Downgrade db schema via Curio Toolkit

cmd/curio/internal/translations/locales/en/out.gotext.json

@@ -92,6 +92,27 @@
             "translatorComment": "Copied from source.",
             "fuzzy": true
         },
+        {
+            "id": "Downgrade a cluster's daatabase to a previous software version.",
+            "message": "Downgrade a cluster's daatabase to a previous software version.",
+            "translation": "Downgrade a cluster's daatabase to a previous software version.",
+            "translatorComment": "Copied from source.",
+            "fuzzy": true
+        },
+        {
+            "id": "If, however, the upgrade has a serious bug and you need to downgrade, first shutdown all nodes in your cluster and then run this command. Finally, only start downgraded nodes.",
+            "message": "If, however, the upgrade has a serious bug and you need to downgrade, first shutdown all nodes in your cluster and then run this command. Finally, only start downgraded nodes.",
+            "translation": "If, however, the upgrade has a serious bug and you need to downgrade, first shutdown all nodes in your cluster and then run this command. Finally, only start downgraded nodes.",
+            "translatorComment": "Copied from source.",
+            "fuzzy": true
+        },
+        {
+            "id": "YYYYMMDD when your cluster had the preferred schema. Ex: 20251128",
+            "message": "YYYYMMDD when your cluster had the preferred schema. Ex: 20251128",
+            "translation": "YYYYMMDD when your cluster had the preferred schema. Ex: 20251128",
+            "translatorComment": "Copied from source.",
+            "fuzzy": true
+        },
         {
             "id": "Manage node config by layers. The layer 'base' will always be applied at Curio start-up.",
             "message": "Manage node config by layers. The layer 'base' will always be applied at Curio start-up.",

cmd/curio/internal/translations/locales/ko/messages.gotext.json

@@ -2931,6 +2931,24 @@
       "translation": "PDP 지갑이 가져와졌습니다",
       "message": "PDP wallet imported",
       "placeholder": null
+    },
+    {
+      "id": "Downgrade a cluster's daatabase to a previous software version.",
+      "translation": "클러스터의 데이터베이스를 이전 소프트웨어 버전으로 다운그레이드합니다.",
+      "message": "Downgrade a cluster's daatabase to a previous software version.",
+      "placeholder": null
+    },
+    {
+      "id": "If, however, the upgrade has a serious bug and you need to downgrade, first shutdown all nodes in your cluster and then run this command. Finally, only start downgraded nodes.",
+      "translation": "그러나 업그레이드에 심각한 버그가 있어 다운그레이드해야 한다면, 먼저 클러스터의 모든 노드를 종료한 뒤 이 명령을 실행하십시오. 마지막으로 다운그레이드된 노드만 다시 시작하십시오.",
+      "message": "If, however, the upgrade has a serious bug and you need to downgrade, first shutdown all nodes in your cluster and then run this command. Finally, only start downgraded nodes.",
+      "placeholder": null
+    },
+    {
+      "id": "YYYYMMDD when your cluster had the preferred schema. Ex: 20251128",
+      "translation": "클러스터에 원하는 스키마가 적용되어 있던 YYYYMMDD 날짜입니다. 예: 20251128",
+      "message": "YYYYMMDD when your cluster had the preferred schema. Ex: 20251128",
+      "placeholder": null
     }
   ]
 }

cmd/curio/internal/translations/locales/zh/messages.gotext.json

@@ -2901,6 +2901,24 @@
       "translation": "1. 使用以下命令测试您的 PDP 服务：pdptool ping --service-url https://your-domain.com --service-name public",
       "message": "1. Test your PDP service with: pdptool ping --service-url https://your-domain.com --service-name public",
       "placeholder": null
+    },
+    {
+      "id": "Downgrade a cluster's daatabase to a previous software version.",
+      "translation": "将集群的数据库降级到之前的软件版本。",
+      "message": "Downgrade a cluster's daatabase to a previous software version.",
+      "placeholder": null
+    },
+    {
+      "id": "If, however, the upgrade has a serious bug and you need to downgrade, first shutdown all nodes in your cluster and then run this command. Finally, only start downgraded nodes.",
+      "translation": "但是，如果升级存在严重缺陷需要降级，请先关闭集群中的所有节点，然后运行此命令。最后，仅启动已降级的节点。",
+      "message": "If, however, the upgrade has a serious bug and you need to downgrade, first shutdown all nodes in your cluster and then run this command. Finally, only start downgraded nodes.",
+      "placeholder": null
+    },
+    {
+      "id": "YYYYMMDD when your cluster had the preferred schema. Ex: 20251128",
+      "translation": "集群拥有首选架构的 YYYYMMDD 日期。例如：20251128",
+      "message": "YYYYMMDD when your cluster had the preferred schema. Ex: 20251128",
+      "placeholder": null
     }
   ]
 }

cmd/curio/toolbox.go

@@ -8,6 +8,7 @@ import (
 	"net/http"
 	"net/url"
 	"strconv"
+	"strings"
 
 	"github.com/docker/go-units"
 	"github.com/ethereum/go-ethereum/common"
@@ -36,6 +37,7 @@ var toolboxCmd = &cli.Command{
 	Subcommands: []*cli.Command{
 		fixMsgCmd,
 		registerPDPServiceProviderCmd,
+		downgradeCmd,
 	},
 }
 
@@ -439,3 +441,34 @@ var registerPDPServiceProviderCmd = &cli.Command{
 		return nil
 	},
 }
+
+var downgradeCmd = &cli.Command{
+	Name:        "downgrade",
+	Usage:       translations.T("Downgrade a cluster's daatabase to a previous software version."),
+	Description: translations.T("If, however, the upgrade has a serious bug and you need to downgrade, first shutdown all nodes in your cluster and then run this command. Finally, only start downgraded nodes."),
+	Flags: []cli.Flag{
+		&cli.IntFlag{
+			Name:     "last_good_date",
+			Usage:    translations.T("YYYYMMDD when your cluster had the preferred schema. Ex: 20251128"),
+			Required: true,
+		},
+	},
+	Action: func(cctx *cli.Context) error {
+		db, err := deps.MakeDB(cctx)
+		if err != nil {
+			return err
+		}
+
+		var runningMachines []string
+		if err := db.Select(cctx.Context, &runningMachines, `SELECT host_and_port FROM harmony_machines 
+		WHERE last_contact > CURRENT_TIMESTAMP - INTERVAL '1 MINUTE' `); err != nil {
+			return err
+		}
+
+		if len(runningMachines) > 0 {
+			return xerrors.Errorf("All machines must be shutdown before downgrading. Machines seen running in the past 60 seconds: %s", strings.Join(runningMachines, ", "))
+		}
+
+		return db.DowngradeTo(cctx.Context, cctx.Int("last_good_date"))
+	},
+}

documentation/en/curio-cli/curio.md

@@ -1296,6 +1296,7 @@ USAGE:
 COMMANDS:
    fix-msg                        Updated DB with message data missing from chain node
    register-pdp-service-provider  Register a PDP service provider with Filecoin Service Registry Contract
+   downgrade                      Downgrade a cluster's daatabase to a previous software version.
    help, h                        Shows a list of commands or help for one command
 
 OPTIONS:
@@ -1337,3 +1338,19 @@ OPTIONS:
    --token-address value   Token contract for payment (IERC20(address(0)) for FIL)
    --help, -h              show help
 ```
+
+### curio toolbox downgrade
+```
+NAME:
+   curio toolbox downgrade - Downgrade a cluster's daatabase to a previous software version.
+
+USAGE:
+   curio toolbox downgrade [command options]
+
+DESCRIPTION:
+   If, however, the upgrade has a serious bug and you need to downgrade, first shutdown all nodes in your cluster and then run this command. Finally, only start downgraded nodes.
+
+OPTIONS:
+   --last_good_date value  YYYYMMDD when your cluster had the preferred schema. Ex: 20251128 (default: 0)
+   --help, -h              show help
+```

documentation/en/versions.md

@@ -36,6 +36,11 @@ Configurations and the number of machines needed: A: Lotus, Curio (numerous), Yu
 * Curio's DEBs include curio-cuda (for Nvidia) and curio-opencl (others like ATI).
   * These can be mixed in a Curio cluster as they only relate to the hardware on the box.
 
+## Database Schema Versions
+* When the latest Curio starts-up, it applies any upgrades & migrations to Yugabyte's schema.
+* This may cause errors on other nodes in your cluster that run the old version (low likelihood), which has the simple solution of completing the upgrade.
+* If, however, the upgrade has a serious bug and you need to downgrade, "curio toolbox downgrade --last_good_date=20250515"
+
 ## Notes
 
 * Forest (0.19+ & Docker Watchtower) is a light alternative to Lotus Client. It meets Curio's needs, but Boost compatibility is in development.

harmony/harmonydb/downgrade/20250811-fix-commit-batching.sql

@@ -0,0 +1,126 @@
+
+CREATE OR REPLACE FUNCTION poll_start_batch_commit_msgs(
+    p_slack_epoch    BIGINT,  -- "Slack" epoch offset
+    p_current_height BIGINT,  -- Current on-chain height
+    p_max_batch      INT,     -- Max sectors per batch
+    p_new_task_id    BIGINT,  -- Task ID to set for a chosen batch
+    p_timeout_secs   INT      -- If earliest_ready + this > now(), condition is met
+)
+-- We return a TABLE of (updated_count BIGINT, reason TEXT),
+-- but in practice it will yield exactly one row.
+RETURNS TABLE (
+    updated_count BIGINT,
+    reason        TEXT
+)
+LANGUAGE plpgsql
+AS $$
+DECLARE
+batch_rec RECORD;
+    cond_slack   BOOLEAN;
+    cond_timeout BOOLEAN;
+    cond_fee     BOOLEAN;
+BEGIN
+    -- Default outputs if we never find a batch
+    updated_count := 0;
+    reason        := 'NONE';
+    /*
+      Single query logic:
+        (1) Select the rows that need commit assignment.
+        (2) Partition them by (sp_id, reg_seal_proof), using ROW_NUMBER() to break
+            them into sub-batches of size p_max_batch.
+        (3) GROUP those sub-batches to get:
+            - batch_start_epoch = min(start_epoch)
+            - earliest_ready_at = min(commit_ready_at)
+            - sector_nums = array of sector_number
+        (4) Loop over results, check conditions, update if found, return count.
+        (5) If we finish the loop, return 0.
+    */
+    FOR batch_rec IN
+        WITH initial AS (
+            SELECT
+                sp_id,
+                sector_number,
+                start_epoch,
+                commit_ready_at,
+                reg_seal_proof
+            FROM sectors_sdr_pipeline
+            WHERE after_porep        = TRUE
+              AND porep_proof        IS NOT NULL
+              AND task_id_commit_msg IS NULL
+              AND after_commit_msg   = FALSE
+              AND start_epoch        IS NOT NULL
+            ORDER BY sp_id, reg_seal_proof, start_epoch
+        ),
+        numbered AS (
+            SELECT
+              l.*,
+              ROW_NUMBER() OVER (
+                PARTITION BY l.sp_id, l.reg_seal_proof
+                ORDER BY l.commit_ready_at
+              ) AS rn
+            FROM initial l
+        ),
+        chunked AS (
+            SELECT
+              sp_id,
+              reg_seal_proof,
+              FLOOR((rn - 1)::NUMERIC / p_max_batch) AS batch_index,
+              start_epoch,
+              commit_ready_at,
+              sector_number
+            FROM numbered
+        ),
+        grouped AS (
+            SELECT
+              sp_id,
+              reg_seal_proof,
+              batch_index,
+              MIN(start_epoch)               AS batch_start_epoch,
+              MIN(commit_ready_at)           AS earliest_ready_at,
+              ARRAY_AGG(sector_number)       AS sector_nums
+            FROM chunked
+            GROUP BY sp_id, reg_seal_proof, batch_index
+            ORDER BY sp_id, reg_seal_proof, batch_index
+        )
+        SELECT
+            sp_id,
+            reg_seal_proof,
+            sector_nums,
+            batch_start_epoch,
+            earliest_ready_at
+        FROM grouped
+    LOOP
+             -- Evaluate conditions separately so we can pick a 'reason' if triggered.
+            cond_slack   := ((batch_rec.batch_start_epoch - p_slack_epoch) <= p_current_height);
+            cond_timeout := (NOW() >= (batch_rec.earliest_ready_at + MAKE_INTERVAL(secs => p_timeout_secs)));
+
+        IF (cond_slack OR cond_timeout OR cond_fee) THEN
+            -- If multiple conditions are true, pick an order of precedence.
+            IF cond_slack THEN
+                reason := 'SLACK (min start epoch: ' || batch_rec.batch_start_epoch || ')';
+            ELSIF cond_timeout THEN
+                reason := 'TIMEOUT (earliest_ready_at: ' || batch_rec.earliest_ready_at || ')';
+            END IF;
+
+            -- Perform the update
+            UPDATE sectors_sdr_pipeline t
+            SET task_id_commit_msg = p_new_task_id
+            WHERE t.sp_id         = batch_rec.sp_id
+              AND t.reg_seal_proof = batch_rec.reg_seal_proof
+              AND t.sector_number = ANY(batch_rec.sector_nums)
+              AND t.after_porep = TRUE
+              AND t.task_id_commit_msg IS NULL
+              AND t.after_commit_msg = FALSE;
+
+            GET DIAGNOSTICS updated_count = ROW_COUNT;
+
+            RETURN NEXT;
+            RETURN;  -- Return immediately with updated_count and reason
+        END IF;
+    END LOOP;
+
+    -- If we finish the loop with no triggered condition, we return updated_count=0, reason='NONE'
+    RETURN NEXT;
+    RETURN;
+END;
+$$;

harmony/harmonydb/downgrade/20250817-balancemgr-pshare.sql

@@ -0,0 +1,2 @@
+ALTER TABLE balance_manager_addresses DROP CONSTRAINT IF EXISTS subject_not_equal_second;
+ALTER TABLE balance_manager_addresses ADD CONSTRAINT subject_not_equal_second CHECK (subject_address != second_address);

harmony/harmonydb/downgrade/20250926-harmony_config_timestamp.sql

@@ -0,0 +1 @@
+#no need to revert