAC,WAAW,EAAE,KAAK,OAAO,IAAI,WAAW,EAAC,MAAM,oBAAoB,CAAC;AAG5E,OAAO,EAEL,KAAK,cAAc,EACnB,KAAK,qBAAqB,EAC1B,KAAK,iBAAiB,EACvB,MAAM,sBAAsB,CAAC;AAG9B,OAAO,EAIL,KAAK,cAAc,EAEnB,KAAK,mBAAmB,EACxB,KAAK,iBAAiB,EACvB,MAAM,gBAAgB,CAAC;AAmBxB,KAAK,OAAO,GAAG,WAAW,GAAG;IAC3B,cAAc,EAAE,MAAM,CAAC;IACvB,MAAM,CAAC,EACH,IAAI,CAAC,oBAAoB,EAAE,eAAe,GAAG,kBAAkB,CAAC,GAChE,SAAS,CAAC;IACd,gBAAgB,CAAC,EAAE,CAAC,MAAM,OAAO,CAAC,MAAM,CAAC,CAAC,GAAG,SAAS,CAAC;CACxD,CAAC;AAEF,qBAAa,wBAAyB,SAAQ,WAAW;;IACvD,QAAQ,CAAC,EAAE,iCAAiC;IAK5C,YACE,EAAE,EAAE,UAAU,EACd,IAAI,EAAE,OAAO,EACb,MAAM,EAAE,MAAM,EACd,cAAc,EAAE,qBAAqB,EAuCtC;IAiLD,UAAmB,QAAQ,IAAI,IAAI,CA2BlC;IAED,UAAyB,OAAO,IAAI,OAAO,CAAC,IAAI,CAAC,CAIhD;CACF;AAED,MAAM,WAAW,gBAAgB;IAC/B,eAAe,CAAC,MAAM,EAAE,MAAM,GAAG,OAAO,CAAC;QACvC,QAAQ,EAAE,cAAc,CAAC;QACzB,QAAQ,EAAE,mBAAmB,CAAC;KAC/B,CAAC,CAAC;CACJ;AAED,MAAM,WAAW,mBAAmB;IAClC,SAAS,CACP,GAAG,EAAE,iBAAiB,GACrB,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,mBAAmB,CAAC,CAAC,CAAC,CAAC;CAChD;AAED,MAAM,WAAW,mBAAoB,SAAQ,mBAAmB;IAC9D;;;OAGG;IACH,QAAQ,CAAC,MAAM,EAAE,WAAW,CAAC;IAE7B;;;;;OAKG;IACH,MAAM,CAAC,MAAM,CAAC,EAAE,KAAK,GAAG,IAAI,CAAC;CAC9B;AAED,qBAAa,wBACX,YAAW,cAAc,EAAE,gBAAgB,EAAE,mBAAmB;;IAOhE,YACE,EAAE,EAAE,UAAU,EACd,OAAO,EAAE,OAAO,EAChB,QAAQ,EAAE,MAAM,EAChB,iBAAiB,EAAE,MAAM,GAAG,SAAS,EAWtC;IAmBK,eAAe,CAAC,MAAM,EAAE,MAAM,GAAG,OAAO,CAAC;QAC7C,QAAQ,EAAE,cAAc,CAAC;QACzB,QAAQ,EAAE,mBAAmB,CAAC;KAC/B,CAAC,CAwBD;IAEK,SAAS,CACb,GAAG,EAAE,iBAAiB,GACrB,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,mBAAmB,CAAC,CAAC,CAAC,CAI7C;IAED;;;;;;;OAOG;IAEG,OAAO,CAAC,MAAM,EAAE,MAAM,GAAG,OAAO,CAAC;QACrC,QAAQ,EAAE,MAAM,CAAC,KAAK,CAAC,mBAAmB,CAAC,CAAC,CAAC;QAC7C,QAAQ,EAAE,YAAY,CAAC,iBAAiB,CAAC,CAAC;KAC3C,CAAC,CAuBD;CACF;AAED,KAAK,cAAc,GAAG,IAAI,CAAC,eAAe,EAAE,KAAK,GAAG,SAAS,CAAC,CAAC;AAE/D,wBAAgB,oBAAoB,CAAC,GAAG,EAAE,cAAc,GAAG,iBAAiB,CAiB3E"}
+{"version":3,"file":"change-streamer-http.d.ts","sourceRoot":"","sources":["../../../../../../zero-cache/src/services/change-streamer/change-streamer-http.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAC,eAAe,EAAC,MAAM,WAAW,CAAC;AAE/C,OAAO,KAAK,EAAC,UAAU,EAAC,MAAM,kBAAkB,CAAC;AAIjD,OAAO,KAAK,EAAC,oBAAoB,EAAC,MAAM,2BAA2B,CAAC;AAGpE,OAAO,EAAC,KAAK,MAAM,EAAC,MAAM,0BAA0B,CAAC;AACrD,OAAO,EAAC,KAAK,OAAO,EAAC,MAAM,uBAAuB,CAAC;AACnD,OAAO,EAKL,KAAK,KAAK,EACV,KAAK,MAAM,EACZ,MAAM,wBAAwB,CAAC;AAChC,OAAO,EAAC,YAAY,EAAC,MAAM,6BAA6B,CAAC;AAIzD,OAAO,EAAC,WAAW,EAAE,KAAK,OAAO,IAAI,WAAW,EAAC,MAAM,oBAAoB,CAAC;AAG5E,OAAO,EAEL,KAAK,cAAc,EACnB,KAAK,qBAAqB,EAC1B,KAAK,iBAAiB,EACvB,MAAM,sBAAsB,CAAC;AAG9B,OAAO,EACL,KAAK,YAAY,EAIjB,KAAK,cAAc,EAEnB,KAAK,mBAAmB,EACxB,KAAK,iBAAiB,EACvB,MAAM,gBAAgB,CAAC;AAmBxB,KAAK,OAAO,GAAG,WAAW,GAAG;IAC3B,cAAc,EAAE,MAAM,CAAC;IACvB,MAAM,CAAC,EACH,IAAI,CAAC,oBAAoB,EAAE,eAAe,GAAG,kBAAkB,CAAC,GAChE,SAAS,CAAC;IACd,gBAAgB,CAAC,EAAE,CAAC,MAAM,OAAO,CAAC,MAAM,CAAC,CAAC,GAAG,SAAS,CAAC;CACxD,CAAC;AAEF,qBAAa,wBAAyB,SAAQ,WAAW;;IACvD,QAAQ,CAAC,EAAE,iCAAiC;IAK5C,YACE,EAAE,EAAE,UAAU,EACd,IAAI,EAAE,OAAO,EACb,MAAM,EAAE,MAAM,EACd,cAAc,EAAE,qBAAqB,EAuCtC;IAmLD,UAAmB,QAAQ,IAAI,IAAI,CA2BlC;IAED;;;;OAIG;IACG,KAAK,IAAI,OAAO,CAAC,IAAI,CAAC,CAI3B;IAED,UAAyB,OAAO,IAAI,OAAO,CAAC,IAAI,CAAC,CAIhD;CACF;AAED,MAAM,WAAW,gBAAgB;IAC/B,eAAe,CACb,MAAM,EAAE,MAAM,EACd,YAAY,CAAC,EAAE,YAAY,GAAG,SAAS,GACtC,OAAO,CAAC;QACT,QAAQ,EAAE,cAAc,CAAC;QACzB,QAAQ,EAAE,mBAAmB,CAAC;KAC/B,CAAC,CAAC;CACJ;AAED,MAAM,WAAW,mBAAmB;IAClC,SAAS,CACP,GAAG,EAAE,iBAAiB,GACrB,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,mBAAmB,CAAC,CAAC,CAAC,CAAC;CAChD;AAED,MAAM,WAAW,mBAAoB,SAAQ,mBAAmB;IAC9D;;;OAGG;IACH,QAAQ,CAAC,MAAM,EAAE,WAAW,CAAC;IAE7B;;;;;OAKG;IACH,MAAM,CAAC,MAAM,CAAC,EAAE,KAAK,GAAG,IAAI,CAAC;CAC9B;AAED,qBAAa,wBACX,YAAW,cAAc,EAAE,gBAAgB,EAAE,mBAAmB;;IAOhE,YACE,EAAE,EAAE,UAAU,EACd,OAAO,EAAE,OAAO,EAChB,QAAQ,EAAE,MAAM,EAChB,iBAAiB,EAAE,MAAM,GAAG,SAAS,EAWtC;IAmBK,eAAe,CACnB,MAAM,EAAE,MAAM,EACd,YAAY,CAAC,EAAE,YAAY,GAAG,SAAS,GACtC,OAAO,CAAC;QACT,QAAQ,EAAE,cAAc,CAAC;QACzB,QAAQ,EAAE,mBAAmB,CAAC;KAC/B,CAAC,CA2BD;IAEK,SAAS,CACb,GAAG,EAAE,iBAAiB,GACrB,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,mBAAmB,CAAC,CAAC,CAAC,CAI7C;IAED;;;;;;;OAOG;IAEG,OAAO,CAAC,MAAM,EAAE,MAAM,GAAG,OAAO,CAAC;QACrC,QAAQ,EAAE,MAAM,CAAC,KAAK,CAAC,mBAAmB,CAAC,CAAC,CAAC;QAC7C,QAAQ,EAAE,YAAY,CAAC,iBAAiB,CAAC,CAAC;KAC3C,CAAC,CAuBD;CACF;AAED,KAAK,cAAc,GAAG,IAAI,CAAC,eAAe,EAAE,KAAK,GAAG,SAAS,CAAC,CAAC;AAE/D,wBAAgB,oBAAoB,CAAC,GAAG,EAAE,cAAc,GAAG,iBAAiB,CAiB3E"}

package/out/zero-cache/src/services/change-streamer/change-streamer-http.js

@@ -93,9 +93,9 @@ var ChangeStreamerHttpServer = class extends HttpService {
 			if (protocolVersion < MIN_SUBSCRIBE_PROTOCOL_VERSION) throw new Error(`the /subscribe endpoint requires protocol v${MIN_SUBSCRIBE_PROTOCOL_VERSION} (client is at v${protocolVersion})`);
 			const instream = await streamInternalStringified(this._lc, ws, subscribeUpstreamSchema, outbound, { batched: true });
 			for await (const msg of instream) if (msg[0] === "reserve-snapshot") {
-				const { taskID } = msg[1];
+				const { taskID, localReplica } = msg[1];
 				this.#ensureChangeStreamerStarted("incoming snapshot reservation");
-				reservation = await this.#changeStreamer.startSnapshotReservation(taskID);
+				reservation = await this.#changeStreamer.startSnapshotReservation(taskID, { localReplica });
 				this.#forwardReservation(reservation, outbound);
 			} else if (msg[0] === "start-subscription") {
 				const ctx = {
@@ -147,6 +147,14 @@ var ChangeStreamerHttpServer = class extends HttpService {
 			this._state.setTimeout(() => this.#ensureChangeStreamerStarted(`startup delay elapsed (${startupDelayMs} ms)`), startupDelayMs);
 		});
 	}
+	/**
+	* Signals (drainable) subscribers to switch over to another
+	* replication-manager, resolving when they have disconnected. Subscribers
+	* continue to be served in the meantime.
+	*/
+	async drain() {
+		if (this.#changeStreamerStarted && this.#changeStreamer.drain) await this.#changeStreamer.drain();
+	}
 	async _onStop() {
 		if (this.#changeStreamerStarted) await this.#changeStreamer.stop();
 	}
@@ -176,9 +184,12 @@ var ChangeStreamerHttpClient = class {
 		this.#lc.info?.(`connecting to change-streamer@${uri}`);
 		return uri;
 	}
-	async reserveSnapshot(taskID) {
+	async reserveSnapshot(taskID, localReplica) {
 		const { instream, outbound } = await this.connect(taskID);
-		outbound.push(["reserve-snapshot", { taskID }]);
+		outbound.push(["reserve-snapshot", {
+			taskID,
+			...localReplica ? { localReplica } : {}
+		}]);
 		const { head, rest } = readHead(instream);
 		const item = await head.next();
 		if (!item.value) throw new Error("stream closed without snapshot reservation");
@@ -240,14 +251,16 @@ function checkProtocolVersion(pathname) {
 	return v;
 }
 function toStartSubscriptionContext(ctx) {
-	const { taskID, id, mode, replicaVersion, watermark, backfills } = ctx;
+	const { taskID, id, mode, replicaVersion, watermark, backfills, drainable, switchover } = ctx;
 	return {
 		taskID,
 		id,
 		mode,
 		replicaVersion,
 		watermark,
-		backfills
+		backfills,
+		...drainable !== void 0 ? { drainable } : {},
+		...switchover !== void 0 ? { switchover } : {}
 	};
 }
 //#endregion