From c6ec3fa8fde6e8572696e6a2cffae30cf593c579 Mon Sep 17 00:00:00 2001
From: Neeraj Gupta <254676+ua741@users.noreply.github.com>
Date: Fri, 29 Nov 2024 12:24:12 +0530
Subject: [PATCH] [server] Fixed delay in next attemp

---
 server/pkg/controller/replication3.go | 10 ++++++++++
 server/pkg/repo/object_copies.go      | 12 ++++++++++--
 2 files changed, 20 insertions(+), 2 deletions(-)

diff --git a/server/pkg/controller/replication3.go b/server/pkg/controller/replication3.go
index c7b52ff870..a57a372fab 100644
--- a/server/pkg/controller/replication3.go
+++ b/server/pkg/controller/replication3.go
@@ -250,6 +250,16 @@ func (c *ReplicationController3) tryReplicate() error {
 			logger.Error(err)
 		}
 
+		if strings.Contains(err.Error(), "size of the uploaded file") {
+			delayErr := c.ObjectCopiesRepo.DelayNextAttemptByDays(context.Background(), objectKey, 7)
+			if delayErr != nil {
+				logger.WithError(delayErr).Error("Failed to delay next attempt by 7 days")
+			} else {
+				discordAlert := fmt.Sprintf("🔥 Size mismatch for object %s, failed to delay next attempt by 7 days", objectKey)
+				c.notifyDiscord(discordAlert)
+			}
+		}
+
 		if err == nil {
 			logger.Info("Replication attempt succeeded")
 		} else {
diff --git a/server/pkg/repo/object_copies.go b/server/pkg/repo/object_copies.go
index 31443e25c5..33245740f8 100644
--- a/server/pkg/repo/object_copies.go
+++ b/server/pkg/repo/object_copies.go
@@ -35,7 +35,6 @@ func (repo *ObjectCopiesRepository) GetAndLockUnreplicatedObject(ctx context.Con
 		}
 	}
 
-	// todo:(neeraj) reduce the gap between last_attempt and now_utc_micro_seconds from 7 days to 1 day
 	row := tx.QueryRowContext(ctx, `
 	SELECT object_key, want_b2, b2, want_wasabi, wasabi, want_scw, scw
 	FROM object_copies
@@ -43,7 +42,7 @@ func (repo *ObjectCopiesRepository) GetAndLockUnreplicatedObject(ctx context.Con
 		(
 			(wasabi IS NULL AND want_wasabi = true) OR
 			(scw IS NULL AND want_scw = true)
-		) AND last_attempt < (now_utc_micro_seconds() - (7* 24::BIGINT * 60 * 60 * 1000 * 1000))
+		) AND last_attempt < (now_utc_micro_seconds() - (24::BIGINT * 60 * 60 * 1000 * 1000))
 	)
 	LIMIT 1
 	FOR UPDATE SKIP LOCKED
@@ -112,6 +111,15 @@ func (repo *ObjectCopiesRepository) RegisterReplicationAttempt(tx *sql.Tx, ctx c
 	return stacktrace.Propagate(err, "")
 }
 
+func (repo *ObjectCopiesRepository) DelayNextAttemptByDays(ctx context.Context, objectKey string, days int) error {
+	_, err := repo.DB.ExecContext(ctx, `
+	UPDATE object_copies
+	SET last_attempt = last_attempt + ($2 * 24::BIGINT * 60 * 60 * 1000 * 1000)
+	WHERE object_key = $1
+	`, objectKey, days)
+	return stacktrace.Propagate(err, "")
+}
+
 // ResetNeedsB2Replication modifies the db to indicate that objectKey should be
 // re-replicated to Backblaze even if it has already been replicated there.
 func (repo *ObjectCopiesRepository) ResetNeedsB2Replication(objectKey string) error {