- Newest
- Most votes
- Most comments
While I cannot provide visibility into RDS backend processes or confirm specific service-side throttles during the "Creating" phase, I can address some aspects of your restore issue based on best practices for RDS snapshot restores.
Your observation about the snapshot containing MyISAM tables is significant. MyISAM tables can indeed cause restore failures or complications. When restoring from snapshots, temporary or MyISAM tables may not restore correctly because they aren't crash-safe and can become corrupted during the snapshot process. This is a known limitation that can prevent successful restoration.
To address slow restore performance in general, consider these approaches:
Optimize the source database before taking snapshots:
- Avoid long-running transactions before snapshot creation, as they generate more undo records that must be processed during restore
- Commit transactions in smaller batches to reduce undo log size
- Create snapshots during low workload periods when there's less database activity
Improve restore performance:
- Use a larger instance class with higher storage throughput and network bandwidth during the restore process
- Monitor the RollbackSegmentHistoryListLength metric in CloudWatch on your source instance and keep it low, as high values indicate numerous undo records that can slow restores
Address the MyISAM table issue: The most reliable solution would be to convert MyISAM tables to InnoDB before taking snapshots, as InnoDB is crash-safe and handles snapshots more reliably. If MyISAM tables are causing the restore to fail or hang indefinitely, this conversion may be necessary for successful restores.
If you continue experiencing issues, you may need to work with AWS Support to investigate the specific restore behavior, as they would have access to backend logs and processes that aren't visible through standard monitoring tools.
Sources
Resolve Amazon Aurora MySQL snapshot restore issues | AWS re:Post
What you’re observing is consistent with engine-internal recovery, not an AWS infrastructure issue. Key points to clarify: • During the “Creating” phase, RDS does not expose engine logs, process state, or SHOW PROCESSLIST, so there is no supported way to verify whether myisamchk or MyISAM index repair is running. • If the snapshot contains MyISAM tables, MariaDB may perform mandatory table/index validation or repair before marking the instance Available. This happens inside the managed service layer and is opaque to customers. • There are no documented service-side throttles specific to MyISAM recovery. However, MyISAM repair is single-threaded and I/O-latency sensitive, which explains: • Low CPU • Flat DiskQueueDepth • Minimal read throughput despite high gp3 limits • The fact that older snapshots restore quickly strongly indicates snapshot state at capture time (e.g., MyISAM tables mid-write) rather than instance class or storage configuration. Practical guidance: • AWS Support is the only channel that can confirm backend recovery activity. • For future reliability, avoid MyISAM on RDS where possible, or ensure snapshots are taken after a clean shutdown / low-write window. • InnoDB-only snapshots do not exhibit this behavior. Bottom line: This is expected behavior when restoring MyISAM-containing snapshots, and there is no customer-visible signal or override during the Creating phase.
answered 9 months ago
Hey, ran into something very similar before so hopefully this helps. What's likely happening is MariaDB is running an internal myisamchk repair before the instance can come online. MyISAM doesn't have crash-safe recovery like InnoDB, so if the snapshot was taken while the instance was active it has to rebuild indexes from scratch. This runs single-threaded so your extra provisioned IOPS basically does nothing here. EBS lazy loading makes it worse blocks are being pulled from S3 on demand during the repair, which is why you see that stuck disk queue, near-zero CPU, and barely any actual IOPS. Classic signature of this problem. You can't verify it directly during the Creating phase but an AWS support ticket will get you host-level log confirmation pretty quickly. For next time, run FLUSH TABLES WITH READ LOCK before taking the snapshot gives MyISAM a clean state and cuts restore time dramatically. Also worth converting tables to InnoDB where possible, restore behavior is just way more reliable. If the restore keeps failing and you suspect actual file-level corruption, try some third party tools like Stellar Repair for MySQL it works directly on the raw .MYI and .MYD files without needing a running instance, which is really useful when the restore just won't complete at all. Hope this helps!
answered 7 months ago
Relevant content
asked 4 years ago
asked 3 years ago
